The Perils and Promises of Using Electronic Health Records for Causal Inference: Case Studies on HIV Treatment Guidelines in Kenya
Electronic health records (EHR) provide rich information on healthcare delivery and health outcomes. The uses of EHR information include billing, patient monitoring, and storage of complex information such as images and physician notes.
More recent interest has focused on the utility of EHR for development of prediction rules, clinical decision support, and comparative effectiveness studies. However, unlike data from clinical trials and cohort studies, data from EHR are not collected according to a protocol or study design; in that sense they are 'experiential' or 'found' data. This talk illustrates some of the challenges of using EHR for drawing inference about causal effects, and describes statistical methods needed to address them. We illustrate using two case studies from a large HIV care program in Kenya: one examining timing of antiviral treatment initiation among those with HIV/TB coinfection, and another looking at the causal effect of recently recommended 'test-and-treat' policy.
Joseph Hogan is Professor of Biostatistics at Brown. He conducts research on statistical methods for causal inference, missing data, and sensitivity analysis for under-identified models, and has collaborated extensively with researchers in HIV/AIDS for the past 20 years. Most of his current work is focused on HIV in Kenya and sub-Saharan Africa.