A new meta-analysis of 75 studies, encompassing 58 prediction models, has identified promising tools for assessing the risk of atrial fibrillation following a stroke. The review evaluated multivariable prediction models designed to guide targeted monitoring. The aim was to optimize stroke care while avoiding unnecessary testing. The findings were reported in Heart Rhythm. 

Among the models analyzed, three demonstrated excellent statistical performance: SAFE (C statistic 0.856), SURF (0.815), and iPAB (0.888). However, when studies with a high risk of bias were excluded, only the SAFE model maintained adequate discrimination. 

None of the models consistently showed strong predictive accuracy when tested in external validation cohorts or in studies with more than 100 AF events.