lfischbe/diss · Datasets at Hugging Face

text stringlengths 1 1.75k
A selection of (unanswered) questions, generalized from the literature:
* How can we tell if AI-PROMs are designed and implemented successfully?
* How can AI systems be designed to incorporate and analyze patient-reported outcome measures effectively?
* What types of PROMs would best suit the target population for a specific study or intervention, considering factors such as disease severity, age range, cultural background, etc.?
* Are there existing validated PROMs that can be used in the context of AI-based interventions, and how do we ensure their applicability to the given scenario?
* How can patients be engaged as partners in designing, testing, and implementing AI-based healthcare solutions that incorporate their experiences and perspectives?
* How has value-based healthcare influenced the development of AI in healthcare?
* What proxies are available—for experience? For expertise?
* What are the impacts of AI-PROMs on patient outcomes, satisfaction, and overall healthcare costs?
* How feasible and effective it is to integrate AI-PROMs into existing clinical workflows? To adjust treatment dynamically, based on patterns in patient-reported feedback?
Conclusion:
The current landscape for both AI and PROMs is characterized by increased adoption and efforts to increase standardization and interoperability. While challenges remain, the benefits of using AI and PROMs to improve patient outcomes and healthcare quality make them each an essential component of modern healthcare.
The integration of AI with PROMs has shown promising results in enhancing healthcare services, improving patient outcomes, and optimizing treatment decisions.
By applying AI tools to the analysis of PROMs, researchers gain a more holistic understanding of the complex relationships between diverse inputs (such as demographics, medical history, and treatment plans) and outputs (patient experiences and quality of life). This information may enable healthcare providers to make better-informed decisions that prioritize patients' individual needs and preferences. Additionally, using PROs as output from AI-based predictive tools may help to support interventions and treatment decisions aligned with desired health outcomes.
Conversely, the use of AI algorithms that incorporate PROM data as an input ensures the inclusion of patients' perspectives in AI-based predictions. This persistent validation of patient experience may help to bridge gaps between patients and healthcare providers, facilitating more transparent communication about patient experiences and preferences. If more collaborative, intersubjective relationships between patients and providers is possible, then better-informed decision-making, improved trust in the healthcare system, and a more equitable distribution of resources may follow.
Suggestions for future work:
Further review of the literature should contain an expanded bibliometric analysis: To what extent does the literature build on a shared research tradition, with shared assumptions? Are there divergent assumptions motivating the research? How collaborative is the research, between and within disciplines? How ‘common’ are the questions, objectives, and methods? Where is the research taking place—how geographically or institutionally diverse are the researchers?
Ethics: Continue evaluation of the ethical implications of using AI and PROMs together, with particular attention to the underlying ethical assumptions that ground issues around patient autonomy, privacy, data ownership, algorithmic transparency, and the potential exacerbation of health disparities due to biased algorithms.
Safety: Develop guidelines and best practices for protecting patient privacy when using PROMs data as training data for AI models.
Real-world implementation: Conduct pilot studies or small-scale implementations of AI-PROM systems in real-world clinical settings to evaluate feasibility, effectiveness, and scalability. A literature review that uses a realist synthesis methodology would provide guidance for implementing and evaluating the technologies in real-world settings. Realist here means somewhere between positivism (everything can be cleanly measured) and constructivism (everything is a sociotechnical construct). Synthesis means tracing lines of relation: unpacking the relationship between contexts, mechanisms, and outcomes. What works in one context may work differently in another. The context of the research includes which questions are asked, the way they are asked, and where the motivation for the research originates, for example, from within medical, engineering, social science, or administrative settings. The kind of medicine (e.g. oncology), the particular patient population, and the characteristics of the patient-generated data all contribute meaningfully to the overall research context. A realist synthesis approach is intended to produce secondary interpretations of primary evidence that “illuminate issues and understand contextual influences on whether, why and how interventions might work.” What are the contexts and settings where AI and PROMs work best together? What are the differences in how AI-PROM mechanisms work across contexts? What contextual factors may inhibit successful implementation of AI-PROMs?
Multidisciplinary collaboration: Foster collaborations between clinicians, researchers, data scientists, industry experts, and regulatory bodies to develop solutions for integrating PROMs with AI, and to build consensus around standardized protocols and guidelines.
Standardization: Establish common standards for collecting, processing, and integrating PGHD with AI. Make use of existing secure standards for data interoperability such as FHIR. Harmonize standards for access and ethical shared use of AI-PROMs across studies, countries, and fields of study. Support for innovative development of new AI-PROM tools must also facilitate open research, continuously improve the validity of results, and protect patients’ privacy.
Apply AI-PROMs to new domains: AI-PROM systems may be particularly well-suited to address unmet needs in areas such as mental health, palliative care, and rare diseases. AI-PROMs may be used to assess the effectiveness of interventions or treatments that do not have established PROMs measures, as well as to expand the scope as to what kinds of outcomes can be measured and evaluated.
Addendum
Figures:

Fig 1: Features of a hypothetical AI-PROM as a flow diagram. Patients, providers, and AI are all in a shared loop, with personalized, adaptable, efficient, accessible, and effective care as the intended goals.
Bibliographic Notes
(link to bibliography of included references)
Total references included:
n=537
Unique sources (Journals, Books, Conferences, etc):
n=318
Individual authors:
n=3863

Publications by year:
2020
67
2021
112
2022
129
2023
158
2024 (1/1—4/10)
71

Publications by type:
peer-reviewed articles
410
reviews
69
conference papers
17
book chapters
10
protocols
21
preprints
6
editorials
4

Research Areas
Research Area
Count
Machine Learning
261
Health Informatics
162
Oncology
112
Orthopedics
58
Medical Informatics
51
Health Care Sciences & Services
49
Surgery
40
Neurosciences & Neurology
25
General & Internal Medicine
23
Sport Sciences
23
Computer Science
21
Engineering
19
Gastroenterology & Hepatology
12
Cardiovascular System & Cardiology
11
Science & Technology - Other Topics
11
Public, Environmental & Occupational Health
10
Rheumatology

A selection of (unanswered) questions, generalized from the literature:

* How can we tell if AI-PROMs are designed and implemented successfully?

* How can AI systems be designed to incorporate and analyze patient-reported outcome measures effectively?

* What types of PROMs would best suit the target population for a specific study or intervention, considering factors such as disease severity, age range, cultural background, etc.?

* Are there existing validated PROMs that can be used in the context of AI-based interventions, and how do we ensure their applicability to the given scenario?

* How can patients be engaged as partners in designing, testing, and implementing AI-based healthcare solutions that incorporate their experiences and perspectives?

* How has value-based healthcare influenced the development of AI in healthcare?

* What proxies are available—for experience? For expertise?

* What are the impacts of AI-PROMs on patient outcomes, satisfaction, and overall healthcare costs?

* How feasible and effective it is to integrate AI-PROMs into existing clinical workflows? To adjust treatment dynamically, based on patterns in patient-reported feedback?

Conclusion:

The current landscape for both AI and PROMs is characterized by increased adoption and efforts to increase standardization and interoperability. While challenges remain, the benefits of using AI and PROMs to improve patient outcomes and healthcare quality make them each an essential component of modern healthcare.

The integration of AI with PROMs has shown promising results in enhancing healthcare services, improving patient outcomes, and optimizing treatment decisions.

By applying AI tools to the analysis of PROMs, researchers gain a more holistic understanding of the complex relationships between diverse inputs (such as demographics, medical history, and treatment plans) and outputs (patient experiences and quality of life). This information may enable healthcare providers to make better-informed decisions that prioritize patients' individual needs and preferences. Additionally, using PROs as output from AI-based predictive tools may help to support interventions and treatment decisions aligned with desired health outcomes.

Conversely, the use of AI algorithms that incorporate PROM data as an input ensures the inclusion of patients' perspectives in AI-based predictions. This persistent validation of patient experience may help to bridge gaps between patients and healthcare providers, facilitating more transparent communication about patient experiences and preferences. If more collaborative, intersubjective relationships between patients and providers is possible, then better-informed decision-making, improved trust in the healthcare system, and a more equitable distribution of resources may follow.

Suggestions for future work:

Further review of the literature should contain an expanded bibliometric analysis: To what extent does the literature build on a shared research tradition, with shared assumptions? Are there divergent assumptions motivating the research? How collaborative is the research, between and within disciplines? How ‘common’ are the questions, objectives, and methods? Where is the research taking place—how geographically or institutionally diverse are the researchers?

Ethics: Continue evaluation of the ethical implications of using AI and PROMs together, with particular attention to the underlying ethical assumptions that ground issues around patient autonomy, privacy, data ownership, algorithmic transparency, and the potential exacerbation of health disparities due to biased algorithms.

Safety: Develop guidelines and best practices for protecting patient privacy when using PROMs data as training data for AI models.

Real-world implementation: Conduct pilot studies or small-scale implementations of AI-PROM systems in real-world clinical settings to evaluate feasibility, effectiveness, and scalability. A literature review that uses a realist synthesis methodology would provide guidance for implementing and evaluating the technologies in real-world settings. Realist here means somewhere between positivism (everything can be cleanly measured) and constructivism (everything is a sociotechnical construct). Synthesis means tracing lines of relation: unpacking the relationship between contexts, mechanisms, and outcomes. What works in one context may work differently in another. The context of the research includes which questions are asked, the way they are asked, and where the motivation for the research originates, for example, from within medical, engineering, social science, or administrative settings. The kind of medicine (e.g. oncology), the particular patient population, and the characteristics of the patient-generated data all contribute meaningfully to the overall research context. A realist synthesis approach is intended to produce secondary interpretations of primary evidence that “illuminate issues and understand contextual influences on whether, why and how interventions might work.” What are the contexts and settings where AI and PROMs work best together? What are the differences in how AI-PROM mechanisms work across contexts? What contextual factors may inhibit successful implementation of AI-PROMs?

Multidisciplinary collaboration: Foster collaborations between clinicians, researchers, data scientists, industry experts, and regulatory bodies to develop solutions for integrating PROMs with AI, and to build consensus around standardized protocols and guidelines.

Standardization: Establish common standards for collecting, processing, and integrating PGHD with AI. Make use of existing secure standards for data interoperability such as FHIR. Harmonize standards for access and ethical shared use of AI-PROMs across studies, countries, and fields of study. Support for innovative development of new AI-PROM tools must also facilitate open research, continuously improve the validity of results, and protect patients’ privacy.

Apply AI-PROMs to new domains: AI-PROM systems may be particularly well-suited to address unmet needs in areas such as mental health, palliative care, and rare diseases. AI-PROMs may be used to assess the effectiveness of interventions or treatments that do not have established PROMs measures, as well as to expand the scope as to what kinds of outcomes can be measured and evaluated.

Addendum

Figures:

Fig 1: Features of a hypothetical AI-PROM as a flow diagram. Patients, providers, and AI are all in a shared loop, with personalized, adaptable, efficient, accessible, and effective care as the intended goals.

Bibliographic Notes

(link to bibliography of included references)

Total references included:

n=537

Unique sources (Journals, Books, Conferences, etc):

n=318

Individual authors:

n=3863

Publications by year:

2020

67

2021

112

2022

129

2023

158

2024 (1/1—4/10)

71

Publications by type:

peer-reviewed articles

410

reviews

69

conference papers

17

book chapters

10

protocols

21

preprints

6

editorials

4

Research Areas

Research Area

Count

Machine Learning

261

Health Informatics

162

Oncology

112

Orthopedics

58

Medical Informatics

51

Health Care Sciences & Services

49

Surgery

40

Neurosciences & Neurology

25

General & Internal Medicine

23

Sport Sciences

23

Computer Science

21

Engineering

19

Gastroenterology & Hepatology

12

Cardiovascular System & Cardiology

11

Science & Technology - Other Topics

11

Public, Environmental & Occupational Health

10

Rheumatology