Nadarajah R Alsaeed E, Hurdus B, et al. - Moderate predictive ability and high risk of bias were displayed by models externally validated for prediction of incident atrial fibrillation (AF) in community-based electronic health record (EHR). A stronger discriminative performance may be achieved with novel methods.

• This is a systematic review and meta-analysis of multivariable prediction models derived and/or validated in EHRs and/or administrative claims databases for incident AF prediction in the community.

• From Ovid Medline and Ovid Embase, 11 eligible studies describing nine prediction models were included; four eligible for meta-analysis comprising 9,289.959 patients.

• The models that offered a c-statistic with a statistically significant 95% prediction interval (PI) and moderate discriminative performance were: CHADS (Congestive heart failure, Hypertension, Age>75, Diabetes mellitus, prior Stroke or transient ischemic attack) (summary c-statistic 0.674, 95% PI 0.526–0.815), CHA ₂ DS ₂ -VASc (summary c-statistic 0.679; 95% PI 0.531–0.811) and HATCH (Hypertension, Age, stroke or Transient ischemic attack, Chronic obstructive pulmonary disease, Heart failure) (summary c-statistic 0.669; 95% PI 0.513–0.803).

• Eligibility for inclusion in meta-analysis was not satisfied by any of the models if studies at high risk of bias were excluded and certainty of effect estimates was ‘low’.

• Strong discriminative performance was displayed by models derived by machine learning but they lacked rigorous external validation.