Online Program

Incorporating Cross-Functional Input to Account for Contextual Factors in the Analytic Design of Real-World Studies That Leverage Health Care Data (309900)

Keywords: electronic health records/EHR, real world evidence/RWE or observational data, clustering methods, linear mixed model or mixed methods

We used electronic health record (EHR) data from the American Academy of Ophthalmology IRIS® Registry (Intelligent Research in Sight), the nation’s first comprehensive eye disease clinical database, to study changes in vision over time. As of January 2021, 367 million patient visits from 66 million unique patients exist in the database. Because EHRs are built for care not clinical research, we needed to design a study that incorporated contextual factors that would have led to incorrect conclusions had they gone unaddressed. We incorporated cross-functional input from clinicians, data scientists, and epidemiologists, and this ideation process revealed that the outcome, visual acuity measurements, is collected in unbalanced time intervals per eye and nested within patients. We implemented an analysis with linear mixed models and incorporated random effects to account for hierarchical clustering at the eye level. Our results were modulated by different clustering methods and by accounting for non-independence of patient eyes. From this case study, we demonstrate the importance of cross-functional perspectives and the downstream impact of analytic decisions on the design of a study.