Leveraging Artificial Intelligence to Enhance the Quality and Efficiency of Real-World Evidence Generation in HEOR
Author(s)
Delaroziere I1, Houvet L2, Susplugas V2, Angeloglou JB1, Karcenty JE2, Lacombe T2, Perez A2, Vergnol T1
1OSPI, Lyon, France, 2OSPI, Paris, France
Presentation Documents
OBJECTIVES: Real-world evidence (RWE) studies rely on data from sources outside traditional randomized controlled trials. French studies have focused on analyzing evidence based on traditional structured databases (eg, SNDS, registries and claims data), but this approach has limitations for understanding clinical outcomes.
Recent RWE studies are shifting towards using unstructured data from electronic health records (EHRs), which provide rich medical information but require a high quality data structuration process to enable valid clinical assertions, based on different type of unstructured data (doctor notes, discharge summaries, lab or imaging reports, consensus conference proceedings, etc.).METHODS: Participating site have been selected across French public and private hospitals with Intelligence For Health solution (I4H). Data structuration has been performed on pseudonymized environments hosted within each site, with an automated pipeline using I4H software to rapidly structure EHR data.
Quality processes based on human checks and consistency tests have been put in place at the various stages of data generation.RESULTS: Our study demonstrates that AI can help identify clinically relevant data points critical to the approval process, which are currently not available in structured data fields.
We developed an automated pipeline using Intelligence for Health software (i4H) to rapidly structure EHR data, reducing data capture time from approximately 120 minutes to 30 seconds per patient. This approach also enables simplified replicability and extension of initial studies, optimizing lead times and costs.The data structuration and algorithms validation process take approximately 6 weeks, with equivalent quality output compared to manual data entry. The possibility of iterating data collection easily once the automated pipeline is set up offers a particular interest for recurring data collection.CONCLUSIONS: We explore how AI can improve HEOR through its ability to better translate real-world unstructured data from EHR into structured database, increasing replicability, lead times and costs.
Conference/Value in Health Info
Value in Health, Volume 27, Issue 12, S2 (December 2024)
Code
RWD171
Topic
Methodological & Statistical Research, Real World Data & Information Systems, Study Approaches
Topic Subcategory
Artificial Intelligence, Machine Learning, Predictive Analytics, Distributed Data & Research Networks, Electronic Medical & Health Records, Reproducibility & Replicability
Disease
No Additional Disease & Conditions/Specialized Treatment Areas