[Skip to Navigation]
December 2016

Risk Prediction With Electronic Health Records: The Importance of Model Validation and Clinical Context

Author Affiliations
  • 1Department of Biostatistics and Bioinformatics, Duke University School of Medicine, Durham, North Carolina
  • 2Center for Predictive Medicine, Duke Clinical Research Institute, Durham, North Carolina
  • 3Division of Cardiology, Duke University School of Medicine, Durham, North Carolina
JAMA Cardiol. 2016;1(9):976-977. doi:10.1001/jamacardio.2016.3826

Historically, risk models have been derived using data from large epidemiologic cohorts or clinical trials. Although these data sources are often high quality, their external generalizability may be limited for at least 2 reasons. First, the populations included in the cohort or trial are often narrowly defined and not representative of all adults.1 Recent efforts to combine data from multiple cohorts have led to risk prediction models with broader external generalizability. The pooled cohort equations used in the 2013 American College of Cardiology/American Heart Association cholesterol guideline2 and the Cohorts for Heart and Aging Research in Genomic Epidemiology–Atrial Fibrillation (CHARGE-AF)3 score were both based on data pooled from multiple observational cohort studies. Second, how data are collected in prospective trials and cohort studies may not match how data are collected in clinical settings.

Add or change institution