Abstract: Electronic health records are used to define health conditions used for risk adjustment in health services research, but the accuracy of electronic health records is often unknown. We sought to determine the extent to which predictive model discrimination would improve when data from the Centers for Medicare and Medicaid Services (CMS) was combined with data from the Veterans’ Affairs (VA) Corporate Data Warehouse. We calculated model performance for 12-month mortality after hospitalization for heart failure using inpatient data files from the VA alone vs. combined data from VA and CMS. The model using combined VA-CMS data had higher predictive performance compared to the model using VA-only comorbidities. When available, researchers may need to consider multiple sources of electronic health data for accurate risk adjustment variables in health services research.
Learning Objective 1: Researchers will learn about the discrepancies in the prevalence of medical conditions from two sources of electronic health data, highlighting the need to consider multiples sources of health data for accurate risk adjustment in health services research.
Craig Meyer (Presenter)
San Francisco VA Health Care System
Ning Zhang, San Francisco VA Health Care System
Mary Whooley, San Francisco VA Health Care System