Phenotyping Diabetes Mellitus on Aggregated Electronic Health Records from Disparate Health Systems

Background: Identifying patients with diabetes mellitus (DM) is often performed in epidemiological studies using electronic health records (EHR), but currently available algorithms have features that limit their generalizability. Methods: We developed a rule-based algorithm to determine DM status the nationally aggregated EHR database. The was validated on two chart-reviewed samples (n = 2813) of (a) atrial fibrillation (AF, n 1194) and (b) randomly sampled hospitalized 1619). Results: diagnosis codes alone resulted sensitivity 77.0% 83.4% AF random samples, respectively. proposed combines blood glucose values medication usage diagnostic exhibits sensitivities between 96.9% 98.0%, while positive predictive (PPV) ranged 61.1% 75.6%. Performances were comparable across sexes, lower specificity observed younger (below 65 versus above) both validation (75.8% vs. 90.8% 60.6% 88.8%). robust for missing laboratory data not data. Conclusions: In this nationwide database analysis, an identifying has been validated. supports quantitative bias analyses future involving EHR-based studies.