Multivariable Prediction Models for Atrial Fibrillation after Cardiac Surgery: A Systematic Review and Critical Appraisal.

Atrial fibrillation is a common complication of cardiac surgery. Multiple models exist to estimate the risk of atrial fibrillation after cardiac surgery (AFACS) and improve targeting of preventative measures, yet none have been consistently adopted into clinical use. This study performed a comprehensive systematic review, assessing quality and risk of bias of studies describing the development or external validation of AFACS prediction models. Although some models performed well in development and external validation (median C-statistic for apparent validation alone, 0.71; range, 0.60 to 0.98; external validation, 0.61; range, 0.51 to 0.77), all model analyses were rated at high risk of bias. Common causes for this were small sample size, data-driven predictor selection, and inadequate internal validation. Overall, no individual model could be recommended for clinical use given the methodologic limitations identified, emphasizing the need for improvements in future AFACS prediction models to facilitate improved targeting of prophylaxis.