A study in JAMA Cardiology suggests that predictive models validated in clinical studies may not prove accurate when used with EMR data.
The researchers investigated whether a risk prediction model for atrial fibrillation, which had been developed by the Cohorts for Heart and Aging Research in Genomic Epidemiology-Atrial Fibrillation, could be validated using EMRs. For this study, they analyzed 33,494 patients with no history of atrial fibrillation, to determine whether the model effectively predicted the development of atrial fibrillation when used with EMRs.
A total of 2,455 patients, or 7.3 percent of the study participants, developed atrial fibrillation. The researchers found that the predictive model had poor calibration in the EMR cohort; overall, the model underpredicted atrial fibrillation among low-risk participants and overpredicted atrial fibrillation among high-risk participants. They concluded that the study "highlights the difficulties of applying a risk model derived from prospective cohort studies to an EMR cohort and suggests that these AF risk prediction models be used with caution in the EMR setting."
The researchers suggest that future predictive models should be developed and validated within EMR cohorts.