Identify
Explore
Discover
Discuss
Summarise
Describe
Last, format your objectives into a numbered list. This is because when you write your thesis or dissertation, you will at times need to make reference to a specific research objective; structuring your research objectives in a numbered list will provide a clear way of doing this.
To bring all this together, let’s compare the first research objective in the previous example with the above guidance:
Research Objective:
1. Develop finite element models using explicit dynamics to mimic mallet blows during cup/shell insertion, initially using simplified experimentally validated foam models to represent the acetabulum.
Checking Against Recommended Approach:
Q: Is it specific? A: Yes, it is clear what the student intends to do (produce a finite element model), why they intend to do it (mimic cup/shell blows) and their parameters have been well-defined ( using simplified experimentally validated foam models to represent the acetabulum ).
Q: Is it measurable? A: Yes, it is clear that the research objective will be achieved once the finite element model is complete.
Q: Is it achievable? A: Yes, provided the student has access to a computer lab, modelling software and laboratory data.
Q: Is it relevant? A: Yes, mimicking impacts to a cup/shell is fundamental to the overall aim of understanding how they deform when impacted upon.
Q: Is it timebound? A: Yes, it is possible to create a limited-scope finite element model in a relatively short time, especially if you already have experience in modelling.
Q: Does it start with a verb? A: Yes, it starts with ‘develop’, which makes the intent of the objective immediately clear.
Q: Is it a numbered list? A: Yes, it is the first research objective in a list of eight.
1. making your research aim too broad.
Having a research aim too broad becomes very difficult to achieve. Normally, this occurs when a student develops their research aim before they have a good understanding of what they want to research. Remember that at the end of your project and during your viva defence , you will have to prove that you have achieved your research aims; if they are too broad, this will be an almost impossible task. In the early stages of your research project, your priority should be to narrow your study to a specific area. A good way to do this is to take the time to study existing literature, question their current approaches, findings and limitations, and consider whether there are any recurring gaps that could be investigated .
Note: Achieving a set of aims does not necessarily mean proving or disproving a theory or hypothesis, even if your research aim was to, but having done enough work to provide a useful and original insight into the principles that underlie your research aim.
Be realistic about what you can achieve in the time you have available. It is natural to want to set ambitious research objectives that require sophisticated data collection and analysis, but only completing this with six months before the end of your PhD registration period is not a worthwhile trade-off.
Each research objective should have its own purpose and distinct measurable outcome. To this effect, a common mistake is to form research objectives which have large amounts of overlap. This makes it difficult to determine when an objective is truly complete, and also presents challenges in estimating the duration of objectives when creating your project timeline. It also makes it difficult to structure your thesis into unique chapters, making it more challenging for you to write and for your audience to read.
Fortunately, this oversight can be easily avoided by using SMART objectives.
Hopefully, you now have a good idea of how to create an effective set of aims and objectives for your research project, whether it be a thesis, dissertation or research paper. While it may be tempting to dive directly into your research, spending time on getting your aims and objectives right will give your research clear direction. This won’t only reduce the likelihood of problems arising later down the line, but will also lead to a more thorough and coherent research project.
Finding a PhD has never been this easy – search for a PhD by keyword, location or academic area of interest.
Join thousands of students.
Join thousands of other students and stay up to date with the latest PhD programmes, funding opportunities and advice.
The research objective of a research proposal or scientific article defines the direction or content of a research investigation. Without the research objectives, the proposal or research paper is in disarray. It is like a fisherman riding on a boat without any purpose and with no destination in sight. Therefore, at the beginning of any research venture, the researcher must be clear about what he or she intends to do or achieve in conducting a study.
How do you define the objectives of a study? What are the uses of the research objective? How would a researcher write this essential part of the research? This article aims to provide answers to these questions.
Definition of a research objective.
“ What does the researcher want or hope to achieve at the end of the research project.”
The uses of the research objective are enumerated below:
The research design serves as the “blueprint” for the research investigation. The University of Southern California describes the different types of research design extensively. It details the data to be gathered, data collection procedure, data measurement, and statistical tests to use in the analysis.
The variables of the study include those factors that the researcher wants to evaluate in the study. These variables narrow down the research to several manageable components to see differences or correlations between them.
Specifying the data collection procedure ensures data accuracy and integrity . Thus, the probability of error is minimized. Generalizations or conclusions based on valid arguments founded on reliable data strengthens research findings on particular issues and problems.
In data mining activities where large data sets are involved, the research objective plays a crucial role. Without a clear objective to guide the machine learning process, the desired outcomes will not be met.
Before forming a research objective, you should read about all the developments in your area of research and find gaps in knowledge that need to be addressed. Readings will help you come up with suitable objectives for your research project.
The following examples of research objectives based on several published studies on various topics demonstrate how the research objectives are written:
Finally, writing the research objectives requires constant practice, experience, and knowledge about the topic investigated. Clearly written objectives save time, money, and effort.
I wrote a detailed, step-by-step guide on how to develop a conceptual framework with illustration in my post titled “ Conceptual Framework: A Step by Step Guide on How to Make One. “
Evans, K. L., Rodrigues, A. S., Chown, S. L., & Gaston, K. J. (2006). Protected areas and regional avian species richness in South Africa. Biology letters , 2 (2), 184-188.
Yeemin, T., Sutthacheep, M., & Pettongma, R. (2006). Coral reef restoration projects in Thailand. Ocean & Coastal Management , 49 (9-10), 562-575.
© 2020 March 23 P. A. Regoniel Updated 17 November 2020 | Updated 18 January 2024
A critique on the cooperative writing response groups, the yaya dub phenomenon: why videos go viral, lingoconomics – an emerging theory in language acquisition, about the author, patrick regoniel, simplyeducate.me privacy policy.
We use cookies to give you the best experience possible. By continuing we’ll assume you’re on board with our cookie policy
Step-by-step research objectives writing guide, step 1: provide the major background of your research, step 2: mention several objectives from the most to least important aspects, step 3: follow your resources and do not promise too much, step 4: keep your objectives and limitations mentioned, step 5: provide action verbs and tone, helpful tips for writing research objectives.
Research objective 1: The study aims to explore the origins and evolution of the youth movements in the Flemish provinces in Belgium, namely Chiro and KSA. This research evaluates the major differences during the post-WW2 period and the social factors that created differences between the movements.
Research objective 2: This paper implements surveys and personal interviews to determine first-hand feedback from the youth members and the team leaders.
Research objective 3: Aiming to compare and contrast, this study determines the positive outcomes of the unity project work between the branches of the youth movement in Belgium, aiming for statistical data to support it.
Receive paper in 3 Hours!
Number of Pages
550 words (double spaced)
Deadline: 10 days left
By clicking "Log In", you agree to our terms of service and privacy policy . We'll occasionally send you account related and promo emails.
Sign Up for your FREE account
Home » Research Paper – Structure, Examples and Writing Guide
Table of Contents
Definition:
Research Paper is a written document that presents the author’s original research, analysis, and interpretation of a specific topic or issue.
It is typically based on Empirical Evidence, and may involve qualitative or quantitative research methods, or a combination of both. The purpose of a research paper is to contribute new knowledge or insights to a particular field of study, and to demonstrate the author’s understanding of the existing literature and theories related to the topic.
The structure of a research paper typically follows a standard format, consisting of several sections that convey specific information about the research study. The following is a detailed explanation of the structure of a research paper:
The title page contains the title of the paper, the name(s) of the author(s), and the affiliation(s) of the author(s). It also includes the date of submission and possibly, the name of the journal or conference where the paper is to be published.
The abstract is a brief summary of the research paper, typically ranging from 100 to 250 words. It should include the research question, the methods used, the key findings, and the implications of the results. The abstract should be written in a concise and clear manner to allow readers to quickly grasp the essence of the research.
The introduction section of a research paper provides background information about the research problem, the research question, and the research objectives. It also outlines the significance of the research, the research gap that it aims to fill, and the approach taken to address the research question. Finally, the introduction section ends with a clear statement of the research hypothesis or research question.
The literature review section of a research paper provides an overview of the existing literature on the topic of study. It includes a critical analysis and synthesis of the literature, highlighting the key concepts, themes, and debates. The literature review should also demonstrate the research gap and how the current study seeks to address it.
The methods section of a research paper describes the research design, the sample selection, the data collection and analysis procedures, and the statistical methods used to analyze the data. This section should provide sufficient detail for other researchers to replicate the study.
The results section presents the findings of the research, using tables, graphs, and figures to illustrate the data. The findings should be presented in a clear and concise manner, with reference to the research question and hypothesis.
The discussion section of a research paper interprets the findings and discusses their implications for the research question, the literature review, and the field of study. It should also address the limitations of the study and suggest future research directions.
The conclusion section summarizes the main findings of the study, restates the research question and hypothesis, and provides a final reflection on the significance of the research.
The references section provides a list of all the sources cited in the paper, following a specific citation style such as APA, MLA or Chicago.
You can write Research Paper by the following guide:
Note : The below example research paper is for illustrative purposes only and is not an actual research paper. Actual research papers may have different structures, contents, and formats depending on the field of study, research question, data collection and analysis methods, and other factors. Students should always consult with their professors or supervisors for specific guidelines and expectations for their research papers.
Research Paper Example sample for Students:
Title: The Impact of Social Media on Mental Health among Young Adults
Abstract: This study aims to investigate the impact of social media use on the mental health of young adults. A literature review was conducted to examine the existing research on the topic. A survey was then administered to 200 university students to collect data on their social media use, mental health status, and perceived impact of social media on their mental health. The results showed that social media use is positively associated with depression, anxiety, and stress. The study also found that social comparison, cyberbullying, and FOMO (Fear of Missing Out) are significant predictors of mental health problems among young adults.
Introduction: Social media has become an integral part of modern life, particularly among young adults. While social media has many benefits, including increased communication and social connectivity, it has also been associated with negative outcomes, such as addiction, cyberbullying, and mental health problems. This study aims to investigate the impact of social media use on the mental health of young adults.
Literature Review: The literature review highlights the existing research on the impact of social media use on mental health. The review shows that social media use is associated with depression, anxiety, stress, and other mental health problems. The review also identifies the factors that contribute to the negative impact of social media, including social comparison, cyberbullying, and FOMO.
Methods : A survey was administered to 200 university students to collect data on their social media use, mental health status, and perceived impact of social media on their mental health. The survey included questions on social media use, mental health status (measured using the DASS-21), and perceived impact of social media on their mental health. Data were analyzed using descriptive statistics and regression analysis.
Results : The results showed that social media use is positively associated with depression, anxiety, and stress. The study also found that social comparison, cyberbullying, and FOMO are significant predictors of mental health problems among young adults.
Discussion : The study’s findings suggest that social media use has a negative impact on the mental health of young adults. The study highlights the need for interventions that address the factors contributing to the negative impact of social media, such as social comparison, cyberbullying, and FOMO.
Conclusion : In conclusion, social media use has a significant impact on the mental health of young adults. The study’s findings underscore the need for interventions that promote healthy social media use and address the negative outcomes associated with social media use. Future research can explore the effectiveness of interventions aimed at reducing the negative impact of social media on mental health. Additionally, longitudinal studies can investigate the long-term effects of social media use on mental health.
Limitations : The study has some limitations, including the use of self-report measures and a cross-sectional design. The use of self-report measures may result in biased responses, and a cross-sectional design limits the ability to establish causality.
Implications: The study’s findings have implications for mental health professionals, educators, and policymakers. Mental health professionals can use the findings to develop interventions that address the negative impact of social media use on mental health. Educators can incorporate social media literacy into their curriculum to promote healthy social media use among young adults. Policymakers can use the findings to develop policies that protect young adults from the negative outcomes associated with social media use.
References :
Appendix : The survey used in this study is provided below.
Social Media and Mental Health Survey
Thank you for your participation!
Research papers have several applications in various fields, including:
Research papers are typically written when a person has completed a research project or when they have conducted a study and have obtained data or findings that they want to share with the academic or professional community. Research papers are usually written in academic settings, such as universities, but they can also be written in professional settings, such as research organizations, government agencies, or private companies.
Here are some common situations where a person might need to write a research paper:
The purpose of a research paper is to present the results of a study or investigation in a clear, concise, and structured manner. Research papers are written to communicate new knowledge, ideas, or findings to a specific audience, such as researchers, scholars, practitioners, or policymakers. The primary purposes of a research paper are:
Research papers have several characteristics that distinguish them from other forms of academic or professional writing. Here are some common characteristics of research papers:
Research papers have many advantages, both for the individual researcher and for the broader academic and professional community. Here are some advantages of research papers:
Research papers also have some limitations that should be considered when interpreting their findings or implications. Here are some common limitations of research papers:
Researcher, Academic Writer, Web developer
Formulating research aim and objectives in an appropriate manner is one of the most important aspects of your thesis. This is because research aim and objectives determine the scope, depth and the overall direction of the research. Research question is the central question of the study that has to be answered on the basis of research findings.
Research aim emphasizes what needs to be achieved within the scope of the research, by the end of the research process. Achievement of research aim provides answer to the research question.
Research objectives divide research aim into several parts and address each part separately. Research aim specifies WHAT needs to be studied and research objectives comprise a number of steps that address HOW research aim will be achieved.
As a rule of dumb, there would be one research aim and several research objectives. Achievement of each research objective will lead to the achievement of the research aim.
Consider the following as an example:
Research title: Effects of organizational culture on business profitability: a case study of Virgin Atlantic
Research aim: To assess the effects of Virgin Atlantic organizational culture on business profitability
Following research objectives would facilitate the achievement of this aim:
Figure below illustrates additional examples in formulating research aims and objectives:
Formulation of research question, aim and objectives
Common mistakes in the formulation of research aim relate to the following:
1. Choosing the topic too broadly . This is the most common mistake. For example, a research title of “an analysis of leadership practices” can be classified as too broad because the title fails to answer the following questions:
a) Which aspects of leadership practices? Leadership has many aspects such as employee motivation, ethical behaviour, strategic planning, change management etc. An attempt to cover all of these aspects of organizational leadership within a single research will result in an unfocused and poor work.
b) An analysis of leadership practices in which country? Leadership practices tend to be different in various countries due to cross-cultural differences, legislations and a range of other region-specific factors. Therefore, a study of leadership practices needs to be country-specific.
c) Analysis of leadership practices in which company or industry? Similar to the point above, analysis of leadership practices needs to take into account industry-specific and/or company-specific differences, and there is no way to conduct a leadership research that relates to all industries and organizations in an equal manner.
Accordingly, as an example “a study into the impacts of ethical behaviour of a leader on the level of employee motivation in US healthcare sector” would be a more appropriate title than simply “An analysis of leadership practices”.
2. Setting an unrealistic aim . Formulation of a research aim that involves in-depth interviews with Apple strategic level management by an undergraduate level student can be specified as a bit over-ambitious. This is because securing an interview with Apple CEO Tim Cook or members of Apple Board of Directors might not be easy. This is an extreme example of course, but you got the idea. Instead, you may aim to interview the manager of your local Apple store and adopt a more feasible strategy to get your dissertation completed.
3. Choosing research methods incompatible with the timeframe available . Conducting interviews with 20 sample group members and collecting primary data through 2 focus groups when only three months left until submission of your dissertation can be very difficult, if not impossible. Accordingly, timeframe available need to be taken into account when formulating research aims and objectives and selecting research methods.
Moreover, research objectives need to be formulated according to SMART principle,
where the abbreviation stands for specific, measurable, achievable, realistic, and time-bound.
Study employee motivation of Coca-Cola | To study the impacts of management practices on the levels of employee motivation at Coca-Cola US by December 5, 2022
|
Analyze consumer behaviour in catering industry
| Analyzing changes in consumer behaviour in catering industry in the 21 century in the UK by March 1, 2022 |
Recommend Toyota Motor Corporation management on new market entry strategy
| Formulating recommendations to Toyota Motor Corporation management on the choice of appropriate strategy to enter Vietnam market by June 9, 2022
|
Analyze the impact of social media marketing on business
| Assessing impacts of integration of social media into marketing strategy on the level of brand awareness by March 30, 2022
|
Finding out about time management principles used by Accenture managers | Identifying main time-management strategies used by managers of Accenture France by December 1, 2022 |
Examples of SMART research objectives
At the conclusion part of your research project you will need to reflect on the level of achievement of research aims and objectives. In case your research aims and objectives are not fully achieved by the end of the study, you will need to discuss the reasons. These may include initial inappropriate formulation of research aims and objectives, effects of other variables that were not considered at the beginning of the research or changes in some circumstances during the research process.
John Dudovskiy
Mar 6, 2019
Have you checked out the rest of The PhD Knowledge Base ? It’s home to hundreds more free resources and guides, written especially for PhD students.
How long does it take the person reading your thesis to understand what you’re doing and how you’re doing it? If the answer is anything other than ’in the opening paragraphs of the thesis’ then keep reading.
If you tell them as early as possible what you’re doing and how you’re doing it – and do so in clear and simple terms – whatever you write after will make much more sense. If you leave them guessing for ten pages, everything they read in those ten pages has no coherence. You’ll know where it is all leading, but they won’t.
Unless you tell them.
If you tell the reader what you’re doing as early as possible in clear and simple terms, whatever you write after will make much more sense.
If you build a house without foundations, it’s pretty obvious what will happen. It’ll collapse. Your thesis is the same; fail to build the foundations and your thesis just won’t work .
Your aims and objectives are those foundations. That’s why we’ve put them right at the top of our PhD Writing Template (if you haven’t already downloaded it, join the thousands who have by clicking here ).
If you write your aims and objectives clearly then you’ll make your reader’s life easier.
A lot of students fail to clearly articulate their aims and objectives because they aren’t sure themselves what they actually are.
Picture this: if there’s one thing that every PhD student hates it’s being asked by a stranger what their research is on.
Use our free PhD structure template to quickly visualise every element of your thesis.
Sounds good, doesn’t it? Be able to call yourself Doctor sooner with our five-star rated How to Write A PhD email-course. Learn everything your supervisor should have taught you about planning and completing a PhD.
Now half price. Join hundreds of other students and become a better thesis writer, or your money back.
Your research aims are the answer to the question, ‘What are you doing?’
1. You need to clearly describe what your intentions are and what you hope to achieve. These are your aims.
2. Your aims may be to test theory in a new empirical setting, derive new theory entirely, construct a new data-set, replicate an existing study, question existing orthodoxy, and so on. Whatever they are, clearly articulate them and do so early. Definitely include them in your introduction and, if you’re smart, you’ll write them in your abstract .
3. Be very explicit . In the opening paragraphs, say, in simple terms, ‘ the aim of this thesis is to …’
4. Think of your aims then as a statement of intent. They are a promise to the reader that you are going to do something. You use the next two hundred pages or so to follow through on that promise. If you don’t make the promise, the reader won’t understand your follow-through. Simple as that.
Because they serve as the starting point of the study, there needs to be a flow from your aims through your objectives (more on this below) to your research questions and contribution and then into the study itself. If you have completed your research and found that you answered a different question (not that uncommon), make sure your original aims are still valid. If they aren’t, refine them.
If you struggle to explain in simple terms what your research is about and why it matters, you may need to refine your aims and objectives to make them more concise.
When writing up your aims, there are a number of things to bear in mind.
1. Avoid listing too many. Your PhD isn’t as long as you think it is and you won’t have time or room for more than around two or three.
2. When you write them up, be very specific. Don’t leave things so vague that the reader is left unsure or unclear on what you aim to achieve.
3. Make sure there is a logical flow between each of your aims. They should make sense together and should each be separate components which, when added together, are bigger than the sum of their parts.
Your aims answer the question, ‘What are you doing?’ The objectives are the answer to the question, ‘How are you doing it?’
Research objectives refer to the goals or steps that you will take to achieve your aims.
When you write them, make sure they are SMART.
You need to be as explicit as possible here. Leave the reader in no doubt about what you will do to achieve your aims. Step by step. Leave no ambiguity. At the same time, be careful not to repeat your methods chapter here. Just hint at your methods by presenting the headlines. You’ll have plenty of space in your methods discussion to flesh out the detail.
Elsewhere in the thesis you will necessarily have to talk in a complex language and juggle complex ideas. Here you don’t. You can write in clear, plain sentences.
The aims of a study describe what you hope to achieve. The objectives detail how you are going to achieve your aims.
Let’s use an example to illustrate.
Objectives:
If you’re still struggling, Professor Pat Thompson’s great blog has a guide that will help.
Leave the reader in no doubt about what you will do to achieve your aims. Step by step. Leave no ambiguity.
Of course your research is complex. That’s the name of the game. But the sign of someone being able to master complexity is their ability to summarise it . Sure, you’re not looking to capture all the richness and detail in a short summary of aims and objectives, but you are looking to tell the reader what you’re doing and how you’re doing it.
If you’re struggling to clearly articulate your aims and objectives, then try the following task. At the top of a Post-it note write the sentence: ‘In this research I will…’. Then keep trying until you can fit an answer onto one single Post-it note. The answer should answer two questions: what are are you doing and how are you doing it?
Remember – whenever you write, make it as clear as possible. Pay attention to the words ‘as possible’ there. That means you should write as clearly as you can given the fact that your subject and research is necessarily complex. Think of it the other way: it’s about not making things more complicated and unclear than they need to be.
In other words, make your reader’s job as easy as you can. They’ll thank you for it.
If you’re still having trouble, get in touch to arrange a one-on-one coaching session and we can work through your aims and objectives together.
32 comments.
The write up is quite inspiring.
My topic is setting up a healing gardens in hospitals Need a aim and objectives for a dissertation
Dis is really good and more understandable thanks
Crisp, concise, and easy to understnad. Thank you for posint this. I now know how to write up my report.
Great. Glad you found it useful.
Good piece of work! Very useful
Great. Glad you found it useful!
The write up makes sense
Great. Thanks!
I love this article. Amazing, outstanding and incredible facts.
Glad you found it useful!
Well written and easy to follow
Thank you for the comment, I’m really glad you found it valuable.
I’m currently developing a dissertation proposal for my PhD in organizational leadership. I need guidance in writing my proposal
Hey – have you checked out this guide? https://www.thephdproofreaders.com/writing/how-to-write-a-phd-proposal/
Indeed I’m impressed and gained a lot from this and I hope I can write an acceptable thesis with this your guide. Bello, H.K
Great. Thanks for the kind words. Good luck with the thesis.
Thumbs up! God job, well done. The information is quite concise and straight to the point.
Glad you thought so – good luck with the writing.
Dear Max, thank you so much for your work and efforts!
Your explanation about Aims and Objectives really helped me out. However, I got stuck with other parts of the Aims and Objectives Work Sheet: Scope, Main Argument, and Contribution.
Could you please explain these as well, preferably including some examples?
Thanks for your kind words. Your question is a big one! Without knowing lots about your topics/subject I’m not able to provide tailored advice, but broadly speaking your scope is the aims/objectives, your main argument is the thread running through the thesis (i.e. what your thesis is trying to argue) and the contribution (again, broadly speaking) is that gap you are filling.
I love your website and you’ve been so SO helpful..
DUMB QUESTION ALERT: Is there supposed to be a difference between aims and research question?
I mean, using your own example.. if the aim of my research is: “To understand the contribution that local governments make to national level energy policy” then wouldn’t the research question be: “How do local governments contribute to energy policy at national level”?
I am sorry if this comes out as completely obvious but I am at that stage of confusion where I am starting to question everything I know.
Sorry it’s taken me so long to reply! It’s not a dumb question at all. The aim of the study is what the study as a whole is seeking to achieve. So that might be the gap it is filling/the contribution it is making. The research questions are your means to achieving that aim. Your aim might be to fill a gap in knowledge, and you then may have a small number of questions that help you along that path. Does that make sense?
Thank you Max for this post! So helpful!
Thanks Anna!
Thanks so much this piece. I have written both bachelor’s and master’s thesis but haven’t read this made me feel like I didn’t know anything about research at all. I gained more insight into aims and objectives of academic researches.
Interesting explanation. Thank you.
I’m glad you found it useful.
Hi… I really like the way it is put “What are you going?” (Aims) and “How are you doing it?” (Objectives). Simple and straightforward. Thanks for making aims and objectives easy to understand.
Thank you for the write up it is insightful. if you are ask to discuss your doctoral aims. that means: what you are doing how you are doing it.
I was totally lost and still in the woods to the point of thinking I am dull, but looking at how you are coaching it tells me that i am just a student who needs to understand the lesson. I now believe that with your guidance i will pass my PhD. I am writing on an otherwise obvious subject, Value addition to raw materials, why Africa has failed to add value to raw materials? Difficult question as answers seem to abound, but that is where i differ and i seem to be against the general tide. However with your guidance I believe i will make it. Thanks.
Thanks for your lovely, kind words. So kind.
Your email address will not be published. Required fields are marked *
Most popular articles from the phd knowlege base.
An official website of the United States government
The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.
The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.
Preview improvements coming to the PMC website in October 2024. Learn More or Try it out now .
Patricia farrugia.
* Michael G. DeGroote School of Medicine, the
† Division of Orthopaedic Surgery and the
‡ Departments of Surgery and
§ Clinical Epidemiology and Biostatistics, McMaster University, Hamilton, Ont
There is an increasing familiarity with the principles of evidence-based medicine in the surgical community. As surgeons become more aware of the hierarchy of evidence, grades of recommendations and the principles of critical appraisal, they develop an increasing familiarity with research design. Surgeons and clinicians are looking more and more to the literature and clinical trials to guide their practice; as such, it is becoming a responsibility of the clinical research community to attempt to answer questions that are not only well thought out but also clinically relevant. The development of the research question, including a supportive hypothesis and objectives, is a necessary key step in producing clinically relevant results to be used in evidence-based practice. A well-defined and specific research question is more likely to help guide us in making decisions about study design and population and subsequently what data will be collected and analyzed. 1
In this article, we discuss important considerations in the development of a research question and hypothesis and in defining objectives for research. By the end of this article, the reader will be able to appreciate the significance of constructing a good research question and developing hypotheses and research objectives for the successful design of a research study. The following article is divided into 3 sections: research question, research hypothesis and research objectives.
Interest in a particular topic usually begins the research process, but it is the familiarity with the subject that helps define an appropriate research question for a study. 1 Questions then arise out of a perceived knowledge deficit within a subject area or field of study. 2 Indeed, Haynes suggests that it is important to know “where the boundary between current knowledge and ignorance lies.” 1 The challenge in developing an appropriate research question is in determining which clinical uncertainties could or should be studied and also rationalizing the need for their investigation.
Increasing one’s knowledge about the subject of interest can be accomplished in many ways. Appropriate methods include systematically searching the literature, in-depth interviews and focus groups with patients (and proxies) and interviews with experts in the field. In addition, awareness of current trends and technological advances can assist with the development of research questions. 2 It is imperative to understand what has been studied about a topic to date in order to further the knowledge that has been previously gathered on a topic. Indeed, some granting institutions (e.g., Canadian Institute for Health Research) encourage applicants to conduct a systematic review of the available evidence if a recent review does not already exist and preferably a pilot or feasibility study before applying for a grant for a full trial.
In-depth knowledge about a subject may generate a number of questions. It then becomes necessary to ask whether these questions can be answered through one study or if more than one study needed. 1 Additional research questions can be developed, but several basic principles should be taken into consideration. 1 All questions, primary and secondary, should be developed at the beginning and planning stages of a study. Any additional questions should never compromise the primary question because it is the primary research question that forms the basis of the hypothesis and study objectives. It must be kept in mind that within the scope of one study, the presence of a number of research questions will affect and potentially increase the complexity of both the study design and subsequent statistical analyses, not to mention the actual feasibility of answering every question. 1 A sensible strategy is to establish a single primary research question around which to focus the study plan. 3 In a study, the primary research question should be clearly stated at the end of the introduction of the grant proposal, and it usually specifies the population to be studied, the intervention to be implemented and other circumstantial factors. 4
Hulley and colleagues 2 have suggested the use of the FINER criteria in the development of a good research question ( Box 1 ). The FINER criteria highlight useful points that may increase the chances of developing a successful research project. A good research question should specify the population of interest, be of interest to the scientific community and potentially to the public, have clinical relevance and further current knowledge in the field (and of course be compliant with the standards of ethical boards and national research standards).
Feasible | ||
Interesting | ||
Novel | ||
Ethical | ||
Relevant |
Adapted with permission from Wolters Kluwer Health. 2
Whereas the FINER criteria outline the important aspects of the question in general, a useful format to use in the development of a specific research question is the PICO format — consider the population (P) of interest, the intervention (I) being studied, the comparison (C) group (or to what is the intervention being compared) and the outcome of interest (O). 3 , 5 , 6 Often timing (T) is added to PICO ( Box 2 ) — that is, “Over what time frame will the study take place?” 1 The PICOT approach helps generate a question that aids in constructing the framework of the study and subsequently in protocol development by alluding to the inclusion and exclusion criteria and identifying the groups of patients to be included. Knowing the specific population of interest, intervention (and comparator) and outcome of interest may also help the researcher identify an appropriate outcome measurement tool. 7 The more defined the population of interest, and thus the more stringent the inclusion and exclusion criteria, the greater the effect on the interpretation and subsequent applicability and generalizability of the research findings. 1 , 2 A restricted study population (and exclusion criteria) may limit bias and increase the internal validity of the study; however, this approach will limit external validity of the study and, thus, the generalizability of the findings to the practical clinical setting. Conversely, a broadly defined study population and inclusion criteria may be representative of practical clinical practice but may increase bias and reduce the internal validity of the study.
Population (patients) | ||
Intervention (for intervention studies only) | ||
Comparison group | ||
Outcome of interest | ||
Time |
A poorly devised research question may affect the choice of study design, potentially lead to futile situations and, thus, hamper the chance of determining anything of clinical significance, which will then affect the potential for publication. Without devoting appropriate resources to developing the research question, the quality of the study and subsequent results may be compromised. During the initial stages of any research study, it is therefore imperative to formulate a research question that is both clinically relevant and answerable.
The primary research question should be driven by the hypothesis rather than the data. 1 , 2 That is, the research question and hypothesis should be developed before the start of the study. This sounds intuitive; however, if we take, for example, a database of information, it is potentially possible to perform multiple statistical comparisons of groups within the database to find a statistically significant association. This could then lead one to work backward from the data and develop the “question.” This is counterintuitive to the process because the question is asked specifically to then find the answer, thus collecting data along the way (i.e., in a prospective manner). Multiple statistical testing of associations from data previously collected could potentially lead to spuriously positive findings of association through chance alone. 2 Therefore, a good hypothesis must be based on a good research question at the start of a trial and, indeed, drive data collection for the study.
The research or clinical hypothesis is developed from the research question and then the main elements of the study — sampling strategy, intervention (if applicable), comparison and outcome variables — are summarized in a form that establishes the basis for testing, statistical and ultimately clinical significance. 3 For example, in a research study comparing computer-assisted acetabular component insertion versus freehand acetabular component placement in patients in need of total hip arthroplasty, the experimental group would be computer-assisted insertion and the control/conventional group would be free-hand placement. The investigative team would first state a research hypothesis. This could be expressed as a single outcome (e.g., computer-assisted acetabular component placement leads to improved functional outcome) or potentially as a complex/composite outcome; that is, more than one outcome (e.g., computer-assisted acetabular component placement leads to both improved radiographic cup placement and improved functional outcome).
However, when formally testing statistical significance, the hypothesis should be stated as a “null” hypothesis. 2 The purpose of hypothesis testing is to make an inference about the population of interest on the basis of a random sample taken from that population. The null hypothesis for the preceding research hypothesis then would be that there is no difference in mean functional outcome between the computer-assisted insertion and free-hand placement techniques. After forming the null hypothesis, the researchers would form an alternate hypothesis stating the nature of the difference, if it should appear. The alternate hypothesis would be that there is a difference in mean functional outcome between these techniques. At the end of the study, the null hypothesis is then tested statistically. If the findings of the study are not statistically significant (i.e., there is no difference in functional outcome between the groups in a statistical sense), we cannot reject the null hypothesis, whereas if the findings were significant, we can reject the null hypothesis and accept the alternate hypothesis (i.e., there is a difference in mean functional outcome between the study groups), errors in testing notwithstanding. In other words, hypothesis testing confirms or refutes the statement that the observed findings did not occur by chance alone but rather occurred because there was a true difference in outcomes between these surgical procedures. The concept of statistical hypothesis testing is complex, and the details are beyond the scope of this article.
Another important concept inherent in hypothesis testing is whether the hypotheses will be 1-sided or 2-sided. A 2-sided hypothesis states that there is a difference between the experimental group and the control group, but it does not specify in advance the expected direction of the difference. For example, we asked whether there is there an improvement in outcomes with computer-assisted surgery or whether the outcomes worse with computer-assisted surgery. We presented a 2-sided test in the above example because we did not specify the direction of the difference. A 1-sided hypothesis states a specific direction (e.g., there is an improvement in outcomes with computer-assisted surgery). A 2-sided hypothesis should be used unless there is a good justification for using a 1-sided hypothesis. As Bland and Atlman 8 stated, “One-sided hypothesis testing should never be used as a device to make a conventionally nonsignificant difference significant.”
The research hypothesis should be stated at the beginning of the study to guide the objectives for research. Whereas the investigators may state the hypothesis as being 1-sided (there is an improvement with treatment), the study and investigators must adhere to the concept of clinical equipoise. According to this principle, a clinical (or surgical) trial is ethical only if the expert community is uncertain about the relative therapeutic merits of the experimental and control groups being evaluated. 9 It means there must exist an honest and professional disagreement among expert clinicians about the preferred treatment. 9
Designing a research hypothesis is supported by a good research question and will influence the type of research design for the study. Acting on the principles of appropriate hypothesis development, the study can then confidently proceed to the development of the research objective.
The primary objective should be coupled with the hypothesis of the study. Study objectives define the specific aims of the study and should be clearly stated in the introduction of the research protocol. 7 From our previous example and using the investigative hypothesis that there is a difference in functional outcomes between computer-assisted acetabular component placement and free-hand placement, the primary objective can be stated as follows: this study will compare the functional outcomes of computer-assisted acetabular component insertion versus free-hand placement in patients undergoing total hip arthroplasty. Note that the study objective is an active statement about how the study is going to answer the specific research question. Objectives can (and often do) state exactly which outcome measures are going to be used within their statements. They are important because they not only help guide the development of the protocol and design of study but also play a role in sample size calculations and determining the power of the study. 7 These concepts will be discussed in other articles in this series.
From the surgeon’s point of view, it is important for the study objectives to be focused on outcomes that are important to patients and clinically relevant. For example, the most methodologically sound randomized controlled trial comparing 2 techniques of distal radial fixation would have little or no clinical impact if the primary objective was to determine the effect of treatment A as compared to treatment B on intraoperative fluoroscopy time. However, if the objective was to determine the effect of treatment A as compared to treatment B on patient functional outcome at 1 year, this would have a much more significant impact on clinical decision-making. Second, more meaningful surgeon–patient discussions could ensue, incorporating patient values and preferences with the results from this study. 6 , 7 It is the precise objective and what the investigator is trying to measure that is of clinical relevance in the practical setting.
The following is an example from the literature about the relation between the research question, hypothesis and study objectives:
Study: Warden SJ, Metcalf BR, Kiss ZS, et al. Low-intensity pulsed ultrasound for chronic patellar tendinopathy: a randomized, double-blind, placebo-controlled trial. Rheumatology 2008;47:467–71.
Research question: How does low-intensity pulsed ultrasound (LIPUS) compare with a placebo device in managing the symptoms of skeletally mature patients with patellar tendinopathy?
Research hypothesis: Pain levels are reduced in patients who receive daily active-LIPUS (treatment) for 12 weeks compared with individuals who receive inactive-LIPUS (placebo).
Objective: To investigate the clinical efficacy of LIPUS in the management of patellar tendinopathy symptoms.
The development of the research question is the most important aspect of a research project. A research project can fail if the objectives and hypothesis are poorly focused and underdeveloped. Useful tips for surgical researchers are provided in Box 3 . Designing and developing an appropriate and relevant research question, hypothesis and objectives can be a difficult task. The critical appraisal of the research question used in a study is vital to the application of the findings to clinical practice. Focusing resources, time and dedication to these 3 very important tasks will help to guide a successful research project, influence interpretation of the results and affect future publication efforts.
FINER = feasible, interesting, novel, ethical, relevant; PICOT = population (patients), intervention (for intervention studies only), comparison group, outcome of interest, time.
Competing interests: No funding was received in preparation of this paper. Dr. Bhandari was funded, in part, by a Canada Research Chair, McMaster University.
Imagine you’re a student planning a vacation in a foreign country. You’re on a tight budget and need to draw…
Imagine you’re a student planning a vacation in a foreign country. You’re on a tight budget and need to draw up a pocket-friendly plan. Where do you begin? The first step is to do your research.
Before that, you make a mental list of your objectives—finding reasonably-priced hotels, traveling safely and finding ways of communicating with someone back home. These objectives help you focus sharply during your research and be aware of the finer details of your trip.
More often than not, research is a part of our daily lives. Whether it’s to pick a restaurant for your next birthday dinner or to prepare a presentation at work, good research is the foundation of effective learning. Read on to understand the meaning, importance and examples of research objectives.
What are the objectives of research, what goes into a research plan.
Research is a careful and detailed study of a particular problem or concern, using scientific methods. An in-depth analysis of information creates space for generating new questions, concepts and understandings. The main objective of research is to explore the unknown and unlock new possibilities. It’s an essential component of success.
Over the years, businesses have started emphasizing the need for research. You’ve probably noticed organizations hiring research managers and analysts. The primary purpose of business research is to determine the goals and opportunities of an organization. It’s critical in making business decisions and appropriately allocating available resources.
Here are a few benefits of research that’ll explain why it is a vital aspect of our professional lives:
One of the greatest benefits of research is to learn and gain a deeper understanding. The deeper you dig into a topic, the more well-versed you are. Furthermore, research has the power to help you build on any personal experience you have on the subject.
Research encourages you to discover the most recent information available. Updated information prevents you from falling behind and helps you present accurate information. You’re better equipped to develop ideas or talk about a topic when you’re armed with the latest inputs.
Research provides you with a good foundation upon which you can develop your thoughts and ideas. People take you more seriously when your suggestions are backed by research. You can speak with greater confidence because you know that the information is accurate.
Take any leading nonprofit organization, you’ll see how they have a strong research arm supported by real-life stories. Research also becomes the base upon which real-life connections and impact can be made. It even helps you communicate better with others and conveys why you’re pursuing something.
As we’ve already established, research is mostly about using existing information to create new ideas and opinions. In the process, it sparks curiosity as you’re encouraged to explore and gain deeper insights into a subject. Curiosity leads to higher levels of positivity and lower levels of anxiety.
Well-defined objectives of research are an essential component of successful research engagement. If you want to drive all aspects of your research methodology such as data collection, design, analysis and recommendation, you need to lay down the objectives of research methodology. In other words, the objectives of research should address the underlying purpose of investigation and analysis. It should outline the steps you’d take to achieve desirable outcomes. Research objectives help you stay focused and adjust your expectations as you progress.
The objectives of research should be closely related to the problem statement, giving way to specific and achievable goals. Here are the four types of research objectives for you to explore:
Also known as secondary objectives, general objectives provide a detailed view of the aim of a study. In other words, you get a general overview of what you want to achieve by the end of your study. For example, if you want to study an organization’s contribution to environmental sustainability, your general objective could be: a study of sustainable practices and the use of renewable energy by the organization.
Specific objectives define the primary aim of the study. Typically, general objectives provide the foundation for identifying specific objectives. In other words, when general objectives are broken down into smaller and logically connected objectives, they’re known as specific objectives. They help define the who, what, why, when and how aspects of your project. Once you identify the main objective of research, it’s easier to develop and pursue a plan of action.
Let’s take the example of ‘a study of an organization’s contribution to environmental sustainability’ again. The specific objectives will look like this:
To determine through history how the organization has changed its practices and adopted new solutions
To assess how the new practices, technology and strategies will contribute to the overall effectiveness
Once you’ve identified the objectives of research, it’s time to organize your thoughts and streamline your research goals. Here are a few effective tips to develop a powerful research plan and improve your business performance.
Your research objectives should be SMART—Specific, Measurable, Achievable, Realistic and Time-constrained. When you focus on utilizing available resources and setting realistic timeframes and milestones, it’s easier to prioritize objectives. Continuously track your progress and check whether you need to revise your expectations or targets. This way, you’re in greater control over the process.
Create a plan that’ll help you select appropriate methods to collect accurate information. A well-structured plan allows you to use logical and creative approaches towards problem-solving. The complexity of information and your skills are bound to influence your plan, which is why you need to make room for flexibility. The availability of resources will also play a big role in influencing your decisions.
After you’ve created a plan for the research process, make a list of the data you’re going to collect and the methods you’ll use. Not only will it help make sense of your insights but also keep track of your approach. The information you collect should be:
Logical, rigorous and objective
Can be reproduced by other people working on the same subject
Free of errors and highlighting necessary details
Current and updated
Includes everything required to support your argument/suggestions
Data analysis is the most crucial part of the process and there are many ways in which the information can be utilized. Four types of data analysis are often seen in a professional environment. While they may be divided into separate categories, they’re linked to each other.
The most commonly used data analysis, descriptive analysis simply summarizes past data. For example, Key Performance Indicators (KPIs) use descriptive analysis. It establishes certain benchmarks after studying how someone has been performing in the past.
The next step is to identify why something happened. Diagnostic analysis uses the information gathered through descriptive analysis and helps find the underlying causes of an outcome. For example, if a marketing initiative was successful, you deep-dive into the strategies that worked.
It attempts to answer ‘what’s likely to happen’. Predictive analysis makes use of past data to predict future outcomes. However, the accuracy of predictions depends on the quality of the data provided. Risk assessment is an ideal example of using predictive analysis.
The most sought-after type of data analysis, prescriptive analysis combines the insights of all of the previous analyses. It’s a huge organizational commitment as it requires plenty of effort and resources. A great example of prescriptive analysis is Artificial Intelligence (AI), which consumes large amounts of data. You need to be prepared to commit to this type of analysis.
Once you’ve collected and collated your data, it’s time to review it and draw accurate conclusions. Here are a few ways to improve the review process:
Identify the fundamental issues, opportunities and problems and make note of recurring trends if any
Make a list of your insights and check which is the most or the least common. In short, keep track of the frequency of each insight
Conduct a SWOT analysis and identify the strengths, weaknesses, opportunities and threats
Write down your conclusions and recommendations of the research
When we think about research, we often associate it with academicians and students. but the truth is research is for everybody who is willing to learn and enhance their knowledge. If you want to master the art of strategically upgrading your knowledge, Harappa Education’s Learning Expertly course has all the answers. Not only will it help you look at things from a fresh perspective but also show you how to acquire new information with greater efficiency. The Growth Mindset framework will teach you how to believe in your abilities to grow and improve. The Learning Transfer framework will help you apply your learnings from one context to another. Begin the journey of tactful learning and self-improvement today!
Explore Harappa Diaries to learn more about topics related to the THINK Habit such as Learning From Experience , Critical Thinking & What is Brainstorming to think clearly and rationally.
Barry Mauer and John Venecek
This portion of the course covers key library resources: literature databases, academic journals, scholarly monographs, and primary source collections. We also discuss key library services for undergraduates as well as connecting with librarians who specialize in English studies, and search tips that will help make your research more efficient. We also cover an often-overlooked skill: citation management, which enables you to compile, organize, and manage your resources efficiently. Managing citations as you go will reduce the stress of the research process
Understanding how to efficiently locate relevant literature will free up time for your reading and writing. You will learn about library services to help you with your search. A key resource is your subject librarian, who is always available to help. In this chapter, you will learn about:
Chapter 6 Objectives Copyright © 2021 by Barry Mauer and John Venecek is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License , except where otherwise noted.
A, Participant flow for the discovery study. B, Participant flow for the replication study. In both studies, during a single visit at the testing site, enrolled participants received expert clinical diagnosis using standardized assessments (reference standard diagnosis) as well as eye-tracking–based measurement of social visual engagement (index test). Index test quality control indicator (QCIs) failures occurred when participants’ data failed to meet automated preset data QCIs (additional details are available in eTables 2 and 3 in Supplement 1 ).
Performance among 711 children in the discovery study and 361 children in the replication study. A, The diamond represents the optimal test positivity threshold for the discovery study (Youden index). B, The test positivity threshold determined in the discovery study was fixed and applied independently in the replication study. The diamond represents the achieved sensitivity and specificity in the replication study using the test positivity threshold from the discovery study. The solid blue circle represents the post hoc theoretical optimal threshold. C, Tabulation corresponds to the diamond in panel A. D, Tabulation corresponds to the diamond in panel B. AUC indicates area under the curve; ROC, receiver operating characteristic. Negative predictive value (NPV) and positive predictive value (PPV) estimates reported here are calculated based on study sample prevalence.
A, Discovery study correlation between eye-tracking–based indices of social disability versus children’s total scores on the Autism Diagnostic Observation Schedule, second edition (ADOS-2). B, Discovery study correlation between eye-tracking–based indices of verbal ability versus children’s verbal age equivalent scores as measured by the Mullen Scales of Early Learning (Mullen). C, Discovery study correlation between eye-tracking–based indices of nonverbal cognitive ability versus children’s nonverbal age equivalent scores as measured by the Mullen. D, E, F, Replication study correlations between eye-tracking–based indices and reference standard assessments. In all scatterplots, circles represent individual data and diamonds represent regression outliers (bivariate outliers identified using Cook distance and difference-in-fits regression diagnostic assessment). The adjusted R 2 values were adjusted for measurement error variance of the reference standard (yielding percentage of reference standard nonerror variance explained by the index test). Additional information is provided in the Secondary End Point Analyses subsection of the eMethods in Supplement 1 .
Measurement of social visual engagement quantifies how a child engages with social and nonsocial cues occurring continuously within naturalistic environmental contexts (left column, shown as still frames from testing videos). In relation to those contexts, normative reference measures provide objective quantification of nonautism age-expected visual engagement (middle columns, shown as density distributions in both pseudocolor format and as color to grayscale fades overlaid on corresponding still frames). The age-expected reference measures can be used to measure and visualize patient comparisons, revealing individual strengths, vulnerabilities, and opportunities for skill building (right columns, sample patient data shown as overlaid circular apertures that encompass the portion of video foveated by each patient [each aperture spans the central 5.2 degrees of a patient’s visual field]). Children with autism present as engaging with toys of interest (1, 3, 5, and 7), color and contrast cues (2, 6, 8, and 9), and objects and background elements not directly relevant to social context (4 and 10-14). Elapsed times at the bottom right of still frames highlight the rapidly changing nature of social interaction in which many hundreds of verbal and nonverbal communicative cues are presented, each eliciting age-expected patterns of engagement and offering corresponding opportunities for objective quantitative comparisons of patient behavior.
eFigure 1. Example Video Stimuli and Coded Regions of Interest
eFigure 2. Eye-Tracking Data Collection Device
eFigure 3. Analysis of Dynamic Visual Scanning and Derivation of Attentional Funnels
eFigure 4. Using Kernel Density Estimation to Derive Attentional Funnels and Quantify Dynamic Visual Scanning
eFigure 5. Eye-Tracking Calibration Accuracy
eFigure 6. Probability Density Functions of Individual Score Numeric Values Used to Make Index Test Categorical Determinations of Autism vs Nonautism, Plotted According to Reference Standard Expert Clinical Diagnosis in Discovery and Replication Studies
eTable 1. Correlation of Eye-Tracking–Based Indices With Reference Standard Assessments
eTable 2. Discovery Study Participants With Missing Eye-Tracking Data (Failed Quality Controls)
eTable 3. Replication Study Participants With Missing Eye-Tracking Data (Failed Quality Controls)
eMethods. Participants, Experimental Design, Experimental Procedures and Data Collection, Data Processing, and Data Analysis and Statistics
eResults. Clinical Outcomes of False Positives and Negatives and Missing Data
eReferences
Data Sharing Statement
Sign up for emails based on your interests, select your interests.
Customize your JAMA Network experience by selecting one or more topics from the list below.
Others also liked.
Jones W , Klaiman C , Richardson S, et al. Development and Replication of Objective Measurements of Social Visual Engagement to Aid in Early Diagnosis and Assessment of Autism. JAMA Netw Open. 2023;6(9):e2330145. doi:10.1001/jamanetworkopen.2023.30145
© 2024
Question Can objective measurements of social visual engagement be developed and replicated to aid in early diagnosis and assessment of autism before age 3 years?
Findings In 2 prospective double-blind studies of diagnostic performance in 1089 children aged 16 to 30 months, 719 in discovery and 370 in replication, eye-tracking–based measurements of social visual engagement relative to expert clinical diagnosis had area under the receiver operating characteristic curve of 0.90, sensitivity of 81.9%, and specificity of 89.9% in discovery; and area under the curve of 0.89, sensitivity of 80.6%, and specificity of 82.3% in replication.
Meaning These results offer the prospect of an objective biomarker to aid in autism diagnosis and assessment.
Importance Autism spectrum disorder is a common and early-emerging neurodevelopmental condition. While 80% of parents report having had concerns for their child’s development before age 2 years, many children are not diagnosed until ages 4 to 5 years or later.
Objective To develop an objective performance-based tool to aid in early diagnosis and assessment of autism in children younger than 3 years.
Design, Setting, and Participants In 2 prospective, consecutively enrolled, broad-spectrum, double-blind studies, we developed an objective eye-tracking–based index test for children aged 16 to 30 months, compared its performance with best-practice reference standard diagnosis of autism (discovery study), and then replicated findings in an independent sample (replication study). Discovery and replication studies were conducted in specialty centers for autism diagnosis and treatment. Reference standard diagnoses were made using best-practice standardized protocols by specialists blind to eye-tracking results. Eye-tracking tests were administered by staff blind to clinical results. Children were enrolled from April 27, 2013, until September 26, 2017. Data were analyzed from March 28, 2018, to January 3, 2019.
Main Outcomes and Measures Prespecified primary end points were the sensitivity and specificity of the eye-tracking–based index test compared with the reference standard. Prespecified secondary end points measured convergent validity between eye-tracking–based indices and reference standard assessments of social disability, verbal ability, and nonverbal ability.
Results Data were collected from 1089 children: 719 children (mean [SD] age, 22.4 [3.6] months) in the discovery study, and 370 children (mean [SD] age, 25.4 [6.0] months) in the replication study. In discovery, 224 (31.2%) were female and 495 (68.8%) male; in replication, 120 (32.4%) were female and 250 (67.6%) male. Based on reference standard expert clinical diagnosis, there were 386 participants (53.7%) with nonautism diagnoses and 333 (46.3%) with autism diagnoses in discovery, and 184 participants (49.7%) with nonautism diagnoses and 186 (50.3%) with autism diagnoses in replication. In the discovery study, the area under the receiver operating characteristic curve was 0.90 (95% CI, 0.88-0.92), sensitivity was 81.9% (95% CI, 77.3%-85.7%), and specificity was 89.9% (95% CI, 86.4%-92.5%). In the replication study, the area under the receiver operating characteristic curve was 0.89 (95% CI, 0.86-0.93), sensitivity was 80.6% (95% CI, 74.1%-85.7%), and specificity was 82.3% (95% CI, 76.1%-87.2%). Eye-tracking test results correlated with expert clinical assessments of children’s individual levels of ability, explaining 68.6% (95% CI, 58.3%-78.6%), 63.4% (95% CI, 47.9%-79.2%), and 49.0% (95% CI, 33.8%-65.4%) of variance in reference standard assessments of social disability, verbal ability, and nonverbal cognitive ability, respectively.
Conclusions and Relevance In two diagnostic studies of children younger than 3 years, objective eye-tracking–based measurements of social visual engagement quantified diagnostic status as well as individual levels of social disability, verbal ability, and nonverbal ability in autism. These findings suggest that objective measurements of social visual engagement can be used to aid in autism diagnosis and assessment.
Approximately 1 in 36 US children is affected by autism. 1 Thirty percent of parents of children with autism had concerns for their child’s development before age 12 months, 50% of parents had concerns by age 18 months, and 80% had concerns by age 2 years. 2 - 4 Despite these early concerns and the manifest behaviors that elicited these concerns, the median age of US diagnosis remains delayed until the age of 4 to 5 years. 5 , 6 The age of diagnosis is even later among those who lack resources or lack access to expert clinicians: diagnoses for US racial minority families, families with low income, and families residing in rural areas lag further. 1 , 6 - 9
The goal of diagnosis in autism is to facilitate timely and targeted support to help a child and family as needed. To that end, there may be an important role for new tools and objective biomarkers that can accurately and efficiently aid in diagnosing children as well as aid in quantifying individual strengths and vulnerabilities. 10 Such tools could enhance health care system capacity and help facilitate timely access to individually appropriate services. 10 , 11
In current best practice, autism is diagnosed behaviorally by symptomatic deficits in social interaction and communication and by the presence of restricted and repetitive behaviors. 12 Current gold (reference) standard 13 diagnostic instruments are standardized validated assessments that measure the presence of autistic social disability through both behavioral observation and parent interview. 14 , 15 Best-practice guidelines also call for standardized assessments of a child’s cognitive and language skills. 16
Unfortunately, there are often long wait lists to access expert clinicians using gold standard instruments (a situation now described as a crisis) 17 and community use of gold standard instruments is limited. 18 , 19 Consequently, many children experience delayed diagnosis, and most receive diagnostic labels without receiving comprehensive evaluations and standardized assessments. 18 , 19
In the present studies, we tested the performance of eye-tracking–based measurements of social visual engagement to accurately predict autism diagnoses and to objectively quantify individual levels of social disability, verbal ability, and nonverbal cognitive ability. In primary analyses, we measured the sensitivity and specificity of eye-tracking assays in comparison with clinician best-estimate diagnosis by expert clinicians. In secondary analyses, we quantified convergent validity between eye-tracking–based indices of social disability, verbal ability, and nonverbal cognitive ability in comparison with standardized assessments thereof as administered by expert clinicians. 20 , 21
The present studies build on prior research using eye tracking to quantify social visual engagement, defined as how children look at and learn from their surrounding social environment. The prior research found that social visual engagement is strongly influenced by individual genetic variation (with monozygotic twin-twin concordance of approximately 0.9), 22 is highly biologically conserved, 23 and is atypical in autism 24 , 25 from very early ages in development (ie, 2-6 months). 26 Here we test the hypothesis that measurements of social visual engagement collected via eye tracking can serve as a robust biomarker to enable early diagnosis and assessment of autism.
The goal of the current studies was to evaluate performance of eye-tracking–based assays to accurately assess categorical presence of autism and to measure dimensional levels of ability or disability. In design, terminology, and reporting, this research followed the Standards for Reporting of Diagnostic Accuracy ( STARD ) guidelines, 27 - 29 with eye-tracking assays referred to as the “index test,” and expert clinical diagnosis using standardized assessments referred to as the “reference standard.” Two observational studies were conducted: a discovery study that was used to develop the data collection tool and algorithms, and a replication study that was used to test performance in an independent sample. Children were consecutively enrolled from April 27, 2013, until September 26, 2017. Data were analyzed from March 28, 2018, to January 3, 2019. The research protocol was approved by the institutional review boards of Emory University and Washington University in St Louis. Written informed consent was obtained from all parents or legal guardians.
To eliminate or minimize design-related bias (as highlighted by Lijmer et al 30 ), data were collected prospectively; participants were enrolled consecutively; enrolled participants had a broad spectrum of case presentation (spanning the full spectrum of symptom severity and absence of symptoms); clinical assessments were blind to eye tracking, eye tracking was blind to clinical assessments; and the index test and reference standard diagnosis were performed with all participants. Only the results of best-practice standardized assessments and reference standard clinician best-estimate diagnosis 31 were used clinically or communicated to parents. In this way, best-practice standard of care was maintained for all participants, blind to eye-tracking results; neither a child’s parents nor expert clinical staff were informed of a child’s eye-tracking results.
A total of 1089 children participated: 719 participated in the discovery study, and 370 participated in the replication study ( Figure 1 ). Eligible participants were identified on the basis of chronological age and were recruited through placement of advertising materials in local media, specialty clinics, and pediatric practices. The studies were designed to develop and test a tool to aid in the diagnosis and assessment of autism, not to test the tool’s utility as a screening instrument. To that end, children for whom there were concerns about autism were recruited (ie, children typically evaluated in specialized clinics for the diagnosis of autism). To be eligible for participation, children could not have clinically meaningful hearing or visual impairments (eg, congenital deafness, blindness, or nystagmus); could not have previously diagnosed genetic conditions associated with autism-related symptoms (eg, not known to have fragile X or Rett syndromes); had to be generally healthy at the time of testing, with no acute illness; had to be either born at or after 37 weeks’ gestational age (discovery study) or born at or after 32 weeks’ gestational age (replication study); and had to be either between the ages of 16 and 30 months (discovery study) or between the ages of 16 and 45 months (replication study). In both studies, the age of participants was guided by future intended use (ie, to align in time with ages that would ideally enable diagnosed children to be referred to early intervention before age 36 months). In the replication study, to test performance among a broader range of children, increased prematurity at birth and older age at enrollment was allowed. For the purpose of sample characterization, patient demographic data (including race, ethnicity, and maternal educational level) were collected by parents’ selection of fixed categories. Race and ethnicity data were collected to enable evaluation of whether test performance varied based on these characteristics. Further details are available in the eMethods in Supplement 1 .
Reference standard diagnosis consisted of clinician best-estimate diagnosis 32 - 35 by experienced licensed clinicians using standardized diagnostic protocols and developmental assessments. 20 Reference standard diagnosis was assigned based on all available clinical information, including developmental assessments as well as medical and developmental history gathered in clinical interviews. At young ages, clinician best-estimate diagnosis (ie, experienced clinicians’ judgments using the totality of information available) is a more stable predictor of later diagnosis than strict reliance on cutoff scores. 20 , 36 For example, while scores on the Autism Diagnostic Observation Schedule, second edition (ADOS-2), may vary during the first 2 to 3 years of life, clinician best estimate is more stable. 32 , 36 The standardized diagnostic protocol was sequential so that a child’s developmental history and scores on screening tests dictated subsequent assessments. A complete description of this protocol is available in the Reference Standard Diagnostic Assessment Procedures subsection of the eMethods in Supplement 1 .
For index test measurements of social visual engagement, eye-tracking data were collected while participants watched video scenes of social interaction (examples are provided in eFigure 1 in Supplement 1 ). Fourteen video scenes were presented, each with a mean (SD) duration of approximately 54.0 (21.5) seconds (range, 21.7 seconds to 1 minute, 29.7 seconds; sum, 12 minutes, 35.5 seconds). Experimental procedures, data collection, and data processing were performed as described in prior studies 22 and in the Experimental Procedures and Data Collection subsection of the eMethods in Supplement 1 . Data collection for the discovery study was performed in an academic medical center laboratory setting. Data collection for the replication study was performed in both an academic medical center laboratory setting and a community clinic using a standalone investigational eye-tracking device (eFigure 2 in Supplement 1 ). Eye-tracking data were collected using near-infrared video-based measurements of eye movements using specialized cameras and hardware (additional details are provided in the Experimental Procedures and Data Collection subsection of the eMethods in Supplement 1 ).
All collected data underwent automated quality control analyses measuring calibration accuracy, integrity of eye movements, duration of data collected, and time spent fixating on video scenes. Data that met or exceeded predefined automated static quality control thresholds proceeded to analysis (Quality Control Indicators subsection of the eMethods in Supplement 1 ). All steps in data processing and analysis were automated, with no manual human review or analysis required.
Time-varying kernel density estimation was used to quantify social visual engagement 37 (eFigure 3 in Supplement 1 ). Probability density functions of visual fixation and scanning were calculated during each moment of collected eye-tracking data (eFigure 4 in Supplement 1 ). Moments in time when the majority of participants with nonautism diagnoses in the discovery study fixated on approximately the same location(s) at the same moments at levels greater than expected by chance were identified by permutation testing. 38 Discovery study data were then mined to identify time points when the majority of participants with autism fixated on alternate locations (defining a classification index). Data were also mined to identify time points when autism and nonautism discovery study data were correlated with measurements of (1) social disability (correlated with ADOS-2 total scores), (2) verbal ability (correlated with verbal age-equivalent scores from the Mullen Scales of Early Learning, hereinafter, Mullen), or (3) nonverbal cognitive ability (correlated with visual reception age-equivalent scores from the Mullen). Data mining for these associations thereby defined 3 indices of individual variability in levels of disability and ability. Further details are provided in the Data Processing subsection of the eMethods in Supplement 1 . Discovery study results were tested by leave-one-out cross validation, 39 with each participant tested as an independent comparison relative to the rest of the sample. All parameters were fixed and then tested again in the independent replication study.
Primary effectiveness analyses were planned as a comparison between the eye-tracking index test results and the reference standard diagnosis results (either autism or nonautism). Sensitivity and specificity were calculated according to standard practice: sensitivity was calculated as the proportion of participants with reference standard autism diagnoses who had eye-tracking results that also indicated autism; specificity was calculated as the proportion of participants with reference standard nonautism diagnoses who had eye-tracking results that also indicated nonautism. The test positivity threshold was derived in the discovery study using the Youden index 40 ; the threshold was then fixed for testing in the independent replication study. Receiver operating characteristic curves, area under the curve, positive predictive value (PPV), negative predictive value (NPV), and accuracy of the eye-tracking index test were also calculated, all with 95% CIs. 41 Primary end point analyses were tested at a 1-sided significance level of α = .025.
Secondary effectiveness analyses were planned as measurements of correlation between eye-tracking–based severity indices and their respective expert clinician–administered reference standard assessments, including the ADOS-2 total score for social disability, the mean of Mullen receptive and expressive language age-equivalent scores for verbal ability, and the Mullen visual reception age-equivalent score for nonverbal ability. For social disability, the correlation was expected to be negative because higher scores on the ADOS-2 denote greater social disability, whereas for the eye-tracking test, lower scores denote greater social disability. Deming regression 42 , 43 was used to quantify the relationships between eye-tracking–based indices and their respective reference standards. Standard regression diagnostics (including Cook distance and difference-in-fits), 44 , 45 Pearson correlation coefficients, and adjusted R 2 coefficients 46 - 48 together with 95% CIs were calculated. Secondary outcome analyses were tested at a 1-sided significance level of α = .025. Data analyses were performed in Matlab R2016a (Mathworks, Inc).
A total of 719 children (mean [SD] age, 22.4 [3.6] months; 224 [31.2%] female and 495 [68.8%] male) were enrolled in the discovery study. A total of 370 children (mean [SD] age, 25.4 [6.0] months; 120 [32.4%] female and 250 [67.6%] male) were enrolled in the replication study. Based on reference standard diagnosis, the discovery study comprised 386 participants (53.7%) with nonautism diagnoses and 333 (46.3%) with autism diagnoses, while the replication study comprised 184 participants (49.7%) with nonautism diagnoses and 186 (50.3%) with autism diagnoses.
Participant characteristics and demographic data are shown in the Table . In both studies, participants with autism had higher ADOS-2 domain and total scores (all t > 24.0, all P < .001). Participants with autism also had lower Mullen verbal age-equivalent scores (both t > 7.4, P < .001) and lower Mullen nonverbal age-equivalent scores (both t > 5.1, P < .001). The ADOS-2 scores in both studies indicated that participants with autism represented the full spectrum of autism symptom severity. Likewise, the Mullen scores in both studies indicated that participants with nonautism and autism diagnoses represented a broad range of verbal and nonverbal abilities, extending from substantially delayed to age-appropriate to advanced abilities. In each study, mean (SD) age of the sample with autism diagnoses was significantly older than the sample with nonautism diagnoses (discovery: 23.1 [3.7] months vs 21.7 [3.4] months; t = 5.2, P <.001; replication: 28.1 [5.8] months vs 22.7 [4.9] months; t = 9.7, P < .001). Sex differences were as expected, 49 with a higher number of boys diagnosed with autism in both studies (both χ 2 > 16.6; P < .001).
Average calibration accuracy was within 1 degree of visual angle and did not differ significantly between diagnostic groups or study samples (eFigure 5 in Supplement 1 ). There were no significant between-group differences in duration of data collected (discovery: t = 1.48; P = .14; replication : t = 0.81 , P = .42). Children with nonautism diagnoses did fixate (discovery: t = 4.97, P < .001; replication: t = 8.51, P < .001) and saccade (discovery: t = 6.75, P < .001; replication: t = 8.10, P < .001) significantly more, and blink less (discovery: t = 4.61, P < .001; replication: t = 4.08, P < .001), than children with autism, which was consistent with expected diagnostic differences in attention to and engagement with social cues in the environment 50 that have been commonly noted in autism. 36
Prespecified primary end point analyses measured the diagnostic accuracy of eye-tracking–based index test results in comparison with reference standard diagnosis. Results are shown in Figure 2 as receiver operating characteristic curves (panels A and B) and diagnostic cross-tabulations with performance measure estimates (panels C and D); underlying score distributions are plotted in eFigure 6 in Supplement 1 . Index test performance had area under the curve statistics equal to 0.90 (95% CI, 0.88-0.92) in the discovery study and 0.89 (95% CI, 0.86-0.93) in the replication study. The test positivity threshold in the discovery study was selected to match the Youden index (represented by the diamond in Figure 2 A). After discovery study determination, the test positivity threshold was fixed and applied in the replication study. Achieved sensitivity and specificity in the replication study are shown in Figure 2 B (represented by the diamond, which corresponds to the cross-tabulation results and performance measure estimates in Figure 2 D). Eye-tracking–based index test results predicted expert clinician reference standard diagnosis with sensitivity equal to 81.9% (95% CI, 77.3%-85.7%) and specificity equal to 89.9% (95% CI, 86.4%-92.5%) in the discovery study, and sensitivity equal to 80.6% (95% CI, 74.1%-85.7%) and specificity equal to 82.3% (95% CI, 76.1%-87.2%) in the replication study. Sensitivity, specificity, PPV, NPV, and accuracy did not differ significantly by sex (all with overlapping 95% CIs of performance estimates). Additional information regarding clinical outcomes is provided in the Clinical Outcomes of False Positives and Negatives subsection in eResults in Supplement 1 .
Prespecified secondary end point analyses measured the strength of association between eye-tracking–based indices and reference standard behavioral assessments of social disability, verbal ability, and nonverbal ability. Results are shown in Figure 3 as scatter plots with Deming regression fitted functions, Pearson R values, and adjusted R 2 coefficients of determination (also summarized in eTable 1 in Supplement 1 ).
In the discovery study ( Figure 3 A-C), the correlation between the index test social disability index and the ADOS-2 total score was −0.74 (95% CI, −0.78 to −0.70), the correlation between the index test verbal ability index and the Mullen verbal age-equivalent score was 0.71 (95% CI, 0.67-0.75), and the correlation between the nonverbal ability index and the Mullen nonverbal age-equivalent score was 0.66 (95% CI, 0.61-0.70). In the replication study ( Figure 3 D-F), the correlation between the index test social disability index and the ADOS-2 total score was −0.72 (95% CI, −0.78 to −0.65), the correlation between the index test verbal ability index and the Mullen verbal age-equivalent score was 0.59 (95% CI, 0.50-0.77), and the correlation between the index test nonverbal ability index and the Mullen nonverbal age-equivalent score was 0.53 (95% CI, 0.43-0.62).
From the replication study, adjusted for reference standard measurement error, the eye-tracking–based social disability index accounted for 68.6% (adjusted R 2 = 0.69; 95% CI, 0.58-0.79) of variance in ADOS-2 total scores. The verbal ability index accounted for 63.4% (adjusted R 2 = 0.63; 95% CI, 0.48-0.79) of variance in Mullen verbal age-equivalent scores. The nonverbal ability index accounted for 49.0% (adjusted R 2 = 0.49; 95% CI, 0.34-0.65) of variance in Mullen nonverbal age-equivalent scores.
In all comparisons, the strength of association between the eye-tracking–based indices and their respective expert clinician–administered assessments was high ( R > 0.5), suggesting strong convergent validity between index and reference standard measures for social disability, verbal ability, and nonverbal ability. There were no significant differences in strength of association by sex. Participants with eye-tracking quality control indicator failures (8 of 719 in the discovery study and 9 of 370 in the replication study), with no results returned, are described further in eTables 2 and 3 in Supplement 1 . Additional details and comparisons are available in the eResults in Supplement 1 .
In 2 prospective double-blind diagnostic studies, the first a discovery study and the second a replication study, 1089 children were tested to measure the diagnostic performance of index test measurements of social visual engagement relative to reference standard expert clinical diagnosis of autism. Children aged 16 to 30 months (discovery study) and 16 to 45 months (replication study) were assessed by expert clinicians to test whether measurements of social visual engagement could accurately predict categorical diagnosis as well as dimensional levels of social disability, verbal ability, and nonverbal ability.
The results, reported in accordance with the STARD initiative, 27 - 29 found that measurements of social visual engagement had 81.9% sensitivity and 89.9% specificity relative to expert clinical diagnosis of autism in the discovery study and 80.6% sensitivity and 82.3% specificity in the replication study. Sensitivity, specificity, PPV, NPV, and accuracy did not differ significantly between the discovery and replication studies, suggesting robust and replicable performance. In addition, measurements of social visual engagement were also predictive of children’s individual scores on gold standard behavioral assessments: measurements of social visual engagement effectively explained 68.6% of variance in individual levels of social disability (ADOS-2 total scores), 63.4% of variance in verbal ability (Mullen verbal age-equivalent scores), and 49.0% of variance in nonverbal cognitive ability (Mullen nonverbal age-equivalent scores).
These results suggest high convergent validity with reference standard assessments that otherwise require highly trained experts to spend multiple hours of assessment time per child. In contrast, for measurements of social visual engagement, biomarker data collection consisted of children watching videos (eFigure 1 in Supplement 1 ), with data collected on a standalone mobile eye-tracking device that was deployed in a clinic and operated by technicians with no required clinical or technical expertise (eFigure 2 in Supplement 1 ).
Once social visual engagement data are collected, although data processing and analysis are computationally intensive, they are also automated, deployed on cloud-based servers, and capable of returning a results report in less than 30 minutes. The index test is objective and quantitative and directly measures thousands of instances of children’s behavior for comparison with age-expected norms (examples are shown in eFigure 4 in Supplement 1 ). Data processing and analysis to derive diagnostic classification and indices of symptom severity are entirely automated, requiring no special expertise or eye-tracking knowledge on the part of clinicians.
It is important to note that the test results derived from measurements of social visual engagement are not intended to replace clinicians with expertise in developmental disabilities; to the contrary, a tool like this could be used by expert clinicians to aid in accurately and efficiently diagnosing autism as well as quantifying children’s strengths and vulnerabilities. Therefore, these results offer important opportunities to enhance health care system capacity and facilitate more rapid progress from the time of first concern to the start of individually appropriate services. 10 While empirically supported services have their own access challenges, those challenges are not a reason to delay diagnosis or to delay initiation of supports for children and families. 51
Finally, the meaning of these measurements resides in what they quantify: repeated divergence from shared social experience with rapid accrual of atypical experience ( Figure 4 ). Shared experience is the foundation for communication and social development. By quantifying the number, extent, and timing of divergence from shared experience, measurements of social visual engagement provide a transactional biomarker: direct objective measurements of a child’s unique biology transacting with specific environmental contexts. Those transactions are the building blocks of learning and brain development. 52
Recognizing the transactional nature of this developmental process is a reminder that the emergence of disability is itself transactional, driven by genetic liabilities but also by atypical learning experiences that are correlated with those liabilities. 53 Recognizing this provides 2 notable reasons for optimism. First, it reminds us that disability is a cocreation, a consequence of individual vulnerabilities transacting with particular environmental contexts, severely disabling in some contexts but less so or not at all in others. 54 Fostering early intervention approaches and contexts that embrace difference and diversity while also augmenting individual adaptive skills is important to reducing disability and optimizing outcomes for all. Second, knowing that the unfolding of disability is transactional means that it can be measured as such to (1) identify children in need of support; (2) monitor specific behaviors and contexts that may exacerbate or ameliorate disability over time; and (3), ideally, intervene more successfully and treat specific individual manifestations and vulnerabilities for disability.
This study has several limitations. Clinical procedures were performed by a relatively small group of expert clinicians, and eye-tracking procedures were implemented under well-controlled laboratory conditions in the discovery study or with a single prototype standalone eye-tracking device in the replication study. The efforts in this study should be complemented by studies collecting reference standard and index test data at multiple different sites with multiple different clinical teams and eye-tracking devices. 55 The results of this study should also be complemented by data quantifying repeatability and reproducibility variance in eye-tracking–based measurements. 56 Previous studies 57 , 58 have also noted expert clinical uncertainty in the reference standard diagnosis of autism in some children. Uncertainty in the reference standard sets an upper limit on the performance measures of any comparison test (eFigure 3 in Jones et al 55 ). In the current study, we did not prospectively track expert clinician certainty of diagnosis in all children. Consequently, we were unable to analyze the effects of clinician certainty in the discovery or replication studies. This limitation was improved in a subsequent multisite study. 55
In 2 diagnostic studies of children aged 16 to 30 months with and without autism, objective measurements of social visual engagement were able to quantify diagnostic status and assess individual levels of social disability, verbal ability, and nonverbal ability. These findings suggest that objective measurements of social visual engagement can be used to aid in autism diagnosis and assessment.
Accepted for Publication: July 11, 2023.
Published: September 5, 2023. doi:10.1001/jamanetworkopen.2023.30145
Open Access: This is an open access article distributed under the terms of the CC-BY License . © 2023 Jones W et al. JAMA Network Open .
Corresponding Author: Warren Jones, PhD, Marcus Autism Center, Children’s Healthcare of Atlanta, 1920 Briarcliff Rd NE, Atlanta, GA 30329 ( [email protected] ).
Author Contributions: Dr Jones had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Concept and design: Jones, Klaiman, Lewis, Shultz, Klin.
Acquisition, analysis, or interpretation of data: All authors.
Drafting of the manuscript: Jones, Klaiman, Edwards, Klin.
Critical review of the manuscript for important intellectual content: Hamner, Paredes.
Statistical analysis: Jones, Edwards.
Obtained funding: Jones, Lewis, Edwards, Klin.
Administrative, technical, or material support: Jones, Klaiman, Richardson, Lambha, Hamner, Beacham, Lewis, Paredes, Edwards, Marrus, Constantino, Shultz, Klin.
Supervision: Jones, Klaiman, Lewis, Edwards, Marrus, Shultz, Klin.
Conflict of Interest Disclosures: Dr Jones reported receiving grants from the Georgia Research Alliance, the Joseph B. Whitehead Foundation, the Marcus Foundation, and the National Institute of Mental Health (NIMH) during the conduct of the study (all via his institution); being a scientific cofounder of and owning equity in EarliTec Diagnostics; holding patents licensed to EarliTec Diagnostics; and receiving personal fees for consulting from EarliTec Diagnostics and a lecture honorarium from Washington University in St Louis School of Medicine outside the submitted work. Dr Klaiman reported receiving grants from the Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD), the Georgia Research Alliance, the Joseph B. Whitehead Foundation, the Marcus Foundation, and the NIMH (all via her institution) during the conduct of the study and personal fees from ABA Centers of America, Dekalb County School District, Cherokee County School District, Fulton County School District, Beaming Health, and EarliTec Diagnostics outside the submitted work. Dr Richardson reported receiving funding and/or equipment from the Georgia Research Alliance, the John B. Whitehead Foundation, and the Marcus Foundation (all via her institution) during the conduct of the study. Dr Lambha reported receiving funding and/or equipment from the Georgia Research Alliance, the Joseph B. Whitehead Foundation, the Marcus Foundation, the NICHD, and the NIMH (all via her institution) during the conduct of the study. Dr Beacham reported receiving funding and/or equipment from the John B. Whitehead Foundation and the Marcus Foundation during the conduct of the study. Mr Lewis reported receiving grants from the Georgia Research Alliance, the Joseph B. Whitehead Foundation, and the Marcus Foundation (all via his institution) during the conduct of the study; being a scientific cofounder of and owning equity in EarliTec Diagnostics; and holding patents licensed to EarliTec Diagnostics outside the submitted work. Mr Paredes reported receiving funding and/or equipment from the Joseph B. Whitehead Foundation and the Marcus Foundation (both via his institution) during the conduct of the study. Dr Marrus reported receiving grants from the NIMH and from Washington University in St Louis School of Medicine during the conduct of the study. Dr Constantino reported receiving grants from the NICHD during the conduct of the study and royalties from Western Psychological Services outside the submitted work. Dr Klin reported receiving grants from the Georgia Research Alliance, the Joseph B. Whitehead Foundation, the Marcus Foundation, the NICHD, and the NIMH (all via his institution) during the conduct of the study; being a scientific cofounder of and owning equity in EarliTec Diagnostics; holding patents licensed to EarliTec Diagnostics; and receiving personal fees from the Alliance for Early Success, the Brazilian Society of Speech Therapy, EarliTec Diagnostics, the McKnight Endowment Fund for Neuroscience, the National Autism Conference, and Washington University in St Louis School of Medicine outside the submitted work. No other disclosures were reported.
Funding/Support: This study was supported by funding from the Marcus Foundation (Drs Jones and Klin), the Joseph B. Whitehead Foundation (Drs Jones and Klin), and the Georgia Research Alliance (Drs Jones and Klin and Mr Lewis); grants HD068479 and U54 HD087011 from the NICHD (Dr Constantino); and grants MH100029 (Drs Jones and Klin) and MH100019 (Dr Marrus) from the NIMH.
Role of the Funder/Sponsor: The funding organizations had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Data Sharing Statement: See Supplement 2 .
Additional Contributions: We thank the families and children for their participation. We also thank the fellows of the Donald J. Cohen Fellowship in Developmental Social Neuroscience, the Simons Fellowship in Computational Neuroscience, and the Simons Fellowship in Design Engineering for providing help with data collection and processing at Marcus Autism Center, Children’s Healthcare of Atlanta, Emory University. We thank Erika Mortenson, BA, Sayli Sant, MD, Teddi Gray, BA, Yi Zhang, MS, Laura Campbell, BA, and Leena Malik, BA, of Washington University in St Louis, for assisting with data collection and processing. We thank Steve Kovar, BFA, of Marcus Autism Center, Children’s Healthcare of Atlanta, for providing help with building the standalone eye-tracking device used in the replication study and Robin Sifre, PhD, of EarliTec Diagnostics, for providing comments on the manuscript. The named individuals were not compensated outside of their normal salary.
Published on 25.6.2024 in Vol 26 (2024)
Authors of this article:
1 Department of Molecular Pharmacology and Experimental Therapeutics, Mayo Clinic, Rochester, MN, United States
2 Mayo Clinic Libraries, Mayo Clinic, Rochester, MN, United States
3 Department of Family Medicine, Mayo Clinic, Phoenix, AZ, United States
4 Department of Psychiatry and Psychology, Mayo Clinic, Rochester, MN, United States
5 Department of Psychiatry and Psychology, Mayo Clinic, Jacksonville, FL, United States
6 Department of Nursing, Mayo Clinic, Rochester, MN, United States
7 Department of Medicine, University of Colorado School of Medicine, Aurora, CO, United States
*these authors contributed equally
Liselotte N Dyrbye, MD, MPHE
Department of Medicine
University of Colorado School of Medicine
Mail Stop C290, Fitzsimons Bldg
13001 E 17th Pl. Rm #E1347
Aurora, CO, 80045
United States
Phone: 1 303 724 4982
Email: [email protected]
Background: The occupational burnout epidemic is a growing issue, and in the United States, up to 60% of medical students, residents, physicians, and registered nurses experience symptoms. Wearable technologies may provide an opportunity to predict the onset of burnout and other forms of distress using physiological markers.
Objective: This study aims to identify physiological biomarkers of burnout, and establish what gaps are currently present in the use of wearable technologies for burnout prediction among health care professionals (HCPs).
Methods: A comprehensive search of several databases was performed on June 7, 2022. No date limits were set for the search. The databases were Ovid: MEDLINE(R), Embase, Healthstar, APA PsycInfo, Cochrane Central Register of Controlled Trials, Cochrane Database of Systematic Reviews, Web of Science Core Collection via Clarivate Analytics, Scopus via Elsevier, EBSCOhost: Academic Search Premier, CINAHL with Full Text, and Business Source Premier. Studies observing anxiety, burnout, stress, and depression using a wearable device worn by an HCP were included, with HCP defined as medical students, residents, physicians, and nurses. Bias was assessed using the Newcastle Ottawa Quality Assessment Form for Cohort Studies.
Results: The initial search yielded 505 papers, from which 10 (1.95%) studies were included in this review. The majority (n=9) used wrist-worn biosensors and described observational cohort studies (n=8), with a low risk of bias. While no physiological measures were reliably associated with burnout or anxiety, step count and time in bed were associated with depressive symptoms, and heart rate and heart rate variability were associated with acute stress. Studies were limited with long-term observations (eg, ≥12 months) and large sample sizes, with limited integration of wearable data with system-level information (eg, acuity) to predict burnout. Reporting standards were also insufficient, particularly in device adherence and sampling frequency used for physiological measurements.
Conclusions: With wearables offering promise for digital health assessments of human functioning, it is possible to see wearables as a frontier for predicting burnout. Future digital health studies exploring the utility of wearable technologies for burnout prediction should address the limitations of data standardization and strategies to improve adherence and inclusivity in study participation.
Burnout is an occupational syndrome characterized by emotional exhaustion, depersonalization, and feelings of reduced personal accomplishment caused by chronic, unmitigated high levels of job-related stress [ 1 ]. Burnout is common among health care professionals (HCPs, also referred to as health care workers), impacting an estimated 35% to 54% of nurses and physicians, and between 45% and 60% of medical students and resident physicians in the United States [ 2 ]. Several studies also reveal a high prevalence of depression and anxiety in HCPs that preceded the coronavirus pandemic [ 3 - 9 ]. Data further suggests that burnout and other forms of distress have increased among HCPs as a result of the COVID-19 pandemic [ 10 - 12 ].
This is concerning because the well-being of HCPs impacts the quality of patient care and patients’ access to care. Several meta-analyses and systematic reviews have reported associations between burnout and negative impacts on the quality of care provided to patients, including increasing the risk of medical errors [ 13 ], malpractice claims [ 14 ], nosocomial infections [ 15 ], and mortality [ 16 ]. Additionally, other studies have found that HCPs who report experiencing burnout are more likely to reduce their time taking care of patients and quit, all of which negatively impact patient’s access to care and add a burden to the global health care system [ 2 ]. The impacts of burnout go beyond the workplace, as HCPs with reported burnout are at increased risk of cardiovascular diseases [ 17 , 18 ], suicidal ideation [ 13 , 19 ], substance use disorders [ 20 ], uncontrolled stress [ 21 ], car accidents [ 22 ], and quality of life [ 23 ].
Contributors of burnout in HCPs are multifactorial and complex. While most factors contributing to burnout originate from system-level factors within the work environment, some risk factors originate from the personal domain or challenges in the personal-professional interface, such as work-home conflict ( Figure 1 ). Due to the complexity of the factors involved, no model exists for predicting when an individual HCP or group of HCPs are at risk for developing burnout or other forms of distress. In response to the negative outcomes of burnout for HCPs and patients, the National Academies of Science, Engineering, and Medicine recommends health care organizations monitor (through frequent surveys) and respond to burnout. This approach is retrospective, as the time required for health care organizations to administer surveys, HCPs to complete them, and the additional time needed to analyze and interpret results all delay any response to burnout. A better approach would be a proactive one, where organizations or individual HCPs could predict and respond to high levels of job stress before the manifestation of burnout and associated personal and professional consequences result.
Previous studies and reviews suggest heart rate (HR) [ 24 ], heart rate variability (HRV) [ 24 ], sleep [ 25 ], and skin temperature [ 26 ] vary in response to stress. Additionally, sleep or fatigue also relates to the risk of burnout [ 27 ], depression [ 28 ], and other related conditions [ 29 ]. These types of data can be collected passively from wearable devices. Over the past 5 years, the adoption of wearable devices worldwide has more than doubled [ 30 ]. Therefore, data collected passively from wearable devices could potentially provide an avenue for detecting individuals at risk for high job stress, burnout, depression, and other related conditions. If predictive, such real-time information obtained passively from wearable devices could dramatically shift the current reactive paradigm to a proactive one, potentially leading to meaningful intervention before patients and HCPs experience adverse health consequences of burnout.
Previous systematic reviews suggest wearable devices may have some utility in predicting depression severity and stress levels [ 31 ]. To our knowledge, there is no review that investigates this relationship among HCPs or explores the ability of wearable devices to detect burnout risk. Hence, a scoping review was conducted to identify and summarize studies exploring associations between burnout, anxiety, depression, and stress, with data obtained from wearable devices in cohorts of HCPs.
A comprehensive search of several databases was performed on June 7, 2022. No date limits were set for the search. The databases (and their coverage periods) were Ovid: MEDLINE (1946 to Present and Epub Ahead of Print, In-Process and Other Non-Indexed Citations and Daily), Embase (1974+), Healthstar (1966+), APA PsycInfo (1987+), Cochrane Central Register of Controlled Trials (1991+), Cochrane Database of Systematic Reviews (2005+), Web of Science Core Collection via Clarivate Analytics (1975+), Scopus via Elsevier (1788+), EBSCOhost: Academic Search Premier, CINAHL with Full Text (1981+), and Business Source Premier.
The search strategy was designed and conducted by a medical librarian (LCH) with input from the study’s investigators (APA and LND). Controlled vocabulary supplemented with keywords was used. The actual strategies listing all search terms used and how they are combined are available in the Multimedia Appendix 1 .
The initial search yielded 505 papers. Two reviewers (MB and SS) independently identified and screened the titles and abstracts of potentially eligible papers. The inclusion criteria of the initial round of screening were as follows: the study must include a validated measure of burnout, stress, anxiety, or depression and the study must include only data from a wearable device worn by an HCP. For this work, we defined HCP as being a medical student, resident, practicing physician, or registered nurse in a hospital or outpatient clinical setting. The full-text reviews of the papers that resulted from the initial screening, data extraction, and quality assessment were also performed independently and in pairs by 2 reviewers (MB and SS). Papers were not excluded due to their calculated quality score. During this process, 475 papers were omitted because they did not satisfy the inclusion criteria (n=472) or were duplicates (n=3). After the initial screening, the full text of 30 papers was assessed for eligibility. Any disagreement was resolved by consensus with other senior reviewers (APA and LND) and the final source list was created, with senior reviewers blinded to reviews of each other and primary reviewers (MB and SS). The study selection process is illustrated in Figure 2 . Tables 1 and 2 provide descriptions of the final 10 papers published from April 2017 to December 2021 included in this review.
Author | Sample characteristics | Wearable-derived measurements | Validated anxiety, burnout, stress, or depression measures | Other measure included |
Feng et al [ ] | 113 Nurses | HR , Sleep, and STC | STAI | Positive and Negative Affect Schedule, Satisfaction with Life Scale, Pittsburgh Sleep Quality Index, Affect EMA , Big Five Inventory-2, and Anxiety and Stress EMA |
Adler et al [ ] | 775 Residents | HR, Sleep, and STC | PHQ-9 | Mood EMA |
Jevsevar et al [ ] | 21 Resident and Physicians | HRV , RHR , RR , and Sleep | MBI-Abbreviated | — |
Silva et al [ ] | 83 Medical students (19 had complete data) | HR and HRV | PSS-4 | — |
Mendelsohn et al [ ] | 59 Residents | Sleep and STC | MBI-HSS | Short-Form Health Survey, Epworth Sleepiness Scale, Satisfaction with Medicine Scale, and International Physical Activity Questionnaire |
Marek et al [ ] | 28 Residents | RHR, Sleep, and STC | Single-item burnout measure | — |
Sochacki et al [ ] | 21 Physicians | Sleep | MBI-HSS, PROMIS-29 (Depression and Anxiety) | — |
Chaukos et al [ ] | 75 Residents (26 had complete data) | Activity level and Sleep | MBI–HSS, PSS-10, and PHQ-9 | Functional Assessment of Chronic Illness Therapy-Fatigue, Penn State Worry Questionnaire, Revised Life Orientation Test, Interpersonal Reactivity Index Perspective-Taking subscale, Measure of Current Status-Part A, and Cognitive Affective Mindfulness Scale |
de Looff et al [ ] | 114 Nurses | SC | MBI–HSS (modified Dutch version) | — |
Weenk et al [ ] | 20 Residents and Physicians | HR and HRV | STAI-short version | — |
a HR: heart rate.
b STC: step count.
c STAI: State-Trait Anxiety Inventory.
d EDA: electrodermal activity.
e EMA: ecological momentary assessment.
f PHQ-9: Patient Health Questionnaire.
g HRV: heart rate variability.
h RHR: resting heart rate.
i RR: respiratory rate.
j Not available.
k PSS: Perceived Stress Scale.
l MBI-HSS: Maslach Burnout Inventory–Human Services Survey.
m PROMIS : Performance of the Patient-Reported Outcomes.
n SC: skin conductance.
Author | Device | Length of data collection | Primary findings | Newcastle Ottawa Scale Score |
Feng et al [ ] | Fitbit Charge 2 | 10 weeks | Baseline STAI score did not relate to sensor-measured physical activity or sleep over the ensuing 10 weeks. | 8 |
Adler et al [ ] | Fitbit Charge 2 | 14 months | Quarterly measurements of change in depressive symptoms related to measured STC , sleep, and HR . | 7 |
Jevsevar et al [ ] | WHOOP | 12 weeks | Being in the operating room related to the next day HRV . Device reported sleep related to next-day HRV. Relationship between baseline burnout score and device measurements not reported. | 8 |
Silva et al [ ] | Microsoft Smart Band 2 | 2 weeks | Stress and HRV were both significantly different between the baseline and stress condition | 8 |
Mendelsohn et al [ ] | Fitbit Charge | 14 days | Baseline burnout score did not relate to average daily sleep or STC over the ensuing 14 days. | 7 |
Marek et al [ ] | Fitbit Charge HR | 16 weeks | Average daily sleep and activity level over a 2-4–week period did not relate to single-item burnout measure score. Average daily resting HR over a 2-4–week period was higher among residents with burnout versus those without burnout | 8 |
Sochacki et al [ ] | WHOOP | 4 weeks | No significant association between weekly burnout score and device-measured hours of sleep over 4 weeks. | 8 |
Chaukos et al [ ] | Basis Health Tracker | 6 months | No association between baseline depressive symptoms or stress levels and device-measured sleep or activity levels over 30 or 90 days of the study. No association between chronic burnout (burnout at 2 time points), never burned out, new burnout (burnout at 2nd but not 1st time point), and unknown burnout status (survey not completed) and devise measured sleep or activity level aggregated over first 30 days. | 6 |
de Looff et al [ ] | Empatica E4 | 1 day or night shift | Skin conductance collected over 1 shift among nursing staff did not correlate with burnout scores collected on questionnaires completed within 2 days of wearing the device (mean 2.4, SD 10 days; range 0-44 days). | 8 |
Weenk et al [ ] | HealthPatch | Up to 3 days (at least 2) | Stress measured by the patch increased during surgery, more so for less experienced trainees, but did not correlate with change in STAI score before or after surgery, perhaps due to small sample size or lack of sensitivity to change. | 8 |
a STAI: State-Trait Anxiety Inventory.
c HR: heart rate.
d HRV: heart rate variability.
Data extraction was mostly completed by a single researcher (MB). Other researchers (APA and SS) helped refine data extraction and review the tables. The following information was extracted from the papers and is included in Tables 1 and 2 : sample population (size and occupation), anxiety, burnout, stress or depression assessment instrument, additional measurements used, wearable device used, measured physiological variable, study duration, primary findings, and the author-determined quality assessment score.
The methodological quality of nonrandomized or observational studies was assessed by 2 reviewers (MB and SS) using the Newcastle Ottawa Quality Assessment Form for Cohort Studies [ 42 ]. The Newcastle-Ottawa Scale is a validated scale of 8 items in 3 domains: selection, comparability, and outcome. Studies are rated from 0 to 9, with those studies rating 0-2 (poor quality), 3-5 (fair quality), and 6-9 (good or high quality). All 10 studies received a Newcastle-Ottawa Scale rating of good or high quality.
Among the 10 reviewed studies, 8 were conducted in the United States, 1 study was conducted in Portugal [ 35 ], and another one was conducted in Canada [ 36 ]. Seven studies recruited either resident physicians (postgraduate medical trainees), practicing physicians, or a combination of both, primarily within the same specialty (eg, orthopedic surgery and emergency medicine). Two studies recruited registered nurses [ 32 , 40 ] and 1 study recruited medical students [ 35 ]. Sample sizes ranged from 20 to 775 participants per study (see Table 1 ). Only 3 studies had more than 100 participants [ 32 , 33 , 40 ].
Table 1 summarizes the sample population, sample size, physiological variables collected from wearable devices, and psychometrics used in the 10 studies. The devices used, length of data collection, and primary findings are listed in Table 2 . Out of the 10 studies, 9 used wrist-worn biosensors, such as the Fitbit Charge (n=4) [ 32 , 33 , 35 , 40 ] WHOOP (n=2) [ 34 , 38 ], Basis B1 (n=1) [ 35 ], Empatica E4 (n=1) [ 40 ], and the Microsoft Smart Band 2 (n=1) [ 35 ]. Sensors embedded within wrist-worn biosensors included optical heart sensors, electrical heart sensors, accelerometers, and skin temperature sensors. The other device used was a HealthPatch, an adhesive patch with 2 ECG electrodes used to measure HR and HRV. A variety of physiological variables were collected, with sleep being the most common, measured in 7 studies. Studies ranged in length of data collection, from a single 12-hour shift to a 14-month period. Only 5 studies collected data for more than 10 weeks [ 32 - 34 , 37 , 39 ].
Only 2 studies explicitly stated the sampling frequency used when processing data from the wearable device [ 33 , 39 ]. Four of the studies discussed how the data were processed; however, the level of detail varied [ 32 , 33 , 35 , 40 ]. Three of the studies indicated the cutoff values for physiological variables or explained how outliers were addressed [ 32 , 33 , 40 ]. Only 4 studies explicitly stated how much raw data were retrieved from the devices [ 32 - 34 , 36 ].
Of the 10 included studies, 6 included a measure of burnout ( Table 1 ) [ 34 , 36 - 40 ]. Four of these 6 studies used the Maslach Burnout Inventory–Human Services Survey (MBI-HSS) [ 43 ]. In a cross-sectional study of 114 nurses, no relationship was found between MBI-HSS score and skin conductance, a measure of autonomic nervous activity, collected through an Empatica E4, for 1 shift [ 40 ]. Another study investigated the relationship between MBI-HSS score, self-reported work hours, physical activity, and sleep, as measured by a Fitbit, in a cohort of 59 residents [ 36 ]. No relationship was found between the change in burnout score and data collected from the Fitbit over 2 weeks. In the third study, no relationship was found between MBI-HSS score and sleep, as measured by a WHOOP, over the course of 4 weeks [ 38 ]. Last, in a study of 75 medicine and psychiatry residents, no relationship was found between burnout score and sleep or activity levels, as measured by Basis B1 health-tracking device, during their first 6 months of residency [ 39 ].
Two studies measured burnout using scales other than the 22-item MBI-HSS (widely considered the gold standard) [ 34 , 37 ]. In a study of 21 orthopedic residents and surgeons, no association was found between baseline abbreviated MBI scores and WHOOP measures collected over 12 weeks [ 34 ]. The final study investigated the association between burnout, as measured by a commonly used single-item measure, and sleep and activity level, as measured by a Fitbit. In this study, of 28 emergency medicine residents, there was no association between burnout scores and sleep or activity levels over the course of the 16-week study [ 37 ].
A 14-month study of 775 medical residents found a relationship between depressive symptoms, as measured by the 9-item Patient Health Questionnaire [ 44 ], and step count (STC) and sleep as measured by a Fitbit Charge 2 [ 33 ]. Medical residents whose depressive symptoms worsened over the period of the study had a significantly higher skew in their hourly STC distributions and spent less time in bed than those whose symptoms did not worsen. In a study of 83 medical students, Perceived Stress Scale-4 scores related to HR and HRV, were measured by a Microsoft Smartband 2, at baseline and during an examination [ 35 ].
In a 10-week study of 113 nurses led by Feng et al [ 32 ], no relationship was found between the level of anxiety, as measured by the State-Trait Anxiety Inventory (STAI) [ 45 ], and wearable sensor data (eg, sleep and HR) collected using Fitbit Charge 2 smartwatch. Weenk et al [ 41 ] conducted a study of 20 surgeons and surgical residents who completed an abbreviated version of the STAI before and after performing surgery, and wore a HealthPatch. This adhesive patch calculates stress using an HR and HRV-dependent algorithm for 48 to 72 hours [ 41 ]. There was no correlation found between the STAI score and HealthPatch data.
Seven studies reported data on participant adherence or experience with wearable devices. Chaukos et al [ 39 ] reported that 25 (40%) of their participants wore their device for more than 50% of the time for the first 3 months of the study, while another 13 (21%) participants wore the device for more than 75% of the time for the first 3 months. Other studies, such as one conducted by Sochacki et al [ 38 ] reported that of the 26 participants, 5 did not complete the minimum WHOOP compliance (4 weeks). Surgeons involved in a study by Jevsevar et al [ 34 ] reported a high percentage of device compliance at 83.2% of the total collection window, similar to the 93% compliance rate reported by Mendelsohn et al [ 36 ] and Sochacki et al [ 38 ]. Weenk et al [ 41 ] reported that 6 of 20 individuals experienced problems with their HealthPatch, similar to Marek et al [ 37 ] who reported 1 of 30 participants dropped out due to fitness tracker intolerance. Problems included connection failure (n=2), loss of skin contact (n=2), and skin irritation (n=2). Feng et al [ 32 ] noted similar compliance between day-shift participants and night-shift participants (number of recordings day-shift: mean 44.6, SD 3.1 sessions; night-shift: mean 45, SD 20.2 sessions).
A risk of bias of assessment was completed for the 8 cohort studies and 1 cross-sectional study ( Figure 3 ). While the risk of bias was generally low across the studies, none included a comparison group of participants who did not wear a device.
To our knowledge, this is the first scoping review to investigate the use of wearable technologies for the prediction of burnout, anxiety, depression, and stress in HCPs. Among the 10 studies identified, a range of wearables collected data on HR, HRV, respiratory rate, skin temperature, sleep, and activity levels from a single shift of work and up to 14 months of data collection in relatively small samples of physicians, medical students, and nurses. In these studies, no relationships were found between collected physiological data from wearables and burnout or anxiety. One study reported a relationship between STC, time in bed, and depressive symptoms, and another between HR, HRV, and acute stress (during an examination). Identified studies had methodological limitations, including short duration which limits the capture of naturalistic variations in the workplace stressors.
In this review, 3 studies measured HRV [ 34 , 35 , 41 ] and only 1 found a significant relationship between HRV and acute stress. A previous systematic review involving non-HCPs identified 2 studies demonstrating relationships between HRV and acute stress-induced conditions and 1 study demonstrating a relationship between HRV and stress levels measured by catecholamine levels [ 31 ]. This previous systematic review also identified 1 study where in a setting of laboratory-induced stress, HRV parameters related to STAI score. These studies, however, differed substantially from the ones included in this review. For example, none of them collected physiological data longer than 24 minutes, stress was induced in a laboratory setting (vs occurring naturally in a work setting), and only 1 study compared physiological data with a self-reported stress measure (ie, STAI score).
Given these early findings, further research focusing on the following elements of rigor are warranted. First, the length of observation should be long enough (at least 2 or 3 consecutive quarters of a calendar year) to allow sufficient quanta of wearable data to capture fluctuations in and chronicity of workplace stress. Studies should systematically collect data using validated instruments measuring burnout (eg, MBI-HSS [ 43 ]), depression (eg, Center for Epidemiologic Studies Depression Scale [ 46 ] and Patient Health Questionnaire-9 [ 44 ]), and anxiety (eg, General Anxiety Disorder-7 [ 47 ]). Investigators may also want to consider designing cohorts comprising groups of HCPs defined by their type of medical specialty or practice location. For example, it is possible that workplace stressors, patient acuity, and job demand fluctuate between primary care and surgical specialties and between outpatient practices and hospital-based practices. Hence, the burnout biomarkers may vary between practices. Considering that burnout is defined as when job demands exceed job resources, it is possible that the workplace (eg, patient acuity and hospital bed size) and related staffing factors (eg, workload, shift length, and availability of support staff) impact physiological biomarkers collected from wearables. Hence, future studies should consider collecting organizational variables to better understand the systemic contributors of burnout. Additionally, given the era of decentralized health care practice (eg, nontraditional shift days/hours and remote care with augmented reality), studies engaging with HCPs may benefit from no-contact passive monitoring and a digital app interface for survey collection (ie, decentralized trail). Finally, there is a bioethics component to understand how wearables can be successfully integrated into workforces’ burnout management. Greater attention needs to be paid to participant engagement, including addressing comfort with wearing the device, resolving discrepancies in wearable-derived data versus self-reported data, and understanding factors that influence perceptions of fatigue but not recorded sleep [ 37 , 48 , 49 ].
The use of wearables to detect the functioning states of human beings is an active and rapidly evolving field. Several wearable-based studies have been shown to aid in the detection of mental health conditions or resilience in quality of life [ 50 ] through mindfulness practices including physical activity [ 51 ] and sleep [ 52 - 54 ] monitoring. Prior work has demonstrated that aspects of physical functioning when combined with data during the day could predict variations in aspects of QoL and mental well-being [ 55 - 58 ]. Work by Campbell et al [ 59 - 64 ] has demonstrated the ability of daily journaling, wearables, and mobile assessments to detect depressive symptoms and mental states in patients with schizophrenia. These prior efforts in the field of mental health and the work summarized in this scoping review demonstrate the promise of wearables in predicting states of one’s functioning, including burnout. However, a consensus is lacking on the best approaches to collecting, processing, and reporting physiological data, much like CONSORT (Consolidated Standards of Reporting Trials) [ 65 ] for reporting randomized trials and STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) [ 66 ] guidelines for reporting observational studies. Standardization of variables should include the creation of a guideline for reporting the sampling frequency, device adherence, and other information regarding device parameters that impact data collection. Such standardization would assist with generalizing findings, validating predictive algorithms, informing meta-analysis, and the use of data for retraining predictive models regardless of the wearable’s make and model. Additionally, there needs to be consensus around approaches to address bioethics, privacy, and confidentiality concerns of participants [ 67 , 68 ]. Predictive technologies, informed by personal biometric or physiologic data, may help improve work conditions but could also place individuals’ privacy or perhaps even their job security at risk.
This study has limitations. Only studies that included physicians, resident physicians, medical students, and nurses and were published in English were included. Following the 2019 pandemic, physicians identifying as 2 or more races experienced the highest levels of burnout onset, according to a report by the American Medical Association [ 69 ]. Furthermore, there are known disparities in the access to, and the use of digital health technologies in underrepresented minorities [ 70 , 71 ]. Therefore, it is vital to understand the factors that cause burnout in these groups of professionals and remove barriers to access to personalized wellness technologies using wearables that may help understand and mitigate burnout. In the context of the use and access of digital health for burnout, 8 of the 10 studies reported the gender breakdown of participants, and only 1 study reported the race of their participants. With the urgent need to broaden access to digital health solutions to study and understand burnout, future efforts should (1) follow reporting guidelines (eg, set by National Institutes of Health in the Human Subjects sections) to report on participant characteristics by ethnicity, race, and gender, and (2) innovate study procedures (eg, decentralized protocols) that improve the recruitment and engagement of underrepresented minorities in digital health studies of burnout. Although we sought to include validated measures of burnout, stress, depression, and anxiety, the instruments used in the studies varied in their psychometric strengths. Finally, most studies lacked power calculations, making findings, effect sizes, or impact of dropouts difficult to interpret from the perspective of the generalizability of biomarkers.
Despite the popularity of wearable devices, only 10 studies were identified that explored relationships between physiological data and burnout, depressive symptoms, stress, or anxiety. Most of these studies had substantial methodological limitations, and nearly all reported limited data collection and processing information, participant experience with the wearable device, and device compliance. Standardizing study procedures, common data elements, and reporting of wearable data are needed to strengthen the rigor of digital health studies. Addressing these limitations will result in improvements in wearable device research, including data standardization and reporting, that will validate their use in providing early intervention for HCP wellness. Additional research is warranted to explore the potential of wearable devices, perhaps augmented with other system-level data (eg, work shift lengths and absenteeism), to predict burnout and other forms of distress, hopefully leading to meaningful action before it has an adverse impact on HCPs and patient care.
This study was partially supported by the Mayo Clinic Summer Undergraduate Research Fellowship, National Science Foundation (grant 2041339); National Institutes of Health (grant R01 NR020362); the Mayo Clinic Center for Individualized Medicine, and the Mayo Clinic Center for Clinical and Translational Science. Any opinions, findings, conclusions, or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation or National Institutes of Health.
APA and LND contributed equally as co-Corresponding Authors. APA may be contacted at [email protected].
PC has received research support from the National Institutes of Health (NIH), National Science Foundation (NSF), Brain and Behavior Research Foundation, and the Mayo Clinic Foundation. PC has received research support from Pfizer, Inc. He has received equipment support from Neuronetics, Inc, and MagVenture, Inc. He received grant-in-kind supplies and genotyping from Assurex Health, Inc for an investigator-initiated study. He served as the primary investigator for a multicenter study funded by Neuronetics, Inc and a site primary investigator for a study funded by NeoSync, Inc. PC served as a paid consultant for Engrail Therapeutics, Sunovion, Procter and Gamble Company, Meta Platforms, Inc, and Myriad Neuroscience. PC is employed by the Mayo Clinic. LD is a coinventor of the Well-Being Index and its derivatives which Mayo Clinic has licensed. LD receives royalties. WB’s research has been supported by the NIMH, NINR, NSF, the Blue Gator Foundation, the Watzinger Foundation, and the Mayo Foundation for Medical Education and Research. He has contributed chapters to UpToDate concerning the pharmacological management of patients with bipolar spectrum disorders. MCs research has been supported by NSF and the Mayo Foundation for Medical Education and Research.
Search strategy.
PRISMA-ScR checklist.
Consolidated Standards of Reporting Trials |
health care professional |
heart rate |
heart rate variability |
Maslach Burnout Inventory–Human Services Survey |
State-Trait Anxiety Inventory |
step count |
Strengthening the Reporting of Observational Studies in Epidemiology |
Edited by T de Azevedo Cardoso; submitted 24.06.23; peer-reviewed by T Pipe, P Punda; comments to author 01.12.23; revised version received 01.01.24; accepted 20.03.24; published 25.06.24.
©Milica Barac, Samantha Scaletty, Leslie C Hassett, Ashley Stillwell, Paul E Croarkin, Mohit Chauhan, Sherry Chesak, William V Bobo, Arjun P Athreya, Liselotte N Dyrbye. Originally published in the Journal of Medical Internet Research (https://www.jmir.org), 25.06.2024.
This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research, is properly cited. The complete bibliographic information, a link to the original publication on https://www.jmir.org/, as well as this copyright and license information must be included.
Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Nature Communications volume 15 , Article number: 4759 ( 2024 ) Cite this article
20k Accesses
2101 Altmetric
Metrics details
Parkinson’s disease is increasingly prevalent. It progresses from the pre-motor stage (characterised by non-motor symptoms like REM sleep behaviour disorder), to the disabling motor stage. We need objective biomarkers for early/pre-motor disease stages to be able to intervene and slow the underlying neurodegenerative process. Here, we validate a targeted multiplexed mass spectrometry assay for blood samples from recently diagnosed motor Parkinson’s patients ( n = 99), pre-motor individuals with isolated REM sleep behaviour disorder (two cohorts: n = 18 and n = 54 longitudinally), and healthy controls ( n = 36). Our machine-learning model accurately identifies all Parkinson patients and classifies 79% of the pre-motor individuals up to 7 years before motor onset by analysing the expression of eight proteins—Granulin precursor, Mannan-binding-lectin-serine-peptidase-2, Endoplasmatic-reticulum-chaperone-BiP, Prostaglaindin-H2-D-isomaerase, Interceullular-adhesion-molecule-1, Complement C3, Dickkopf-WNT-signalling pathway-inhibitor-3, and Plasma-protease-C1-inhibitor. Many of these biomarkers correlate with symptom severity. This specific blood panel indicates molecular events in early stages and could help identify at-risk participants for clinical trials aimed at slowing/preventing motor Parkinson’s disease.
Introduction.
Parkinson’s disease (PD) is a complex and increasingly prevalent neurodegenerative disease of the central nervous system (CNS). It is clinically characterised by progressive motor and non-motor symptoms that are caused by α-synuclein aggregation predominantly in dopaminergic cells, which leads to Lewy body (LB) formation 1 . The failure of neuroprotective strategies in preventing disease progression is due, in part, to the clinical heterogeneity of the disease—it has several phenotypes—and to the lack of objective biomarker readouts 2 . To facilitate the approval of neuroprotective strategies, governing agencies and pharmaceutical companies need regulatory pathways that use objectively measurable markers—potential therapeutical targets as well as state and rate biomarkers—directly associated with PD pathophysiology and clinical phenotypes 3 .
The recently emerged α-synuclein seed amplification assays (SAA) can identify α-synuclein pathology in vivo and support stratification purposes but still rely on cerebrospinal fluid (CSF) obtained through relatively invasive lumbar punctures 4 . Therefore, this test remains specialised and not readily suitable for large-scale clinical use. As peripheral fluid biomarkers are less invasive and easier to obtain, they could be used in repeated and long-term monitoring, which is necessary for population-based screenings for upcoming neuroprotective trials. While the only emerged serum biomarker in the last years, axonal marker neurofilament light chain (NfL), increases longitudinally and correlates with motor and cognitive PD progression 5 , it is non-specific to the disease process.
Growing data support evidence of PD pathology in the peripheral system, which increases the likelihood of finding a source of matrices for less invasive biomarkers. We know α-synuclein aggregation induces neurodegeneration, which is propagated throughout the CNS. Evidence indicates that additional inflammatory events are an early and potentially initial step in a pathophysiological cascade leading to downstream α-synuclein aggregation that activates the immune system 6 . Inflammatory risk factors in circulating blood (i.e. C-reactive-protein and Interleukin-6 and α-synuclein-specific T-cells) are associated with motor deterioration and cognitive decline in PD 7 , 8 . These inflammatory blood markers can even be identified in plasma/serum samples of individuals with isolated REM sleep behaviour disorder (iRBD), the early stage of a neuronal synuclein disease (NSD), and the most specific predictor for PD and dementia with Lewy bodies (DLB) 6 . NSD was recently proposed as a biologically defined term, for a spectrum of clinical syndromes, including iRBD, PD and DLB, that follow an integrated clinical staging system of progressing neuronal α-synuclein pathology (NSD-ISS) 9 .
In this study, we used mass spectrometry-based proteomic phenotyping to identify a panel of blood biomarkers in early PD. In the initial discovery stage, we analysed samples from a well-characterised cohort of de novo PD patients and healthy controls (HC) who had been subjected to rigorous collection protocols 10 . Using unbiased state-of-the-art mass spectrometry, we identified putatively involved proteins, suggesting an early inflammatory profile in plasma. We thereafter moved on to the validation phase by creating a high-throughput and targeted proteomic assay that was applied to samples from an independent replication cohort, consisting of de novo PD, HC and iRBD patients. Finally, after refining the targeted proteomic panel to include a multiplex of only the biomarkers which were reliably measured, an independent analysis was performed on a larger and independent cohort of longitudinal, high-risk subjects who had been confirmed as iRBD by state-of-the-art video-recorded polysomnography (vPSG), including follow-up sampling of up to 7 years.
In summary, using a panel of eight blood biomarkers identified in a machine-learning approach, we were able to differentiate between PD and HC with a specificity of 100%, and to identify 79% of the iRBD subjects, up to 7 years before the development of either DLB or motor PD (NSD stage 3). Our identified panel of biomarkers significantly advances NSD research by providing potential screening and detection markers for use in the earliest stages of NSD for subject identification/stratification for the upcoming prevention trials.
We performed a bottom-up proteomics analysis of plasma, which had been depleted of the major blood proteins, using two-dimensional in-line liquid chromatography fractionation into ten fractions and label-free mass spectrometric analysis by QTOF MS E . The discovery cohort consisted of ten randomly selected drug-naïve patients with PD and ten matched HC from the de novo Parkinson’s disease (DeNoPa 10 ) cohort (details can be found in Supplementary Table 1 ). This analysis identified 1238 proteins when restricting identification to originate from at least one peptide per protein and at least two fragments per peptide. After excluding proteins with less than two unique peptides or with an identification score below a set threshold (see method section below), 895 distinct proteins remained. Of these proteins, 47 were differentially expressed between the de novo PD and control groups on a nominal significance level of 95%. Pathway analysis suggested enrichment in several inflammatory pathways. Workflow and Results are shown in Fig. 1 , and 2 Supplementary Figs. 1 , 2.
The study included three phases. Phase 0 consisted of discovery proteomics by untargeted mass spectrometry to identify putative biomarkers, followed by phase I in which targets from the discovery phase were transferred to a targeted, mass spectrometric MRM method and applied to a new and larger cohort of samples, and finally phase II in which the targeted MRM method was refined and a larger number of samples were analysed to evaluate the clinical feasibility of the targeted protein panel.
The circle radii in the Volcano plot represent the identification certainty, where large radii represent proteins identified by at least two unique peptides and an identification score >15, smaller radii are given for proteins identified by two or more unique peptides or a confidence score >15. The horizontal axis shows log 2 of the average fold-change and the vertical axis shows −log 10 of the p values. The significantly different proteins are annotated by gene name and coloured in pink, while the non-significant proteins are coloured in grey. GO annotations for the significant proteins are shown, the dashed line represents p = 0.05. Disease and function annotations from IPA are shown, divided into annotations with a positive or negative activation score. Source data are provided as a Source Data file.
We next developed a validatory, high-throughput and multiplexed, mass spectrometric targeted proteomic assay based on the potential biomarkers identified in the discovery phase. Additional proteins were also included in the assay, several of which had been identified in previous discovery studies of PD, Alzheimer’s disease (AD), and ageing 11 . In addition, we also included several known pro- and anti-inflammatory proteins identified in the literature 12 , 13 , 14 , 15 , which had been previously developed into an in-house targeted proteomic neuroinflammatory panel. Using this approach, we created a targeted proteomic panel, including biomarkers from current scientific developments and preliminary findings from our own work 16 , 17 . This targeted proteomic and multiplexed assay included 121 proteins and aimed to validate biomarkers and probe the pathways identified as being perturbed in the discovery phase. Details can be found in Supplementary Table 2 and Fig. 3 .
Workflow and overview of the results of the targeted proteomic analysis of de novo Parkinson’s disease (PD) subjects, healthy controls (HC), and the validation cohorts of other neurological disorders (OND) and isolated REM sleep behaviour disorder (iRBD). A A targeted mass spectrometric proteomic assay was developed and optimised. The assay was then applied to plasma samples from cohorts comprising de novo PD ( n = 99) and HC ( n = 36), and validated in patients with OND ( n = 41) and prodromal subjects with iRBD ( n = 18). The protein expression difference between the groups was compared using Mann–Whitney’s two-sided U -test with Benjamini–Hochberg FDR adjustment at 5%. The lollipop charts show the log 10 p values, signed according to fold-changes. Pink icons represent a protein upregulated in an affected group and grey represents a protein upregulated in controls. B Significantly differentially expressed proteins in the comparison between de novo PD and healthy controls. C Significantly differentially expressed proteins between iRBD, OND and HC. Source data are provided as a Source Data file.
For the targeted proteomics analysis, we used plasma samples, independent from the proteomic discovery step, from 99 individuals recently diagnosed with de novo PD (48 men, 50%, mean age 67 years) and 36 healthy controls (HC; 20 men, 57%, mean age 64 years). This was the main cohort, to which we added further samples for validation that consisted of a heterogeneous group of 41 patients with other neurological diseases (OND) (29 men, 71%, mean age 70 years) and 18 patients with vPSG-confirmed iRBD (10 men, 56%, mean age 67 years). Further details can be found in Table 1 and Fig. 3 .
Our targeted proteomic assay was developed for 121 proteins, 32 of which we consistently and reliably detected in plasma. Of these 32 markers, 23 were confirmed as being significantly and differentially expressed between PD and HC. We identified six differentially expressed proteins in the comparison between iRBD patients and HC and between OND and HC (Fig. 3 ). Both the de novo PD and iRBD groups demonstrated an upregulated expression of the serine protease inhibitors SERPINA3, SERPINF2 and SERPING1, and of the central complement protein C3. Granulin precursor protein was shown to be downregulated in all three patient groups (PD, iRBD and OND) compared to HC. The OND and PD groups had a shared and upregulated expression of the proteins PTGDS, CST3, VCAM1 and PLD3. Detailed information about the diagnoses of the OND group can be found in Table 1 , and detailed information about the proteins can be found in Supplementary Table 2 . Figure 4 shows the significantly different proteins as Box-scatter plots.
The data are displayed as Box and Whisker plots overlaid with scatter plots of the individual measurements. The whiskers show the minimum and maximum, and the boxes show the 25th percentile, the median and the 75th percentile. The protein expression difference between the groups was compared using Mann–Whitney’s two-sided U -test with Benjamini–Hochberg multiple testing correction (FDR adjustment at 5%). ns not significant, * p < 0.05, ** p < 0.01, *** p < 0.001 and **** p < 0.0001. The proteins are represented by gene names. Source data are provided as a Source Data file.
The involvement of the differentially expressed proteins and their impact on biological processes were evaluated using pathway analysis (Ingenuity Pathway Analysis [IPA], Qiagen). The significantly differentially expressed proteins between PD and HC were used as input, with a fold-change set as the expression observation. We considered pathways as significant if they had an enrichment p value <0.05. At least two of the input proteins were included. Three major pathway clusters were identified and consisted of (i) the expression of serine protease inhibitors or serpins and complement and coagulation components, (ii) endoplasmic reticulum (ER) stress/heat shock-related proteins and (iii) the expression of VCAM1, SELE and PPP3CB. The highest enrichment scores were identified in the pathways acute phase response signalling ( p = 7.8 E −10 ), coagulation system ( p = 7.4 E −6 ), complement system ( p = 8.1 E −6 ), LXR/RXR activation ( p = 9.1 E −6 ), FXR/RXR activation ( p = 9.8 E −6 ) and glucocorticoid receptor signalling ( p = 2.0 E −5 ). These are all pathways involved in inflammatory responses. We also identified pathways related to the unfolded protein response ( p = 0.004) and neuroinflammation ( p = 0.04), although with lower enrichment scores. For details, see Supplementary Fig. 1 .
Inflammation-related pathways (including both the complement system and the acute phase response) demonstrated the highest significance levels, followed by pathways regulating protein folding, ER stress, and heat shock proteins. A network representation of proteins and pathways showed clusters consisting of inflammation/coagulation/lipid metabolism (FXR/RXR and LXR/RXR), heat shock proteins/protein misfolding, and more heterogenous pathway clusters related to Wnt-signalling and extracellular matrix proteins. Figure 5 illustrates the potential detrimental and protective mechanisms suggested to be taking place based on the protein expressions observed in this study, leading to oligomerisation and accumulation of α-synuclein in neuronal Lewy body inclusions and, finally, dopaminergic neuronal cell loss.
Oligomerisation and accumulation of α-synuclein in Lewy body inclusions is a key process in the pathophysiology of neuronal synuclein disease, i.e. Parkinson’s disease and dementia with Lewy bodies from aggregation and accumulation, the pathological pathway includes different steps finally leading to the loss of dopaminergic neurons. Protective and detrimental mechanisms influence these processes, based on the differently expressed protein profiles, assessed by targeted mass spectrometry in our study. Detailed information about the proteins can be found in Supplementary Table 2 .
Principal component analysis (PCA) demonstrated that the HC and PD groups formed two clusters separate from each other over the first and second principal components (PC), attributed with 23.5% and 13.9% of the model’s total variance, respectively. The iRBD group was situated in the middle of HC and PD, and the OND group varied considerably with no evident clustering, as expected due to the heterogeneity of diseases. The corresponding loadings of PC1 and PC2 demonstrated that those with PD correlated with lower levels of PPP3CB, DKK3, SELE and GRN, and higher levels of most of the other proteins. The loadings plot had a high level of covariation in the expression of the PPP3CB, DKK3 and SELE proteins, which were all downregulated in PD. These proteins correlated negatively with the expression of SERPINs, complement C3 and HPX, which all showed a high degree of covariation, and were upregulated in the PD group. Data are displayed in Supplementary Fig. 2 .
We next applied machine learning to construct a discriminant OPLS-DA model using the PD and HC samples from the validation phase. The samples clustered into two distinct and well-separated classes, and evaluation of the model showed that it was highly significant ( p = 2.3E −27 permutations p = <0.001). The proteins with the greatest influence on the class separations were GRN, DKK3, C3, SERPINA3, HPX, SERPINF2, CAPN2, SERPING1 and SELE. We predicted the iRBD samples in the model, which resulted in 13 subjects classified as PD (72%) and five not belonging to either group. None of the iRBD samples were classified as controls. We additionally predicted the OND samples, out of which nine were classified as HC, 12 as PD and 19 were not classified as belonging to either group. The 12 samples predicted as PD did not demonstrate enrichment according to the OND groups. The random distribution of the OND samples between PD and HC indicates that the heterogenous group of OND individuals does not share a distinct protein expression with either the HC or PD groups. The iRBD samples that were classified as PD, and not as HC, strongly suggest a shared proteomic profile between iRBD and the protein expression observed in the newly diagnosed PD patients.
We subsequently explored if the observed protein expressions could be used to build a regression model capable of predicting whether individuals belonged to the PD or HC groups. We identified a panel of proteins that discriminated between PD and HC with 100% accuracy and then constructed a linear support vector classification model and applied recursive feature elimination to pinpoint the most discriminating variables. The data were divided into two parts: one consisting of 70% for model training and one containing 30% for testing. The proportion of PD and control samples was maintained in each part. The number of features included in the model was determined by feature ranking with cross-validated recursive feature elimination in the training dataset. The feature selection resulted in a model with eight predictors: GRN, MASP2, HSPA5, PTGDS, ICAM1, C3, DKK3 and SERPING1. The training data were predicted in the model and resulted in all samples being classified in the correct class. We further constructed receiver operating characteristic (ROC) and precision-recall (PR) curves to illustrate the ability of each protein to distinguish between PD and HC and compared this with the ability of the combined multiplexed protein panel. The combined panel achieved an AUC of 1.0 on both ROC and PR curves. The AUC of the individual predictors ranged from 0.53 to 0.92 in the ROC curve, and from 0.79 to 0.96 in the PR curve (Fig. 6 ). We further evaluated the whole dataset by performing repeated cross-validation with six splits of the data and 40 repetitions. The resulting classification metrics (Supplementary Fig. 3 ) demonstrated average and standard deviation for precision, recall, F1 score, and balanced accuracy score of 0.87 ± 0.09, 0.87 ± 0.08, 0.86 ± 0.09 and 0.82 ± 0.12, respectively, thereby indicating a highly robust classification model. Testing the model’s specificity for PD, we predicted the heterogenous group of OND, resulting in 26 of the 42 samples being classified as PD-like. Prediction of the prodromal iRBD group resulted in 17 of 18 samples being classified as PD-like. We compared the prediction of the OND and iRBD samples between the OPLS-DA and SVM models, finding that most of the samples were classified in the same group in both models (out of the samples with a classification in the OPLS-DA model: 82% in OND and 100% in iRBD). The proportion of iRBD samples classified as PD in our models (72% in the OPLS-DA model and 94% in the SVM model) is in line with clinical evidence based on longitudinal cohort studies, reporting that over 80% of iRBD subjects will develop an advanced NSD with motor impairment and/or cognitive decline 18 . We evaluated the influence of age and sex on the proteins included in the support vector model and found that neither influenced the model’s classification ability (see Supplementary Methods 2 for details).
The model was trained on 70% of the samples to establish the most discriminating features. Applying cross-validated recursive feature elimination, the top predictors were determined as a granulin precursor, mannan-binding lectin-serine peptidase 2, endoplasmic reticulum chaperone-BiP, prostaglandin-H2 d -isomerase, intercellular adhesion molecule-1, complement C3, dickkopf-3 and plasma protease C1 inhibitor. The remaining 30% of samples were predicted in the model and resulted in 100% prediction accuracy. Receiver operating characteristics (ROC) and precision-recall (PR) curves of the individual and combined proteins in the test set demonstrated that the individual proteins achieved ROC area under the curve (AUC) values 0.53–0.92 and PR values 0.79–0.96, while the combined predictors reached an area under the curve = 1.0. Source data are provided as a Source Data file.
To evaluate the results from the initial prediction models focusing on at-risk subjects, we developed and refined our targeted and multiplexed proteomic test to quantitate only those proteins that were readily and reliably detectable from the initial targeted proteomic assay ( n = 32). Next, we analysed an additional set of 146 longitudinal samples from an independent cohort of 54 individuals with iRBD. This cohort was available from continuing recruitment at the same centre and consisted of longitudinally followed iRBD subjects. Deep phenotyping revealed 100% (54/54) had RBD on PSG, 88.9% (48/54) had hyposmia as identified with the Sniffin’ Stick Identification Test, and 91.7 % (22/24) had neuronal α-synuclein positivity as shown by α-synuclein Seed Amplification Assay (SAA) in cerebrospinal fluid (CSF) 19 . Longitudinal follow-up was available for up to 10 years, during which 16 subjects (20%) phenoconverted to either PD ( n = 11) or dementia with Lewy bodies (DLB; n = 5). Since only serum samples were available from the independent replication cohort (further details can be found in Supplementary Table 3 ), we investigated how the proteins in our assay correlated between plasma, serum, and CSF and found good correlations between plasma and serum, but poor correlations between these blood matrices and CSF. The limited correlations between blood and CSF proteins correspond to those of other studies comparing the protein expression between plasma/serum and CSF 20 , 21 and underscore that our test does not necessarily reflect a prodromal and PD-specific proteomic signature of the protein expression in the CSF in proximity to the brain, but rather shows an earlier change in the blood protein expression between healthy status and very early PD patients (Details from this comparison can be found in Supplementary Methods 1 and Supplementary Fig. 4 ).
We applied all available longitudinal iRBD samples ( n = 146) from phase II to the two machine-learning models (OPLS-DA and support vector machine) constructed in phase I (PD vs. HC). The OPLS-DA model, based on all 32 detected proteins, identified 70% of the iRBD samples as PD, while the SVM model, which was based on a panel of eight proteins, identified 79% of the samples as PD. As mentioned above, at the time of analysis, 16 of the 54 subjects in our longitudinal iRBD validation cohort had developed PD/DLB. The earliest correct classification was 7.3 years prior to phenoconversion and the latest was 0.9 years prior to diagnosis (average 3.5 ± 2.4 years). Detailed information can be found in Fig. 7 and Supplementary Methods 3 .
146 new serum samples from individuals diagnosed with iRBD, several with longitudinal follow-up samples, were predicted in the OPLS-DA model. 70% of the samples were predicted as Parkinson’s disease (PD), and 23 of 40 individuals had all their longitudinal samples predicted as PD. In the more refined support vector machine (SVM) model, 79% of the 146 new samples were predicted as PD and 27 of 40 individuals consistently had all their longitudinal samples predicted as PD. Source data are provided as a Source Data file.
We next evaluated the relationship between proteins and clinical data by correlating the protein expression in PD and HC (from phase I) with clinical scores (Mini-Mental State Examination [MMSE], Hoehn & Yahr stage [H&Y] and UPDRS [Unified Parkinson’s Disease Rating Scale; I–III, and total score]). We found negative correlations for GRN, DKK3, PPP3CB, and SELE with H&Y and UPDRS parts II, III, and total score, possibly indicating a connection between a more severe clinical (especially motor) impairment and lower expression of markers in the Wnt-signalling pathways (DKK3 and PPP3CB). Higher Cystatin C plasma levels correlated with higher numbers in UPDRS part III (motor performance) and UPDRS total score. The same was found for PTGDS plasma levels, which were also negatively correlated with MMSE. The central complement cascade protein, C3, negatively correlated with MMSE, and positively correlated with H&Y, UPDRS part III, and total score. The UPR-regulating protein BiP (HSPA5) correlated negatively with MMSE, and positively with H&Y and UPDRS parts II, III, and total score. The ERAD-associated proteins, HSPAIL and adiponectin, were positively correlated with H&Y, and UPDRS parts II, III, and total score. SERPINs (SERPINA3, SERPINF2 and SERPING1) and hemopexin (HPX) correlated negatively with MMSE and positively with H&Y and UPDRS parts II, III, and total score. In general, the MMSE score was inversely correlated with H&Y stage and UPDRS scores. For detailed information, see Fig. 8 and Table 2 .
The correlation was performed using Spearman’s procedure, and the clustering method was set to average. The clustering metric was Euclidean. The heatmap is coloured by correlation coefficient where red represents positive and blue negative correlations. The proteins are represented by gene names. Detailed information about the protein correlations can be found in Supplementary Table 3 . De novo Parkinson’s disease ( n = 99) and healthy controls ( n = 36). MMSE mini-mental state examination, UPDRS unified Parkinson’s disease rating Scale. Source data are provided as a Source Data file.
The longitudinal expression in the iRBD samples was evaluated using linear mixed-effects models. Conditional growth models with random slopes and random intercepts between the individuals were constructed. After adjusting the p values for multiple testing by applying the Benjamini–Hochberg (BH) procedure with alpha = 0.05, we found that Butyrylcholinesterase (BCHE) was significantly decreased over the timepoints in the iRBD individuals ( p = 0.01). We next focused only on the iRBD samples with at least two timepoints and for which PD had consistently been predicted in the SVM model ( n = 90). This produced comparable results to the initial model with BCHE significantly related with time since baseline ( p = 0.01), but also TUBA4A was nominally significantly increased ( p = 0.04) although not passing the BH FDR threshold. The modelling also demonstrated that the clinical measurements H&Y ( p = 0.02), UPDRS I–III ( p = 0.02), and UPDRS I and III ( p = 0.03 and 0.03, respectively), were significantly related to the time since baseline in the iRBD group post multiple testing correction. PD non-motor symptoms, as measured on the PD NMS sum score, were strongly correlated with longitudinal motor progression ( p = 5E −8 ). Similarly, the questionnaire for quality of life PDQ-39’s mean values also correlated with longitudinal motor progression ( p = 0.005). From available routine blood values, cholesterol was associated with longitudinal timepoints ( p = 0.02). Details can be found in Supplementary Table 4 . Correlating the clinical measurements with the targeted proteomic data, we applied Spearman’s correlation and found that cholesterol was positively correlated with six of the identified proteins (Supplementary Table 5 ), including HSPA8, APOE and MASP2 ( p = 5E −9 , 0.0003 and 0.003, respectively). Also significantly correlated, but to a lesser degree and not passing the BH FDR threshold, were the PD NMS sum which correlated negatively with TUBA4A (p unadjusted = 0.01) and the PDQ-39 mean values, which correlated negatively with CST3 and PTGDS ( p unadjusted = 0.03 and 0.05, respectively).
PD has emerged as the world’s fastest-growing neurodegenerative disorder and currently affects close to 10 million people worldwide. Consequently, there is an urgent need for disease-modifying and prevention strategies 22 , 23 . The development of such strategies is hampered by two limitations: there are major gaps in our understanding of the earliest events in the molecular pathophysiology of PD, and we lack reliable and objective biomarkers and tests in easily accessible bio-fluids. We, therefore, need biomarkers that can identify PD earlier, preferably a significant time before an individual develops significant neuronal loss and disabling motor and/or cognitive disease. Such biomarkers would advance population-based screenings to identify individuals at risk and who could be included in upcoming prevention trials.
In the last years, CSF SAA emerged as the most specific indicator for NSD, in prodromal stages like iRBD, with an impressively high sensitivity and specificity of up to 74 and 93%, respectively, across various cohorts 9 , 24 . Despite the many questions surrounding SAA that need to be answered, including the ultimate understanding of its functionality, it is a true milestone for advancing prevention trials. It is, however, hampered by having only been shown to be robust in CSF and by the slow development and high variability of SAA in peripheral blood 25 , as well as by the lack of quantification capabilities. An easier and more accessible biofluid test would enable screening large population-based cohorts for at-risk status to develop an NSD. Therefore, the identification of additional biomarkers is needed, as is further knowledge of the biomarkers and pathways of the underlying pathophysiology (e.g. inflammation) during the earliest stage of NSD.
Other emerging multiplex technologies are increasingly used to identify individual proteomic biomarkers. However, these techniques are not true proteomic or ‘eyes open’ methods, as they rely on selected large panels of specific antibodies/and other (e.g. aptamer)-based assay technologies. These techniques, although useful, have not provided consistent results 3 , 26 . Proteomics using mass spectrometry measures all expressed proteins in an unbiased fashion as opposed to those selectively included in a panel that also includes variability due to cross-reactivity. Therefore, proteomic screening using mass spectrometry-based techniques is much more likely to identify pathways or biomarkers and provides more meaningful insights into the disease mechanisms involved in PD. We found a discrepancy between the detected markers during the discovery and the targeted phases. This is a known phenomenon in biomarker translation 27 that is also reflected in the low number of biomarkers having received FDA approval 28 . We addressed this by using previously reported successful improvement strategies in proteomic approaches, namely by refining our panel, reducing the number of markers, and increasing the sample size 29 . Furthermore, the validation of potential biomarkers was performed on a second and different type of mass spectrometer (triple quadrupole), which has the advantage of being available in all large hospitals.
Targeted MS has been previously applied in PD, including by the current authors, but the biological fluid used in the majority of studies is CSF 30 and not peripheral fluids such as blood. Here we demonstrate that even with a very low required volume of plasma/serum (10 µl) targeted proteomic is feasible.
The targeted proteomic assay presented here was developed from proteins identified in an unbiased discovery study, from our previous research, and from the literature. It included several inflammatory markers, Wnt-signalling members, and proteins indicative of protein misfolding. When analysing PD, OND, iRBD and HC in the targeted proteomic validation phase, we identified and confirmed 23 distinct and differentially expressed proteins between PD and HC. Our analysis moreover demonstrated that iRBD possesses a significantly different protein profile compared to HC, consisting of decreased levels of GRN and MASP2 and increased levels of the complement factor C3 and SERPINs (SERPINA3, SERPINF2 and SERPING1), thus indicating early involvement of inflammatory pathways in the initial pathophysiological steps of PD. Comparing these results to previous findings by our and other groups 8 , 31 highlights the link between these proteins and the pathways of complement activation, coagulation cascades, and Wnt-signalling.
By applying machine-learning models, we classified and separated de novo PD or control samples with 100% accuracy based on the expression of eight proteins (GRN, MASP2, HSPA5, PTGDS, ICAM1, C3, DKK3 and SERPING1).
With an independent validation, we added (a) a larger sample set and (b) longitudinal samples from the most interesting subgroup with 54 iRBD subjects and a total of 146 serum samples. We were able to validate our previous panel with a high prediction rate (79%) of these individuals as seen in PD in the targeted approach. Interestingly, the biomarker panel itself did not correlate with longitudinal expression but remained robust after the initial classification of iRBD. So far, 16 of the 54 iRBD subjects converted to PD/DLB (stage 3 NSD). Out of these samples, the SVM model predicted ten individuals with all their timepoints classified as PD, and of the 11 iRBD subjects who converted to PD/DLB, eight were identified as PD by the proteome analysis. Our panel, therefore, identified a PD-specific change in blood up to 7 years before the development of the stage 3 NSD.
The main shortcoming with many previously explored PD biomarkers is weak or no correlation with clinical progression data. So far, outcome measures in clinical trials are primarily based on motor progression, often by a clinical rating scale such as the UPDRS and/or wearable technologies. More objective biomarkers correlating with or reflecting the progression of the pathophysiology and clinical symptoms would be of the utmost importance. We, therefore, calculated correlations with clinical parameters and identified an association with multiple markers, including DKK3, PPP3CB and C3, indicating downregulation of Wnt-signalling pathways. Increased activity of the complement cascade correlated with higher scores in symptom severity (UPDRS part III and total score) and lower scores in cognitive performance (MMSE).
Protein (i.e. α-synuclein) misfolding is a well-known component of PD pathology and is believed to be the key factor behind Lewy body formation 32 . The transport of excessive amounts of misfolded proteins or increased folding cycles can induce ER stress. A cellular defence mechanism to alleviate ER stress is the unfolded protein response (UPR) reducing ER protein influx and increasing protein folding capacity 33 . The UPR is mainly activated by BiP-bound misfolded proteins 34 . The higher expressed markers HSPA5 (UPR-regulating protein BiP) and HSPA1L in our plasma samples of early PD indicate ER stress as a significant factor in the disease process and has been previously linked to PD in both mouse models and brain tissue studies 35 , 36 .
As mentioned by other groups and confirmed in our results, increasing evidence suggests inflammation is a specific feature in early PD. Complement activation has been associated with the formation of α-synuclein and Lewy bodies in PD and deposits of the complement factors iC3b and C9 have been found in Lewy bodies 37 . C3 is a central molecule in the complement cascade and was highly upregulated in blood in both PD and both independent iRBD sample sets analysed in this study. This upregulation in the earliest phase of motor PD (stage 3 NSD), and even in the prodromal phase (stage 2 NSD), clearly indicates inflammation as an early, if not the initial, event in PD neurodegeneration. Complement C3 levels correlated positively with indicators of motor dysfunction (H&Y stage and UPDRS)—indicating a direct connection between high plasma levels of inflammatory proteins and motor symptoms—and negatively with cognitive decline, here with the MMSE.
The protein Mannan-binding serine peptidase 2 (MASP2), an initiator of the lectin part of the complement cascade, was significantly downregulated in PD and iRBD. MASP1 and MASP2 proteins are inhibited by plasma protease C1 inhibitor SERPING1 in the lectin pathway, with SERPING1 modulating the complement cascade as it belongs to the SERPIN family of acute phase proteins 38 . In experimental PD mice models, increased SERPING1 levels are associated with dopaminergic cell death 39 . Acting as a serine/cysteine proteinase inhibitor, SERPING1 can increase serine levels, which could also affect αSyn phosphorylation. This can play a crucial role in PD pathology, as almost 90% of αSyn in Lewy bodies is phosphorylated on Serine129 40 , 41 . We identified increased SERPING1 plasma levels in both PD and iRBD in our analysis (compared to HC), thus contributing to conditions with increased αSyn phosphorylation, consecutive aggregation, Lewy body formation, and finally degeneration of dopaminergic neurons. Furthermore, we observed a strong correlation of SERPING1 plasma levels with UPDRS II, III and total score, as a direct measure of dopaminergic cell loss 39 .
Alpha-2-antiplasmin (SERPINF2) was also significantly upregulated in PD and iRBD. SERPINF2 is a major regulator of the clotting pathway, acting as an inhibitor of plasmin, a serine protease formed upon the proteolytic cleavage of its precursor, plasminogen, by tissue-type plasminogen activator (t-PA) or by the urokinase-type plasminogen activator (u-PA). Plasmin has been reported to cleave and degrade extracellular and aggregated αSyn 42 . Recently, we showed that activation of the plasminogen/plasmin system is decreased in PD, indicated by decreased plasma levels of uPA and its corresponding receptor uPAR, while t-PA was associated with faster disease progression 8 . The upregulation of SERPINF2 observed here is another indicator of decreased plasmin activity. Alpha-1-antichymotrypsin (SERPINA3), a third member of the SERPIN family, was also upregulated in the PD subjects. In the CNS, the primary source of SERPINA3 is astrocytes, where its expression is upregulated by various inflammatory receptor complexes 38 .
Overall, independent upregulation of these three members of the SERPIN (SERPING1, SERPINF2, SERPINA3) family is also indicative of increased inflammatory activity, combined with less activation of the plasmin system, and correlation with motor and non-motor symptom severity. In addition, a strong downregulation of progranulin ( GRN ) was detected, indicating a potential loss of neuroprotection and increased susceptibility to neuroinflammation. GRN may act as a neurotrophic factor, promoting neuronal survival and modulating lysosomal function. Loss-of-function mutations in the GRN gene are a cause of frontotemporal dementia and familial DLB. GRN gene variants are also known to increase the risk of developing Alzheimer’s disease (AD) and PD 43 . The main characteristics of neurodegeneration related to GRN are TDP43(-Transactive response DNA binding protein 43) inclusions, but Lewy body pathology is also very common. Loss of progranulin has further been linked to increased production of pro-inflammatory species such as tumour necrosis factor (TNF) and IL-6 in microglia 15 . A study in mice showed that Grn -/- mice had elevated levels of complement proteins, including C3, even before the onset of neurodegeneration 44 . Additionally, previous studies have found GRN downregulated in serum samples of advanced PD compared to AD and healthy individuals 45 .
As a possible compensatory reaction to the described increased inflammatory markers, the levels of Prostaglandin-H 2 d -isomerase (PTGDS)/Prostaglandin-D 2 synthase (PGDS2), better known as β-trace protein, were upregulated. PDGDS is an important brain enzyme producing prostaglandin D2 (PGD2), which has a neuroprotective and anti-inflammatory function. The upregulation reported here could be a reaction to the amount of neuronal cell loss, which is also seen in the significant correlation with the clinical motor and cognitive scales (see below). Furthermore, β-trace protein is a marker for CSF and is used to identify the fluid in clinical routine diagnostics, thus helping detect CSF leakage 46 . Increased plasma levels could be indicative of a disrupted blood–brain barrier (BBB), often discussed in PD pathology 47 and demonstrated in our cohorts.
Our study shows that the Wnt-related proteins DKK3 and PPP3CB are strongly downregulated in de novo PD. DKK3 is an activator of the canonical Wnt/β-catenin branch and PPP3CB is a component of the non-canonical Wnt/Ca 2+ signalling pathway. Wnts are secreted, cysteine-rich glycoproteins that act as ligands to locally stimulate receptor-mediated signal transduction of the Wnt-pathway 48 . Wnt-signalling is crucial for the development and maintenance of dopaminergic neurons 49 , shows protective effects on midbrain dopaminergic neurons 50 , and seems to be involved in the maintenance of the BBB 48 , 51 . Wnt-ligands and agonists trigger a “Wnt-On” stage, characterised by neuronal plasticity and protection, while the opposite “Wnt-Off” stage, potentially leading to neurodegeneration, triggered by the phosphorylation activity of glycogen synthetase kinase-3β (GSK-3beta) 50 , 52 . Wnt-inhibitors are separated into secreted Frizzled-related proteins (sFRP) and Dickkopf proteins (DKK). DKK1, DKK2 and DKK4 act as antagonists, while DKK3 is an agonist and activator 53 . Adult neurogenesis is primarily governed by canonical Wnt/β-catenin signaling 54 and downregulation of Wnt-signalling promotes dysfunction and/or death of dopaminergic neurons. Restoration of dopaminergic neurons was shown in mice where β-catenin was activated in situ 52 and neural stem cells transplanted to the substantia nigra of medically PD-induced mice induced re-expression of Wnt1 and repair dopaminergic neurons 55 . DKK3 and PPP3CB were strongly downregulated in de novo PD, removing an important line of defence against the detrimental loss of dopaminergic neurons. The downregulation of the Wnt-signalling pathways was further correlated with higher motor scores (UDPRS and H&Y stages).
Wnt-signalling in PD is not only promising as a potential biomarker. In oncology, drugs can modify Wnt-pathways, which is of interest to the PD field 56 . Some substances show no BBB-permeability. As a disrupted BBB seems to be apparent in PD, these drugs may be effective. Furthermore, these substances are also relevant for PD treatment: research points towards a peripheral starting point of PD and future therapies should be administered as early as possible 57 . These promising substances include DKK- as well as GSK inhibitors, but to date, no drugs targeting the Wnt-signalling pathways have been effectively tested in clinical trials, including in those with neurodegenerative diseases. Progress and clinical trials are urgently needed here.
The transfer of multi-omics analysis to clinically meaningful results that directly impact future drug trial planning and biomarker validation, depends fundamentally on correlating these results and altered pathway regulations with established clinical scores. The markers we analysed in our targeted mass spectrometry panel did not only show different expression patterns between HC, PD, and in both of our independent iRBD sample sets, but most of the markers also robustly correlated with important clinical scores (UPDRS and MMSE, see Table 1 ). Cognitive decline correlated negatively with the SERPINs and complement factor C3. The burden of motor and non-motor symptoms and overall symptom severity rated by UPDRS and its subscores correlated positively with the SERPINs, Complement C3, and negatively with DKK3, GRN, and SELE. So, increased inflammatory activity and downregulation of Wnt-signalling seem to strongly affect the clinical picture of PD subjects.
The iRBD subjects showed decreased levels of BCHE over time compared to controls. BCHE has been reported as decreased in serum samples of PD with cognitive impairment 58 . Validation of this easily assessable marker in serum is needed to evaluate its predictive potential.
While we did not find significant differences when we compared paired serum and plasma samples; the analysis of paired samples of plasma/serum and CSF only correlated weakly with the marker concentrations in these peripheral and central compartments. This discrepancy has been reported by several groups 20 , 21 . One reason is that mass spectrometry-based proteome analysis is always biased towards quantification and detection of the most abundant proteins in each sample matrix, and the total protein concentrations in human plasma/serum are more than two orders of magnitude higher than that in CSF. Further, the regulatory function of the blood–brain barrier seems to play a different role for different proteins, as some, like c-reactive protein, show a strong correlation between CSF and plasma, but most of the proteins do not. CSF and blood proteome show complex dynamics influenced by multiple and still mostly unknown factors. The protein shift in samples with a known BBB dysfunction (determined by the CSF/serum albumin index or the CSF/plasma ratio) can not be determined for individual proteins nor the dysfunction be localised by mass spectrometry 20 .
Our model could not correctly predict phenoconversion in all cases. The reasons for this can be varied: The proteome pattern changes over time and the period between sampling and phenconversion may play a role. The three PD phenoconverters that were not predicted as PD neither differ clinically or demographically from the phenoconverters, nor from the non-phenoconverters. iRBD diagnosis in our study was confirmed by vPSG, supported by a high percentage of additional measurements including hyposmia and CSF SAA positivity. Therefore, even those iRBD cases that do not show the PD-proteome pattern still have a high-risk constellation of converting to PD/DLB on three different levels (PSG, olfaction, and SAA). Continuing further longitudinal follow-up of these subjects will elucidate our understanding of when and potentially why conversion occurs/does not occur. It is known that around 80% of iRBD subjects develop NSD, i.e. PD/DLB, with a rate of 6% per year, as shown in a multicenter cohort including ours 59 . To a lesser extent, iRBD subjects develop the intracytoplasmic glial α-synuclein aggregation disorder Multiple Systems Atrophy (MSA) 59 , 60 . Although RBD is common in MSA (summary prevalence of 73% 61 ), none of our iRBD subjects have, as yet converted to MSA. Recruiting and following large longitudinal at-risk cohorts is, therefore, very important and future studies will not only identify biomarkers for phenoconversion from stage 1 or 2 to eventually stage 3 NSD or MSA, but also identify the many possible factors of resilience (including genetics, etc.) of NON-conversion which will be as, if not more important than identifying indicators for phenoconversion. Both direction progression biomarkers from stage 1 and 2 cohorts will have tremendous implications for future neuroprevention trials as phenoconversion itself is (due to the low annual rate) unlikely to be an outcome measure.
A significant strength of our biomarker discovery to translation pipeline is that it allows for the developed test to be easily validated and translated to any clinical laboratory equipped with a tandem LC-MS instrument. One advantage of using triple quadrupole platforms is that additional and better biomarkers can easily be augmented into the test described in this manuscript. Thus, any test could be refined and optimised over time with very little modification to the assay as additional biomarkers are discovered. Clinical testing for neurological disorders is limited to the use of a selected few well-characterised individual markers and translating biomarkers to eventual clinical application is notoriously challenging. The power of using multiplexed biomarker technologies with machine learning enables biomarkers to be evaluated in context with other markers of pathological events, thereby creating a ‘disease profile’ as opposed to individual markers. This approach opens the biomarker discovery field for many disorders and increases the specificity and sensitivity of testing, as demonstrated in this study. The combination of multiplexed analysis of biomarker panels analysed on triple quadrupole platforms can advance biomarker translation to clinical application; this mass spectral technology is already embedded in many clinical diagnostics labs for routine small molecule analyses.
Our peripheral blood protein pattern for PD helps not only to classify but also to predict the earliest stage of the disease. We find differently expressed proteins in pre-motor iRBD and early motor stages of the disease compared to HC. Multiple markers also correlated with the progression of motor and non-motors symptoms. Thus, our blood panel can also identify subjects at risk (stage 2) to develop PD up to 7 years before advancing to motor stage 3. Next steps will be the independent validation in other (and even earlier) non-motor cohorts, e.g. in subjects with hyposmia also at-risk for PD 62 and in our population-based Healthy Brain Ageing cohort in Kassel 63 . It would further be interesting to evaluate the predictive potential of these identified markers with continuing clinical follow-up and together with other established PD progression markers like serum neurofilament light chain 5 and dopamine transporter imaging in a longitudinal analysis.
Our work was predominantly focused on the similarities between PD and iRBD. The authors are unaware of any study that has analysed longitudinally collected samples and prodromal cohorts, including iRBD and phenoconverters. Future work would include (i) validation of our findings in independent cohorts consisting of iRBD and other at-risk subjects for the synuclein aggregation disorders in neurons (PD, DLB) and oligodendrocytes (MSA), (ii) refinement of the panels of biomarkers developed in this study including sensitivity and technical performance, (iii) and using the pipeline described in this manuscript, the identification and validation of additional biomarkers that could distinguish between the different clinical syndromes with the ultimate goal of identifying progression biomarkers as outcome measures for prevention trials.
In summary, instead of single biomarkers, in a univariate approach, we have created a pipeline using a targeted proteomic test of a multiplexed panel of proteins, together with machine learning. This powerful combination of multiple well-selected biomarkers with state-of-the-art machine-learning bioinformatics, allowed us to use a panel of eight biomarkers that could distinguish early PD from HC. This biomarker panel provided a distinct signature of protective and detrimental mechanisms, finally triggering oxidative stress and neuroinflammation, leading to α-synuclein aggregation and LB formation. Moreover, this signature was already present in the prodromal non-motor (stage 2 NSD), up to 7 years before the development of motor/cognitive symptoms (stage 3), supporting the high specificity of iRBD and its high conversion rate to PD/DLB 18 . Most importantly, this blood panel can, in the future, upon further validation help identify subjects at risk of developing PD/DLB and stratify them for upcoming prevention trials.
Our research complies with all relevant ethical regulations. Institutional review board statements were obtained from the University Medical Centre in Goettingen, Germany, Approval No. 9/7/04 and 36/7/02. The study was conducted according to the Declaration of Helsinki, and all participants gave written informed consent. All plasma, serum and CSF samples from subjects were selected from known cohorts using identical sample processing protocols designed by the Movement Disorder Center Paracelsus-Elena-Clinic.
Patients with de novo PD were diagnosed according to the UK Brain Bank Criteria, without PD-specific medication. Diagnosis in all subjects was supported by (1) a positive (i.e. >30% improvement of UPDRS III after 250 mg of levodopa) acute levodopa challenge testing 64 in all PD subjects, (2) hyposmia by smell identification test (Sniffin Sticks 65 ) in all PD subjects and (3) 1.5-tesla Magnetic Resonance Imaging (MRI) without significant abnormalities or evidence for other diseases in all but three subjects who were excluded (due to significant vascular lesions or evidence for hydrocephalus) from the analysis. Participants not fulfilling the above criteria and meeting criteria for other neurological disorders were named as other neurological disorders (OND). OND consists of subjects with vascular parkinsonism ( n = 10), essential tremor ( n = 7), progressive supranuclear palsy; PSP ( n = 7), multiple system atrophy; MSA ( n = 3), corticobasal syndrome; CBS ( n = 2), DLB ( n = 2), drug-induced tremor ( n = 2), dystonic tremor ( n = 2), restless legs syndrome ( n = 1), hemifacial spasm ( n = 1), motoneuron disease ( n = 1), amyotrophic shoulder neuralgia ( n = 1), and Alzheimer’s disease ( n = 1). The initial exploratory cohort consisted of ten PD subjects (8 men, mean age 67.1 ± 10.6) and ten healthy controls (5 men, mean age 65,7, SD ± 8,6.). For details, see Supplementary Table 3 ). The validation cohort included 99 PD subjects (49 men, mean age 66,1, SD ± 10,8), 36 healthy controls (20 men, mean age 63.7, SD ± 6,5.) and the described (see above) 41 OND subjects (29 men, mean age 70, SD ± 8.9. For details, see Supplementary Table 1 . The prodromal validation cohort consisted of 54 patients with iRBD (27 men, mean age 67.5, SD ± 8.1, for details, see Supplementary Table 4 ). RBD was diagnosed with two nights of state-of-the-art vPSG. Samples from HC were selected from the DeNoPa cohort 10 and matched for age and sex with the PD patients, had to be between 40 and 85 years old, without any active known/treated CNS condition, and with a negative family history of idiopathic PD. Antipsychotic drugs were an exclusion criterion. The provided data for sex are based on self-report.
The paired sample analysis of CSF, plasma and serum was applied in samples from subjects with OND 7 men, mean age 74 years, SD ± 7; diagnosis: four Alzheimer’s disease, three vascular Parkinsonism, one essential tremor, one multiple system atrophy one progressive supranuclear palsy).
Clinical assessments included the UPDRS subscores (parts I–III), the sum (UPDRS total score), and cognitive screening using the MMSE 10 .
Plasma and serum samples for both cohorts were collected in the morning under fasting conditions using Monovette tubes (Sarstedt, Nümbrecht, Germany) for EDTA plasma and serum collection by venipuncture. Tubes were centrifuged at 2500× g at room temperature (20 °C) for 10 min and aliquoted and frozen within 30 min of collection at −80 °C until analysis 10 , 66 . Single- use aliquots were used for all analyses presented here. For further details, we refer to the following publication 67 .
CSF was collected in polypropylene tubes (Sarstedt, Nümbrecht, Germany) directly after the plasma collection by lumbar puncture in the sitting position. Tubes were centrifuged at 2500× g at room temperature (20 °C) for 10 min and aliquoted and frozen within 30 min after collection at −80 °C until analysis. Before centrifugation, white and red blood cell counts in CSF were determined manually 10 , 66 . CSF β-amyloid 1–42, total tau protein (t-tau), phosphorylated tau protein (p-tau181) and neurofilament light chains (NFL) concentrations were measured by board-certified laboratory technicians, who were blinded to clinical data, using commercially available INNOTEST ELISA kits for the tau and Aβ markers (Fujirebio Europe, Ghent, Belgium) and the UmanDiagnostics NF-light® assay (UmanDiagnostics, Umeå, Sweden) for NFL. Total protein and albumin levels were measured by nephelometry (Dade Behring/Siemens Healthcare Diagnostics) 66 .
For the α-synuclein seeding aggregation assay (αSyn-SAA) the CSF samples were blindly analyzed in triplicate (40 μL/well) in a reaction mixture (0.3 mg/mL recombinant α-Syn (Amprion [California, USA]; catalogue number S2020), 100 mM piperazine- N , N ′-bis(2-ethanesulfonic acid) (PIPES) pH 6.50, 500 mM sodium chloride, 10 μM thioflavin T, and one bovine serum albumin (BSA)–blocked 2.4-mm silicon nitride G3 bead (Tsubaki-Nakashima [Georgia, USA]). Beads were blocked in 1% BSA 100 mM PIPES pH 6.50 and washed with 100 mM PIPES pH 6.50. The assay was performed in 96-well plates (Costar [New York, USA], catalogue number 3916) using a FLUOstar Omega fluorometer (BMG [Ortenberg, Germany]). Plates were orbitally shaken (800 rpm for 1 min every 29 min at 37 °C). Results from the triplicates were considered input for a three-output probabilistic algorithm with sample labelling as “positive,” “negative,” or “inconclusive”, based on the parameters: Maximum fluorescence (Fmax), time to reach 50% Fmax (T50), slope, and the coefficient of determination for the fitting were calculated for each replicate using a sigmoidal equation available in Mars data analysis software (BMG). The time to reach the 5000 relative fluorescence units (RFU) threshold (TTT) was calculated with a user-defined equation in Mars 19 .
In the mass spectrometry-based proteomic discovery analysis of plasma, we depleted the control and de novo PD samples from the twelve most abundant plasma proteins using Pierce Top12 columns (Thermo Fisher Scientific, Waltham, MA, USA) according to the manufacturer’s instructions. The depleted samples were freeze-dried before the addition of 20 µL of lysis buffer (100 mM Tris pH 7.8, 6 M urea, 2 M thiourea, and 2% ASB-14). The samples were shaken on an orbital shaker for 60 min at 1500 rpm. To break disulphide bonds, 45 µg DTE was added, and the samples were incubated for 60 min. To prevent disulphide bonds from reforming, 108 µg IAA was added, and the samples were incubated for 45 min covered in light. About 165 µL MilliQ water was added to dilute the concentration of urea and 1 µg trypsin gold (Promega, Mannheim, Germany) was added before 16 h of incubation at +37 °C to digest the proteins into peptides. To purify the peptides, solid phase extraction was performed using 100 mg C18 cartridges (Biotage, Uppsala, Sweden). The cartridges were washed with two 1 mL aliquots of 60% ACN, and 0.1% TFA before equilibration by two 1 mL aliquots of 0.1% TFA. The concentration of TFA in the samples was adjusted to 0.1%. The samples were loaded, and the flow-through was captured and re-applied. Salts were washed away from the bound peptides by two 1 mL aliquots of 0.1% TFA. The peptides were eluted by two 250 µL aliquots of 60% ACN, and 0.1% TFA. Solvents were evaporated using a vacuum concentrator. The samples were re-suspended in 50 µL 3% ACN, 0.1% FA prior to analysis. About 4 µL was injected into a 2D-NanoAquity liquid chromatography system (Waters, Manchester, UK). All samples were fractionated online into ten fractions over 12 h. The mobile phase in the first chromatographic system consisted of A1: 10 mM ammonium hydroxide titrated to pH 9 and B1: acetonitrile. The second chromatographic system’s mobile phase was A2: 5% dimethylsulfoxide (DMSO) + 0.1% formic acid, B2: acetonitrile with 5% DMSO + 0.1% formic acid. 2D-liquid chromatography fractionation was performed by loading the sample onto a 300 µm × 50 mm, 5 µm Peptide BEH C18 column (Waters). The peptides were eluted from the first column at a flow rate of 2 µL/min. The initial condition of the gradient elution was 3% B, held over 0.5 minutes before linearly increasing the proportion of organic solvent B, fraction per fraction over 0.5 min. The conditions thereafter remained static for 4 min before returning to the initial conditions over 0.5 min and equilibration prior to the next elution for 10 min. The eluted peptides from the first-dimensional column were loaded into a 180 µm × 20 mm, 5 µm Symmetry C18 trap column (Waters) before entering the analytical column, a 75 µm × 150 mm, 1.7 µm Peptide BEH C18 (Waters). The column temperature was +45 °C. The gradient elution applied to the analytical column started at 3% B and was linearly increased to 40% B over 40 min after which it was increased to 85% B over 2 min and washed for 2 min before returning to initial conditions over 2 min followed by equilibration for 15 min before the subsequent injection. The eluted peptides were detected using a Synapt-G2-S i (Waters) equipped with a nano-electrospray ion source. Data were acquired in positive MS E mode from 0 to 60 min within the m/z range 50−2000. The capillary voltage was set to 3 kV and the source temperature to +100 °C. The desolvation gas consisted of nitrogen with a flow rate of 50 L/h, and the desolvation temperature was set to +200 °C. The purge and desolvation gas consisted of nitrogen, operated at a flow rate of 600 mL/h and 600 L/h, respectively. The gas in the IMS cell was helium, with a flow rate of 90 mL/h. The low energy acquisition was performed by applying a constant collision energy of 4 V with a 1-s scan time. High energy acquisition was performed by applying a collision energy ramp, from 15 to 40 V, and the scan time was 1 s. The lock mass consisted of 500 fmol/µL [glu1]-fibrinopeptide B, continuously infused at a flow rate of 0.3 µL/min and acquired every 30 s. The doubly charged precursor ion, m/z 785.8426, was utilised for mass correction. After acquisition, data were imported to Progenesis QI for proteomics (Waters), and the individual fractions were processed before all results were merged into one experiment. The Ion Accounting workflow was utilised, with UniProt Canonical Human Proteome as a database (build 2016). The digestion enzyme was set as trypsin. Carbamidomethyl on cysteines was set as a fixed modification; deamidation of glutamine and asparagine, and oxidation of tryptophan and pyrrolidone carboxylic acid on the N-terminus were set as variable modifications. The identification tolerance was restricted to at least two fragments per peptide, three fragments per protein, and one peptide per protein. A FDR of 4% or less was accepted. The resulting identifications and intensities were exported and variables with a confidence score less than 15 and only one unique peptide were filtered out.
The peptides included in the targeted assay were selected from several proteomic screening studies in which we analysed plasma, serum, urine, and CSF in ageing, PD and AD. The analytical method is described by ref. 17 . Furthermore, due to the suggested involvement of inflammation in neurodegenerative diseases, several known pro- and anti-inflammatory proteins identified from the literature were included in the multiplexed assay. The final panel consisted of 121 proteins (Supplementary Table 2 ), out of which a number were measured with two peptides, leading to a total of 167 unique peptides. When possible, the peptides were chosen to have an amino acid sequence length between 7 and 20. The amino acid sequences were confirmed to be unique to the proteins by using the Basic Local Alignment Search Tool (BLAST) provided by UniProt 68 . Synthetic peptide standards were purchased from GenScript (Amsterdam, Netherlands). To establish the most optimal transitions, repeated injections of 1 pmol peptide standard onto a Waters Acquity ultra-performance liquid chromatography (UPLC) system coupled to a Waters Xevo-TQ-S triple quadrupole MS were performed. The most high-abundant precursor-to-product ion transitions and their optimal collision energies were determined manually or using Skyline 69 . Detection was performed in positive ESI mode. The capillary voltage was set to 2.8 kV, the source temperature to 150 °C, the desolvation temperature to 600 °C, and the cone gas and desolvation gas flows to 150 and 1000 L/h, respectively. The collision gas consisted of nitrogen and was set to 0.15 mL/min. The nebuliser operated at 7 bar. Two transitions were chosen, one quantifier for relative concentration determination and one qualifier for identification, totally rendering 334 analyte transitions. Cone and collision energies varied depending on the optimal settings for each peptide. Each peptide was measured with a minimum of 12 points per peak and a dwell time of 10 ms or more to ensure adequate data acquisition. The optimised transitions were distributed over two multiple reaction monitoring (MRM) methods, always keeping the quantifier and qualifier for each peptide in the same MRM segment. Plasma, serum, and CSF samples were depleted from albumin and IgG using Pierce Top2 cartridges (Thermo Fisher Scientific, Waltham, MA, USA) following the manufacturer’s instructions. About 150 µg whole protein yeast enolase (ENO1) was added to the cartridges as an internal standard to account for digestion efficiency. Digestion was performed as described above. Solid phase extraction was carried out on BondElute 100 mg C18 96-well plates (Agilent, Santa Clara, USA) using the same methodology as in the preparation of untargeted proteomic analyses. Quality control samples were prepared from acetone-precipitated plasma, digested and solid phase extracted. Calibration curves ranging from 0 to 1 pmol/μL were constructed in blank and matrix by spiking increasing amounts of peptides into blank and QC samples. Before analysis, the samples were reconstituted in 30 µL 3% ACN, 0.1% FA containing 0.1 μM heavy isotope labelled peptides from the following proteins (annotated by gene name): ALDOA, C3, GSTO1, RSU1 and TSP1. About 5 µL were injected. The peptides were separated and detected on an Acquity UPLC system coupled to a Xevo-TQ-S triple quadrupole mass spectrometer (Waters, Manchester, UK). Chromatographic separation of the peptides was performed using a 1 × 100 mm, 1.7 μm ACQUITY UPLC Peptide CSH C18 column (Waters).
The mobile phase consisted of A: 0.1% formic acid and B: 0.1% formic acid in acetonitrile pumped at a flow rate of 0.2 mL/min. The column temperature was set to +55 °C. The initial mobile phase composition was 3% B, which was kept static for 0.8 min before initialising the linear gradient, running for 7.6 min to 25% B, eluting most of the peptides. B was thereafter linearly increased to 80% over 0.5 min and held for 1.9 min, eluting the most apolar peptides and washing the column before returning to the initial conditions over 0.1 minutes followed by equilibration for 6 min prior to the subsequent injection. Two subsequent injections of each sample were performed, each paired with one of the two MRM acquisition methods.
After acquisition, peak-picking and integration were performed using TargetLynx (version 4.1, Waters) or an in-house application ('mrmIntegrate') written in Python (version 3.8). mrmIntegrate is publicly available to download via the GitHub repository https://github.com/jchallqvist/mrmIntegrate . The application takes text files as input (.raw files are transformed into text files through the application 'MSConvert' from ProteoWizard 70 and applies a LOWESS filter over five points of the chromatogram. The integration method to produce areas under the curve is trapezoidal integration. The application enables retention time alignment and simultaneous integration of the same transition for all samples. Peptide peaks were identified by the blank and matrix calibration curves. The integrated peak areas were exported to Microsoft Excel, where first, the ratio between quantifier and qualifier peak areas were evaluated to ensure that the correct peaks had been integrated. The digestion efficiency was evaluated by monitoring the presence of baker’s yeast ENO1 in the samples, all samples without a signal were excluded from further analysis. After the initial quality assessment, the quantifier area was divided by the area of one of the internal standards, ALDOA or GSTO1 to yield a ratio used for the determination of relative concentrations. Any compound that also showed an intensity signal in the blank samples had the blank signal subtracted from the analyte peak intensity. Pooled plasma quality control samples were additionally evaluated to assess the robustness of the run.
The rapid and refined targeted proteomics LC-MS/MS method contained only peptides from the 31 proteins observed in the original targeted proteomics method (121 proteins). We utilised a Waters Acquity (UPLC) system coupled to a Waters Xevo-TQ-XS triple quadrupole operating in positive ESI mode. The column was an ACQUITY Premier Peptide BEH C18, 300 Å, 1.7 µm, maintained at 40 °C. The mobile phase was A: 0.1% formic acid in water, and B: 0.1% formic acid in acetonitrile. The gradient elution profile was initiated with 5% B and held for 0.25 min before linearly increasing to 40% B over 9.75 min to elute and separate the peptides. The column was washed for 1.6 min with 85% B before returning to the initial conditions and equilibrating for 0.4 min. The flow rate was 0.6 mL/min. The settings of the mass spectrometer and the peak-picking method were the same as described in the prior section. Baker’s yeast ENO1 was utilised to monitor digestion efficiency and as an internal standard.
Most of the statistical analyses were performed in Python (version 3.8.5). The untargeted and targeted datasets were inspected for outliers and instrumental drift using principal component analysis (PCA) and orthogonal projection to latent variables (OPLS) in SIMCA, version 17 (Umetrics Sartorius Stedim, Umeå, Sweden). Outliers exceeding ten median deviations from each variable’s median were excluded. Instrumental drift was corrected by applying a non-parametric LOWESS filter from statsmodels (version 0.14.0) using 0.5 fractions of the data to estimate the LOWESS curve 71 . The data were evaluated for normal distribution using D’Agostino and Pearson’s method from SciPy (version 1.9.3) 72 . The non-normally distributed variables in the untargeted data were transformed to normality by the Box-Cox procedure using the SciPy function 'boxcox'. Significance testing between the independent groups of HC and PD/OND/iRBD individuals was performed by Student’s two-tailed t -test for the untargeted proteomic data and by Mann–Whitney’s non-parametric U -test (SciPy) for the targeted data. Due to the limited sample numbers, no multiple testing correction was performed in the untargeted data. In the targeted data, the Benjamini–Hochberg multiple testing correction procedure (statsmodels) was applied with an accepted false discovery rate of 5%. Fold-changes were calculated by dividing the means of the affected groups by the control group. Correlation analyses in the targeted data were performed by Spearman’s correlation (SciPy) and the correlation p values were adjusted variable-wise by the Benjamini–Hochberg procedure (FDR = 5%).
We implemented a support vector classifier model to discriminate between PD and HC and to predict new samples. The data were first z-scored protein-wise and any 'not a number'-values were replaced by the median. We used the 'LinearSVC' method from SciKit Learn and applied cross-validated recursive feature elimination to determine the number of variables to use in the model. The most discriminating variables for distinguishing between controls and PD were thereafter chosen by recursive feature elimination 73 . Feature selection and model training were performed on 70% of the data, partitioned using the SciKit Learn function “train_test_split”, and cross-validation was performed using a stratified k-fold with five splits. The remaining 30% of the data were predicted in the model. PR and ROC curves were constructed from the test data and consisted of each predictor and from the combined predictors, the packages precision_recall_curve and roc_curve from SciKit Learn were implemented. Linear mixed models were performed using the R-to-Python bridge software pymer4 (version 0.8.0), where individual was set as a random effect and the correlations between the MS measured proteins and clinical variables were evaluated for significance post Benjamini–Hochberg’s procedure for multiple testing correction. Plots of the data were constructed using the Seaborn and Matplotlib packages (versions 0.12.2 and 3.6.0, respectively) 74 .
All multivariate analyses were performed in SIMCA, version 17. OPLS and OPLS-discriminant analysis (OPLS-DA) models were evaluated for significance by ANOVA p values and by permutation tests applying 1000 permutations, where p < 0.05 and p < 0.001 were deemed significant, respectively.
Data were analysed for pathway enrichment using IPA (QIAGEN Inc. Data were analysed for pathway enrichment using IPA (QIAGEN Inc., https://digitalinsights.qiagen.com/products-overview/discovery-insights-portfolio/analysis-and-visualization/qiagen-ipa/ .). Input variables were set to proteins demonstrating a significant difference between PD individuals and HC, with fold-change as expression observation. The accepted output pathways were restricted to p < 0.05 and at least two proteins were included in the pathways. Gene Ontology (GO) annotations were extracted using DAVID Bioinformatics Resources (2021 build) 75 , 76 . Networks were built in Cytoscape 77 (version 3.8.0) by applying the “Organic layout” from yFiles 77 .
Patient samples can be provided to other researchers for certain projects after contact with the corresponding authors and upon availability approval of the team in Kassel, Germany.
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
The chromatograms from the targeted mass spectrometric data generated in this study have been deposited in the ProteomeXchange database under accession code PXD041419 and in the Panorama repository ( https://panoramaweb.org/DNP_Pub.url , https://doi.org/10.6069/p9cy-h335 ). The integrated targeted mass spectrometric data generated in this study are provided in the Supplementary Information. Source data for all data presented in graphs within the figures are provided in a source data file. Source data are provided with this paper.
Peak-picking and integrations were performed in TargetLynx (part of the MassLynx suite, version 4.1), or using an in-house application written in Python which can be found on GitHub ( https://github.com/jchallqvist/mrmIntegrate ). The data visualisation and statistical analyses were performed in Python (version 3.8.5) using the packages SciPy (version 1.9.3), statsmodels (version 0.14.0), SciKit Learn (version 1.1.2), Seaborn (version 13.0) and Matplotlib (version 3.6.0). The code used can be found on GitHub ( https://github.com/jchallqvist/DNP_Pub/blob/main/DNP_Code , https://doi.org/10.5281/zenodo.11130369 ).
Simuni, T. et al. Baseline prevalence and longitudinal evolution of non-motor symptoms in early Parkinson’s disease: the PPMI cohort. J. Neurol. Neurosurg. Psychiatry 89 , 78–88 (2018).
Article PubMed Google Scholar
Michell, A. W., Lewis, S. J., Foltynie, T. & Barker, R. A. Biomarkers and Parkinson’s disease. Brain 127 , 1693–1705 (2004).
Article CAS PubMed Google Scholar
Kieburtz, K., Katz, R., McGarry, A. & Olanow, C. W. A new approach to the development of disease-modifying therapies for PD; fighting another pandemic. Mov. Disord. 36 , 59–63 (2021).
Shahnawaz, M. et al. Development of a biochemical diagnosis of Parkinson disease by detection of α-synuclein misfolded aggregates in cerebrospinal fluid. JAMA Neurol. 74 , 163–172, (2017).
Mollenhauer, B. et al. Validation of serum neurofilament light chain as a biomarker of Parkinson’s disease progression. Mov. Disord . https://doi.org/10.1002/mds.28206 (2020).
Lindestam Arlehamn, C. S. et al. α-Synuclein-specific T cell reactivity is associated with preclinical and early Parkinson’s disease. Nat. Commun. 11 , 1875 (2020).
Article ADS CAS PubMed PubMed Central Google Scholar
Mollenhauer, B. et al. Baseline predictors for progression 4 years after Parkinson’s disease diagnosis in the De Novo Parkinson Cohort (DeNoPa). Mov. Disord. 34 , 67–77 (2019).
Bartl, M. et al. Blood markers of inflammation, neurodegeneration, and cardiovascular risk in early Parkinson’s disease. Mov. Disord . https://doi.org/10.1002/mds.29257 (2022).
Simuni, T. et al. A biological definition of neuronal α-synuclein disease: towards an integrated staging system for research. Lancet Neurol. 23 , 178–190 (2024).
Mollenhauer, B. et al. Nonmotor and diagnostic findings in subjects with de novo Parkinson disease of the DeNoPa cohort. Neurology 81 , 1226–1234 (2013).
Hällqvist, J. et al. A multiplexed urinary biomarker panel has potential for Alzheimer’s disease diagnosis using targeted proteomics and machine learning. Int. J. Mol. Sci . https://doi.org/10.3390/ijms241813758 (2023).
Hu, W., Ralay Ranaivo, H., Craft, J. M., Van Eldik, L. J. & Watterson, D. M. Validation of the neuroinflammation cycle as a drug discovery target using integrative chemical biology and lead compound development with an Alzheimer’s disease-related mouse model. Curr. Alzheimer Res. 2 , 197–205 (2005).
Notter, T. et al. Translational evaluation of translocator protein as a marker of neuroinflammation in schizophrenia. Mol. Psychiatry 23 , 323–334 (2018).
Jonsson, M., Gerdle, B., Ghafouri, B. & Backryd, E. The inflammatory profile of cerebrospinal fluid, plasma, and saliva from patients with severe neuropathic pain and healthy controls-a pilot study. BMC Neurosci. 22 , 6 (2021).
Article PubMed PubMed Central Google Scholar
Chen, X. et al. Progranulin does not bind tumor necrosis factor (TNF) receptors and is not a direct regulator of TNF-dependent signaling or bioactivity in immune or neuronal cells. J. Neurosci. 33 , 9202–9213 (2013).
Article CAS PubMed PubMed Central Google Scholar
Captur, G. et al. Plasma proteomic signature predicts who will get persistent symptoms following SARS-CoV-2 infection. EBioMedicine 85 , 104293 (2022).
Doykov, I. et al. The long tail of Covid-19’ - The detection of a prolonged inflammatory response after a SARS-CoV-2 infection in asymptomatic and mildly affected patients. F1000Res 9 , 1349 (2020).
Hu, M. T. REM sleep behavior disorder (RBD). Neurobiol. Dis. 143 , 104996 (2020).
Concha-Marambio, L. et al. Accurate detection of α-synuclein seeds in cerebrospinal fluid from isolated rapid eye movement sleep behavior disorder and patients with Parkinson’s disease in the de novo Parkinson (DeNoPa) cohort. Mov. Disord. 38 , 567–578 (2023).
Dayon, L. et al. Proteomes of paired human cerebrospinal fluid and plasma: relation to blood-brain barrier permeability in older adults. J. Proteome Res. 18 , 1162–1174 (2019).
Whelan, C. D. et al. Multiplex proteomics identifies novel CSF and plasma biomarkers of early Alzheimer’s disease. Acta Neuropathol. Commun. 7 , 169 (2019).
Bloem, B. R., Okun, M. S. & Klein, C. Parkinson’s disease. Lancet 397 , 2284–2303 (2021).
Dorsey, E. R., Sherer, T., Okun, M. S. & Bloem, B. R. The emerging evidence of the Parkinson pandemic. J. Parkinsons Dis. 8 , S3–s8 (2018).
Grossauer, A. et al. α-Synuclein seed amplification assays in the diagnosis of synucleinopathies using cerebrospinal fluid-A systematic review and meta-analysis. Mov. Disord. Clin. Pr. 10 , 737–747 (2023).
Article Google Scholar
Okuzumi, A. et al. Propagative α-synuclein seeds as serum biomarkers for synucleinopathies. Nat. Med. 29 , 1448–1455 (2023).
Raffield, L. M. et al. Comparison of proteomic assessment methods in multiple cohort studies. Proteomics 20 , e1900278 (2020).
Hernández, B., Parnell, A. & Pennington, S. R. Why have so few proteomic biomarkers “survived” validation? (Sample size and independent validation considerations). Proteomics 14 , 1587–1592 (2014).
Füzéry, A. K., Levin, J., Chan, M. M. & Chan, D. W. Translation of proteomic biomarkers into FDA approved cancer diagnostics: issues and challenges. Clin. Proteom. 10 , 13 (2013).
Bader, J. M., Albrecht, V. & Mann, M. MS-based proteomics of body fluids: the end of the beginning. Mol. Cell Proteom. 22 , 100577 (2023).
Article CAS Google Scholar
Pan, C. et al. Targeted discovery and validation of plasma biomarkers of Parkinson’s disease. J. Proteome Res. 13 , 4535–4545 (2014).
Qin, X. Y., Zhang, S. P., Cao, C., Loh, Y. P. & Cheng, Y. Aberrations in peripheral inflammatory cytokine levels in Parkinson disease: a systematic review and meta-analysis. JAMA Neurol. 73 , 1316–1324, (2016).
Choi, M. L. & Gandhi, S. Crucial role of protein oligomerization in the pathogenesis of Alzheimer’s and Parkinson’s diseases. FEBS J. 285 , 3631–3644 (2018).
Walter, P. & Ron, D. The unfolded protein response: from stress pathway to homeostatic regulation. Science 334 , 1081–1086 (2011).
Article ADS CAS PubMed Google Scholar
Bertolotti, A., Zhang, Y. H., Hendershot, L. M., Harding, H. P. & Ron, D. Dynamic interaction of BiP and ER stress transducers in the unfolded-protein response. Nat. Cell Biol. 2 , 326–332 (2000).
Colla, E. Linking the endoplasmic reticulum to Parkinson’s disease and alpha-synucleinopathy. Front. Neurosci. 13 , 560 (2019).
Mercado, G., Castillo, V., Soto, P. & Sidhu, A. ER stress and Parkinson’s disease: pathological inputs that converge into the secretory pathway. Brain Res. 1648 , 626–632 (2016).
Loeffler, D. A., Camp, D. M. & Conant, S. B. Complement activation in the Parkinson’s disease substantia nigra: an immunocytochemical study. J. Neuroinflamm 3 , 29 (2006).
Zattoni, M. et al. Serpin signatures in Prion and Alzheimer’s diseases. Mol. Neurobiol. 59 , 3778–3799 (2022).
Seo, M. H. & Yeo, S. Association of increase in Serping1 level with dopaminergic cell reduction in an MPTP-induced Parkinson’s disease mouse model. Brain Res. Bull. 162 , 67–72 (2020).
Anderson, J. P. et al. Phosphorylation of Ser-129 is the dominant pathological modification of alpha-synuclein in familial and sporadic Lewy body disease. J. Biol. Chem. 281 , 29739–29752 (2006).
Fujiwara, H. et al. alpha-Synuclein is phosphorylated in synucleinopathy lesions. Nat. Cell Biol. 4 , 160–164 (2002).
Kim, K. S. et al. Proteolytic cleavage of extracellular alpha-synuclein by plasmin implications for Parkinson disease. J. Biol. Chem. 287 , 24862–24872 (2012).
Reho, P. et al. GRN mutations are associated with Lewy body dementia. Mov. Disord. 37 , 1943–1948 (2022).
Kao, A. W., Mckay, A., Singh, P. P., Brunet, A. & Huang, E. J. Progranulin, lysosomal regulation and neurodegenerative disease. Nat. Rev. Neurosci. 18 , 325–333 (2017).
Mateo, I. et al. Reduced serum progranulin level might be associated with Parkinson’s disease risk. Eur. J. Neurol. 20 , 1571–1573 (2013).
Bachmann-Harildstad, G. Diagnostic values of beta-2 transferrin and beta-trace protein as markers for cerebrospinal fluid fistula. Rhinology 46 , 82–85 (2008).
PubMed Google Scholar
Pediaditakis, I. et al. Modeling alpha-synuclein pathology in a human brain-chip to assess blood-brain barrier disruption. Nat. Commun. 12 , 5907 (2021).
Serafino, A., Giovannini, D., Rossi, S. & Cozzolino, M. Targeting the Wnt/β-catenin pathway in neurodegenerative diseases: recent approaches and current challenges. Expert Opin. Drug Discov. 15 , 803–822 (2020).
Arenas, E. Wnt signaling in midbrain dopaminergic neuron development and regenerative medicine for Parkinson’s disease. J. Mol. Cell Biol. 6 , 42–53 (2014).
L’Episcopo, F. et al. A Wnt1 regulated Frizzled-1/β-Catenin signaling pathway as a candidate regulatory circuit controlling mesencephalic dopaminergic neuron-astrocyte crosstalk: therapeutical relevance for neuron survival and neuroprotection. Mol. Neurodegener. 6 , 49 (2011).
Sweeney, M. D., Sagare, A. P. & Zlokovic, B. V. Blood-brain barrier breakdown in Alzheimer disease and other neurodegenerative disorders. Nat. Rev. Neurol. 14 , 133–150 (2018).
L’Episcopo, F. et al. Wnt/beta-catenin signaling is required to rescue midbrain dopaminergic progenitors and promote neurorepair in ageing mouse model of Parkinson’s disease. Stem Cells 32 , 2147–2163 (2014).
Marchetti, B. Wnt/beta-catenin signaling pathway governs a full program for dopaminergic neuron survival, neurorescue and regeneration in the MPTP mouse model of Parkinson’s disease. Int. J. Mol. Sci. 19 , 3743 (2018).
Marchetti, B. et al. Parkinson’s disease, aging and adult neurogenesis: Wnt/beta-catenin signalling as the key to unlock the mystery of endogenous brain repair. Aging Cell 19 , e1310110 (2020).
L’Episcopo, F. et al. Neural stem cell grafts promote astroglia-driven neurorestoration in the aged Parkinsonian brain via Wnt/beta-catenin signaling. Stem Cells 36 , 1179–1197 (2018).
Serafino, A. et al. Developing drugs that target the Wnt pathway: recent approaches in cancer and neurodegenerative diseases. Expert Opin. Drug Discov. 12 , 169–186 (2017).
Harms, A. S., Ferreira, S. A. & Romero-Ramos, M. Periphery and brain, innate and adaptive immunity in Parkinson’s disease. Acta Neuropathol. 141 , 527–545 (2021).
Dong, M. X. et al. Serum butyrylcholinesterase activity: a biomarker for Parkinson’s disease and related dementia. Biomed. Res. Int. 2017 , 1524107 (2017).
Postuma, R. B. et al. Risk and predictors of dementia and parkinsonism in idiopathic REM sleep behaviour disorder: a multicentre study. Brain 142 , 744–759 (2019).
Zhang, H. et al. Risk factors for phenoconversion in rapid eye movement sleep behavior disorder. Ann. Neurol. 91 , 404–416 (2022).
Palma, J. A. et al. Prevalence of REM sleep behavior disorder in multiple system atrophy: a multicenter study and meta-analysis. Clin. Auton. Res. 25 , 69–75 (2015).
Jennings, D. et al. Imaging prodromal Parkinson disease: the Parkinson Associated Risk Syndrome Study. Neurology 83 , 1739–1746 (2014).
Schade, S. et al. Identifying prodromal NMS in a population-based recruitment strategy: Kassel data of Healthy Brain Ageing. Zenodo (2023).
Schade, S. et al. Acute levodopa challenge test in patients with de novo Parkinson’s disease: data from the DeNoPa cohort. Mov. Disord. Clin. Pr. 4 , 755–762 (2017).
Hummel, T., Sekinger, B., Wolf, S. R., Pauli, E. & Kobal, G. ‘Sniffin’ sticks’: olfactory performance assessed by the combined testing of odor identification, odor discrimination and olfactory threshold. Chem. Senses 22 , 39–52 (1997).
Mollenhauer, B. et al. Monitoring of 30 marker candidates in early Parkinson disease as progression markers. Neurology 87 , 168–177 (2016).
Mollenhauer, B. et al. α-Synuclein and tau concentrations in cerebrospinal fluid of patients presenting with parkinsonism: a cohort study. Lancet Neurol. 10 , 230–240 (2011).
UniProt. BLAST https://www.uniprot.org/blast/ . Accessed January 2024.
MacLean, B. et al. Skyline: an open source document editor for creating and analyzing targeted proteomics experiments. Bioinformatics 26 , 966–968 (2010).
Chambers, M. C. et al. A cross-platform toolkit for mass spectrometry and proteomics. Nat. Biotechnol. 30 , 918–920 (2012).
Seabold, S. & Perktold, J. Statsmodels: econometric and statistical modeling with Python. In Proc. 9th Python in Science Conference (SciPy, 2010).
Virtanen, P. et al. SciPy 1.0: fundamental algorithms for scientific computing in Python. Nat. Methods 17 , 261–272 (2020).
Pedregosa, F. et al. Scikit-learn: machine learning in Python. J. Mach. Learn. Res. 12 , 2825–2830 (2011).
MathSciNet Google Scholar
Hunter, J. D. Matplotlib: A 2D graphics environment. Comput. Sci. Eng. 9 , 90–95 (2007).
Sherman, B. T. et al. DAVID: a web server for functional enrichment analysis and functional annotation of gene lists (2021 update). Nucleic Acids Res. 50 , W216–221, (2022).
Huang Da, W., Sherman, B. T. & Lempicki, R. A. Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat. Protoc. 4 , 44–57 (2009).
Shannon, P. et al. Cytoscape: a software environment for integrated models of biomolecular interaction networks. Genome Res. 13 , 2498–2504 (2003).
Download references
This work was supported by the Michael J Fox Foundation, PDUK, The Peto Foundation, The TMSRG (UCL), The BRC at Great Ormond Street Hospital, and the Horizon 2020 Framework Programme (Grant number 634821, PROPAG-AGING). We thank the PROPAG-AGING consortium, a full list of the members can be found in the supplementary material.
Open Access funding enabled and organized by Projekt DEAL.
These authors contributed equally: Jenny Hällqvist, Michael Bartl.
These authors jointly supervised this work: Kevin Mills, Brit Mollenhauer.
UCL Institute of Child Health and Great Ormond Street Hospital, London, UK
Jenny Hällqvist, Ivan Doykov, Justyna Śpiewak, Héloїse Vinette & Wendy E. Heywood
UCL Queen Square Institute of Neurology, Clinical and Movement Neurosciences, London, UK
Jenny Hällqvist & Kevin Mills
Department of Neurology, University Medical Center Goettingen, Goettingen, Germany
Michael Bartl, Mohammed Dakna, Mary Xylaki, Sandrina Weber & Brit Mollenhauer
Institute for Neuroimmunology and Multiple Sclerosis Research, University Medical Center Goettingen, Goettingen, Germany
Michael Bartl
Paracelsus-Elena-Klinik, Kassel, Germany
Sebastian Schade, Maria-Lucia Muntean, Friederike Sixel-Döring, Claudia Trenkwalder & Brit Mollenhauer
Department of Experimental, Diagnostic, and Specialty Medicine (DIMES), University of Bologna, Bologna, Italy
Paolo Garagnani, Chiara Pirazzini & Claudio Franceschi
IRCCS Istituto delle Scienze Neurologiche di Bologna, Bologna, Italy
Maria-Giulia Bacalini
National Hospital for Neurology & Neurosurgery, Queen Square, WC1N3BG, London, UK
Kailash Bhatia & Sebastian Schreglmann
Institute of Diagnostic and Interventional Neuroradiology, University Medical Center Goettingen, Goettingen, Germany
Marielle Ernst
Department of Neurology, Philipps-University, Marburg, Germany
Friederike Sixel-Döring
UCL: Food, Microbiomes and Health Institute Strategic Programme, Quadram Institute Bioscience, Norwich Research Park, Norwich, UK
Héloїse Vinette
Department of Neurosurgery, University Medical Center Goettingen, Goettingen, Germany
Claudia Trenkwalder
You can also search for this author in PubMed Google Scholar
J.H., M.B., K.M., and B.M. conceptualised, planned and oversaw all aspects of the study. J.H., K.M., J.S., H.V., M.B. and S. Schreglmann performed and analyzed most of the experiments. S. Schade, S.W. and M.B. consented to the subjects and collected the samples. M.-L.M., F.S.-D. and S. Schade analyzed the sleep lab data and diagnosed the iRBD subjects. J.H. and M.D. performed the statistical data analysis. J.H. applied the machine learning methods and designed the figures. W.H., I.D., C.F., M.-G.B., P.G., C.P., K.B. and M.X. provided substantial contributions to the conception of the work, acquisition and interpretation of the data, particularly for the mass spectrometry setup and the refinement of the targeted panel. S. Schade, S.W., C.T., M.B., B.M., M.-L.M. and F.S.D. conceptualised the clinical study, analyzed the clinical data and reevaluated the diagnosis. M.E. provided substantial contributions to the clinical data analyzes, particularly the imaging patient data in regard to differential diagnosis. J.H., M.B., K.M. and B.M. wrote the manuscript with input and substantial revisions from all authors.
Correspondence to Jenny Hällqvist or Michael Bartl .
Competing interests.
JH, MD, MX, SW, KB, ME, PG, MGB, CP, KM, ID, WH, JS, HV and CF and have no competing interests to report. MB has received funding from the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) – 413,501,650. CT has received honoraria for consultancy from Roche, and honoraria for educational lectures from UCB, and has received research funding for the PPMI study from the Michael J. Fox Foundation and funding from the EU (Horizon 2020) and stipends from the (International Parkinson’s and Movement Disorder Society) IPMDS. BM has received honoraria for consultancy from Roche, Biogen, AbbVie, UCB, and Sun Pharma Advanced Research Company. BM is a member of the executive steering committee of the Parkinson Progression Marker Initiative and PI of the Systemic Synuclein Sampling Study of the Michael J. Fox Foundation for Parkinson’s Research and has received research funding from the Deutsche Forschungsgemeinschaft (DFG), EU (Horizon 2020), Parkinson Fonds Deutschland, Deutsche Parkinson Vereinigung, Parkinson’s Foundation and the Michael J. Fox Foundation for Parkinson’s Research. MLM has received honoraria for speaking engagements from Deutsche Parkinson Gesellschaft e.V., and royalties from Gesellschaft fur Medien + Kommunikation mbH + Co. FSD has received honoraria for speaking engagements from AbbVie, Bial, Ever Pharma, Medtronic and royalties from Elsevier and Springer. She served on an advisory board for Zambon and Stada Pharma. FSD participated in Ad Boards for consultation: Abbvie, UCB, Bial, Ono, Roche and got honorary for lecturing: Stada Pharm, AbbVie, Alexion, Bial. S. Schade received institutional salaries supported by the EU Horizon 2020 research and innovation programme under grant agreement No. 863664 and by the Michael J. Fox Foundation for Parkinson’s Research under grant agreement No. MJFF-021923. He is supported by a PPMI Early Stage Investigators Funding Programme fellowship of the Michael J. Fox Foundation for Parkinson’s Research under grant agreement No. MJFF-022656. S. Schreglmann received institutional salaries supported by the EU Horizon 2020 research and innovation programme under grant agreement No. 863664, support from the Advanced Clinician Scientist programme by the Interdisciplinary Centre for Clinical Research, Wuerzburg, Germany, and from the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) Project-ID 424778381-TRR 295. He is a fellow of the Thiemann Foundation. He serves as a scientific adviser to Elemind Inc.
Peer review information.
Nature Communications thanks Kanta Horie and the other, anonymous, reviewers for their contribution to the peer review of this work. A peer review file is available.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary information, peer review file, reporting summary, source data, source data, rights and permissions.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ .
Reprints and permissions
Cite this article.
Hällqvist, J., Bartl, M., Dakna, M. et al. Plasma proteomics identify biomarkers predicting Parkinson’s disease up to 7 years before symptom onset. Nat Commun 15 , 4759 (2024). https://doi.org/10.1038/s41467-024-48961-3
Download citation
Received : 06 April 2023
Accepted : 20 May 2024
Published : 18 June 2024
DOI : https://doi.org/10.1038/s41467-024-48961-3
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.
Provided by the Springer Nature SharedIt content-sharing initiative
By submitting a comment you agree to abide by our Terms and Community Guidelines . If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.
Sign up for the Nature Briefing: Translational Research newsletter — top stories in biotechnology, drug discovery and pharma.
Help | Advanced Search
Title: towards natural language-driven assembly using foundation models.
Abstract: Large Language Models (LLMs) and strong vision models have enabled rapid research and development in the field of Vision-Language-Action models that enable robotic control. The main objective of these methods is to develop a generalist policy that can control robots with various embodiments. However, in industrial robotic applications such as automated assembly and disassembly, some tasks, such as insertion, demand greater accuracy and involve intricate factors like contact engagement, friction handling, and refined motor skills. Implementing these skills using a generalist policy is challenging because these policies might integrate further sensory data, including force or torque measurements, for enhanced precision. In our method, we present a global control policy based on LLMs that can transfer the control policy to a finite set of skills that are specifically trained to perform high-precision tasks through dynamic context switching. The integration of LLMs into this framework underscores their significance in not only interpreting and processing language inputs but also in enriching the control mechanisms for diverse and intricate robotic operations.
Subjects: | Robotics (cs.RO); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG) |
Cite as: | [cs.RO] |
(or [cs.RO] for this version) |
Access paper:.
Code, data and media associated with this article, recommenders and search tools.
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs .
You are accessing a machine-readable page. In order to be human-readable, please install an RSS reader.
All articles published by MDPI are made immediately available worldwide under an open access license. No special permission is required to reuse all or part of the article published by MDPI, including figures and tables. For articles published under an open access Creative Common CC BY license, any part of the article may be reused without permission provided that the original article is clearly cited. For more information, please refer to https://www.mdpi.com/openaccess .
Feature papers represent the most advanced research with significant potential for high impact in the field. A Feature Paper should be a substantial original Article that involves several techniques or approaches, provides an outlook for future research directions and describes possible research applications.
Feature papers are submitted upon individual invitation or recommendation by the scientific editors and must receive positive feedback from the reviewers.
Editor’s Choice articles are based on recommendations by the scientific editors of MDPI journals from around the world. Editors select a small number of articles recently published in the journal that they believe will be particularly interesting to readers, or important in the respective research area. The aim is to provide a snapshot of some of the most exciting work published in the various research areas of the journal.
Original Submission Date Received: .
Find support for a specific problem in the support section of our website.
Please let us know what you think of our products and services.
Visit our dedicated information section to learn more about MDPI.
Towards an all-ireland diamond open access publishing platform: the publishoa.ie project—2022–2024.
Data availability statement, conflicts of interest.
1 | |
2 | . Note the statement that ‘[p]ublishers are committed to delivering on the Open Access agenda, but do require the necessary time, flexibility and funding to do so’. See also UK Research and Innovation (UKRI) policy statement on Open Access. Available online: (accessed on 16 March 2024). |
3 | (accessed on 16 March 2024). |
Click here to enlarge figure
The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
Mahony, J. Towards an All-Ireland Diamond Open Access Publishing Platform: The PublishOA.ie Project—2022–2024. Publications 2024 , 12 , 19. https://doi.org/10.3390/publications12030019
Mahony J. Towards an All-Ireland Diamond Open Access Publishing Platform: The PublishOA.ie Project—2022–2024. Publications . 2024; 12(3):19. https://doi.org/10.3390/publications12030019
Mahony, Jane. 2024. "Towards an All-Ireland Diamond Open Access Publishing Platform: The PublishOA.ie Project—2022–2024" Publications 12, no. 3: 19. https://doi.org/10.3390/publications12030019
Article access statistics, further information, mdpi initiatives, follow mdpi.
Subscribe to receive issue release notifications and newsletters from MDPI journals
IMAGES
VIDEO
COMMENTS
Your research objectives may evolve slightly as your research progresses, but they should always line up with the research carried out and the actual content of your paper. Research aims. A distinction is often made between research objectives and research aims. A research aim typically refers to a broad statement indicating the general purpose ...
Characteristics of research objectives. Research objectives must start with the word "To" because this helps readers identify the objective in the absence of headings and appropriate sectioning in research papers. 5,6. A good objective is SMART (mostly applicable to specific objectives): Specific—clear about the what, why, when, and how
Examples of Specific Research Objectives: 1. "To examine the effects of rising temperatures on the yield of rice crops during the upcoming growth season.". 2. "To assess changes in rainfall patterns in major agricultural regions over the first decade of the twenty-first century (2000-2010).". 3.
Research Objectives. Research objectives refer to the specific goals or aims of a research study. They provide a clear and concise description of what the researcher hopes to achieve by conducting the research.The objectives are typically based on the research questions and hypotheses formulated at the beginning of the study and are used to guide the research process.
The golden thread simply refers to the collective research aims, research objectives, and research questions for any given project (i.e., a dissertation, thesis, or research paper). These three elements are bundled together because it's extremely important that they align with each other, and that the entire research project aligns with them.
The objectives provide a clear direction and purpose for the study, guiding the researcher in their data collection and analysis. Here are some tips on how to write effective research objective: 1. Be clear and specific. Research objective should be written in a clear and specific manner.
To develop a set of research objectives, you would then break down the various steps involved in meeting said aim. For example: This study will investigate the link between dehydration and the incidence of urinary tract infections (UTIs) in intensive care patients in Australia. To achieve this, the study objectives w ill include:
The writing of effective research aims and objectives can cause confusion and concern to new and experienced researchers and learners. This step in your research journey is usually the first written method used to convey your research idea to your tutor. Therefore, aims and objectives should clearly convey your topic, academic foundation, and ...
A research objective is defined as a clear and concise statement of the specific goals and aims of a research study. It outlines what the researcher intends to accomplish and what they hope to learn or discover through their research. Research objectives are crucial for guiding the research process and ensuring that the study stays focused and ...
Once you've decided on your research objectives, you need to explain them in your paper, at the end of your problem statement. Keep your research objectives clear and concise, and use appropriate verbs to accurately convey the work that you will carry out for each one. Example: Verbs for research objectives I will assess … I will compare …
There are typically three main types of objectives in a research paper: Exploratory Objectives: These objectives are focused on gaining a deeper understanding of a particular phenomenon, topic, or issue. Exploratory research objectives aim to explore and identify new ideas, insights, or patterns that were previously unknown or poorly understood.
Research Objectives Examples in Different Fields. The application of research objectives spans various academic disciplines, each with its unique focus and methodologies. To illustrate how the objectives of the study guide a research paper across different fields, here are some research objective examples:
Summary. One of the most important aspects of a thesis, dissertation or research paper is the correct formulation of the aims and objectives. This is because your aims and objectives will establish the scope, depth and direction that your research will ultimately take. An effective set of aims and objectives will give your research focus and ...
5 Examples of Research Objectives. The following examples of research objectives based on several published studies on various topics demonstrate how the research objectives are written: This study aims to find out if there is a difference in quiz scores between students exposed to direct instruction and flipped classrooms (Webb and Doman, 2016).
Research objective 2: This paper implements surveys and personal interviews to determine first-hand feedback from the youth members and the team leaders. Research objective 3: Aiming to compare and contrast, this study determines the positive outcomes of the unity project work between the branches of the youth movement in Belgium, aiming for ...
Here are three simple steps that you can follow to identify and write your research objectives: 1. Pinpoint the major focus of your research. The first step to writing your research objectives is to pinpoint the major focus of your research project. In this step, make sure to clearly describe what you aim to achieve through your research.
A research paper is a piece of academic writing that provides analysis, interpretation, and argument based on in-depth independent research. About us; Disclaimer; ... Formal and objective: Research papers are written in a formal and objective tone, with an emphasis on clarity, precision, and accuracy. They avoid subjective language or personal ...
Formulating research aim and objectives in an appropriate manner is one of the most important aspects of your thesis. This is because research aim and objectives determine the scope, depth and the overall direction of the research. Research question is the central question of the study that has to be answered on the basis of research findings.
Aim: To understand the contribution that local governments make to national level energy policy. Objectives: Conduct a survey of local politicians to solicit responses. Conduct desk-research of local government websites to create a database of local energy policy.
The development of the research question, including a supportive hypothesis and objectives, is a necessary key step in producing clinically relevant results to be used in evidence-based practice. A well-defined and specific research question is more likely to help guide us in making decisions about study design and population and subsequently ...
An in-depth analysis of information creates space for generating new questions, concepts and understandings. The main objective of research is to explore the unknown and unlock new possibilities. It's an essential component of success. Over the years, businesses have started emphasizing the need for research.
The objective of this paper is to provide step by step guides to analyse quantitative data. A common research topic on diabetes is used to illustrate the approach in data analysis.
Why written objectives need to be really SMART. July 2017. British Journal of Healthcare Management 23 (7):324-336. DOI: 10.12968/bjhc.2017.23.7.324. Authors: Osahon Ogbeiwi. South West Yorkshire ...
Objectives This portion of the course covers key library resources: literature databases, academic journals, scholarly monographs, and primary source collections. We also discuss key library services for undergraduates as well as connecting with librarians who specialize in English studies, and search tips that will help make your research more ...
Paper presented at: International Meeting for Autism Research; May 9, 2009; Chicago, IL. Accessed December 6, 2022. ... Objective To develop an objective performance-based tool to aid in early diagnosis and assessment of autism in children younger than 3 years.
The procurement strategy for a construction project should provide the framework to achieve secondary procurement and socio-economic development objectives [2]. However, little attention has been focused on this in theory and practice. This paper addresses that gap by presenting a case study of the innovative targeting strategy developed and successfully implemented on a New Universities ...
Background: The occupational burnout epidemic is a growing issue, and in the United States, up to 60% of medical students, residents, physicians, and registered nurses experience symptoms. Wearable technologies may provide an opportunity to predict the onset of burnout and other forms of distress using physiological markers. Objective: This study aims to identify physiological biomarkers of ...
Schreglmann received institutional salaries supported by the EU Horizon 2020 research and innovation programme under grant agreement No. 863664, support from the Advanced Clinician Scientist ...
Large Language Models (LLMs) and strong vision models have enabled rapid research and development in the field of Vision-Language-Action models that enable robotic control. The main objective of these methods is to develop a generalist policy that can control robots with various embodiments. However, in industrial robotic applications such as automated assembly and disassembly, some tasks ...
The Government of Ireland has set a target of achieving 100% open access to publicly funded scholarly publications by 2030. As a key element of achieving this objective, the PublishOA.ie project was established to evaluate the feasibility of establishing an all-island [Republic of Ireland and Northern Ireland] digital publishing platform for Diamond Open Access journals and monographs designed ...