Development of a computable phenotype using electronic health records for venous thromboembolism in medical inpatients: the Medical Inpatient Thrombosis and Hemostasis study

Abstract

BackgroundAccurate and efficient methods to identify venous thromboembolism (VTE) events in hospitalized people are needed to support large-scale studies. Validated computable phenotypes using a specific combination of discrete, searchable elements in electronic health records to identify VTE and distinguish between hospital-acquired (HA)–VTE and present-on-admission (POA)–VTE would greatly facilitate the study of VTE, obviating the need for chart review. ObjectivesTo develop and validate computable phenotypes for POA- and HA-VTE in adults hospitalized for medical reasons. MethodsThe population included admissions to medical services from 2010 to 2019 at an academic medical center. POA-VTE was defined as VTE diagnosed within 24 hours of admission, and HA-VTE as VTE identified more than 24 hours after admission. Using discharge diagnosis codes, present-on-admission flags, imaging procedures, and medication administration records, we iteratively developed computable phenotypes for POA-VTE and HA-VTE. We assessed the performance of the phenotypes using manual chart review and survey methodology. ResultsAmong 62,468 admissions, 2693 had any VTE diagnosis code. Using survey methodology, 230 records were reviewed to validate the computable phenotypes. Based on the computable phenotypes, the incidence of POA-VTE was 29.4 per 1000 admissions and that of HA-VTE was 3.6 per 1000 admissions. The POA-VTE computable phenotype had positive predictive value and sensitivity of 88.8% (95% CI, 79.8%-94.0%) and 99.1% (95% CI, 94.0%- 99.8%), respectively. Corresponding values for the HA-VTE computable phenotype were 84.2% (95% CI, 60.8%-94.8%) and 72.3% (95% CI, 40.9%-90.8%). ConclusionWe developed computable phenotypes for HA-VTE and POA-VTE with adequate positive predictive value and sensitivity. This phenotype can be used in electronic health record data–based research.