Prediabetes as a phenotype, not an independent risk factor, of atrial fibrillation

We thank Fauchier and colleagues1 for their thoughtful appraisal of our meta-analysis,2 which raises three concerns: high heterogeneity, residual confounding, and overconfident causal language. We welcome the opportunity to clarify how prediabetes should be interpreted in the context of atrial fibrillation (AF). On heterogeneity, we agree that an I2 of 99.3% limits interpretation of the pooled hazard ratio (HR 1.15), which is why we reported a prediction interval of 0.77–1.72. However, I2 quantifies the proportion of variability attributable to between-study differences rather than chance. In meta-analyses with many or highly precise studies, Cochran’s Q test becomes overpowered, detecting trivial heterogeneity and inflating I2 even when between-study differences are small. The interpretive question is whether the association is directionally consistent and mechanistically plausible, not if heterogeneity exists. In our analysis, point estimates in the basic-adjustment stratum were uniformly positive with I2 = 0%, whereas the comprehensive-adjustment stratum showed I2 = 99.3%, reflecting inconsistent covariate selection across studies rather than true biological divergence. Effects remained consistent across geography and definitions, except impaired glucose tolerance (single cohort), with stable leave-one-out estimates of HR 1.11–1.17. Two factors likely bias the pooled estimate downward. First, the ascertainment of AF via ICD codes or intermittent monitoring misses paroxysmal AF (non-differential misclassification towards the null) and second, the trim-and-fill adjustment shifted the estimate upward to HR 1.36 (1.20–1.53), suggesting under-representation of larger positive effects.