Reviewing Current Trends Regarding Use of Artificial Intelligence in Generating Real-World Evidence From Patient Medical Charts
Author(s)
Sneha S. Kelkar, MSc., MPH1, Viktor Chirikov, MS, PhD2;
1OPEN Health, Director, New York, NY, USA, 2OPEN Health, Bethesda, MD, USA
1OPEN Health, Director, New York, NY, USA, 2OPEN Health, Bethesda, MD, USA
Presentation Documents
OBJECTIVES: Medical charts are considered the ideal source for contemporary real-world evidence on patient characteristics, treatments, and clinical outcomes. However, medical charts are complex systems generating vast amounts of structured and raw, unstructured data. There has been a surge recently in utilizing artificial intelligence (AI) to optimize aspects of chart reviews. Our study aimed to evaluate the current progress in the application of AI to medical chart reviews and understand the role human intelligence needs to play in this process.
METHODS: We conducted a literature review in MEDLINE (01/01/2023-12/31/2024) combining terms for ‘AI/machine learning’ and ‘medical chart review’ and supplemented it with a targeted review of policy documents, regulatory/HTA guidelines, and stakeholder’s websites. Two experienced researchers reviewed the results; emerging themes and insights were interpreted from the perspective of the ISPOR Strategic Plan 2030.
RESULTS: Our results highlighted that AI is useful in optimizing the following aspects of chart review: i) data collection speed and volume through reducing the burden of manual extraction from unstructured records across multiple sites; ii) richness of the data collected, e.g. for biomarkers and other clinical data obviating the need for external data curation; iii) addressing data missingness via the use of synthetic data to impute incomplete fields. However, significant help of the on-site research team is required to adapt AI to multi-country/multi-system with differential level of health record systems to avoid spurious interpretation. This is particularly relevant to studies assessing highly nuanced outcomes or PRO data often dependent on the version and format of instrument used.
CONCLUSIONS: The observed trends align with ISPOR’s effort for affordability and long-term sustainability of healthcare systems. As multiple agencies advocate for the use of real-world data to advance patient-centered care, future efforts should emphasize the application of AI to chart reviews ensuring reliable, transparent, and unbiased evidence generation.
METHODS: We conducted a literature review in MEDLINE (01/01/2023-12/31/2024) combining terms for ‘AI/machine learning’ and ‘medical chart review’ and supplemented it with a targeted review of policy documents, regulatory/HTA guidelines, and stakeholder’s websites. Two experienced researchers reviewed the results; emerging themes and insights were interpreted from the perspective of the ISPOR Strategic Plan 2030.
RESULTS: Our results highlighted that AI is useful in optimizing the following aspects of chart review: i) data collection speed and volume through reducing the burden of manual extraction from unstructured records across multiple sites; ii) richness of the data collected, e.g. for biomarkers and other clinical data obviating the need for external data curation; iii) addressing data missingness via the use of synthetic data to impute incomplete fields. However, significant help of the on-site research team is required to adapt AI to multi-country/multi-system with differential level of health record systems to avoid spurious interpretation. This is particularly relevant to studies assessing highly nuanced outcomes or PRO data often dependent on the version and format of instrument used.
CONCLUSIONS: The observed trends align with ISPOR’s effort for affordability and long-term sustainability of healthcare systems. As multiple agencies advocate for the use of real-world data to advance patient-centered care, future efforts should emphasize the application of AI to chart reviews ensuring reliable, transparent, and unbiased evidence generation.
Conference/Value in Health Info
2025-05, ISPOR 2025, Montréal, Quebec, CA
Value in Health, Volume 28, Issue S1
Code
MSR29
Topic
Methodological & Statistical Research
Topic Subcategory
Artificial Intelligence, Machine Learning, Predictive Analytics
Disease
No Additional Disease & Conditions/Specialized Treatment Areas