Statistical heterogeneity in meta-analysis of hypertension and exercise training: A meta-review.

ABSTRACT

We aim to evaluate the prevalence of reporting and the extent of statistical heterogeneity of systematic reviews with meta-analyses (SRMAs) of exercise training for hypertension and to provide practical recommendations for dealing with statistical heterogeneity. We systematically searched on four databases (from 2002 to September of 2023) for SRMAs comparing exercise interventions vs. a non-exercise control on blood pressure (BP) control in adults with hypertension. Fifty-nine SRMAs, with a median of 14 studies, were analysed. Cochran's Q (41%), I2 (24%), forest plots (44%), and particularly τ2 (54%) and prediction intervals (96.6%) frequently were not reported for the hypertension subgroup. The recalculated prediction intervals were discrepant (i.e., crossed the null effect) of significant 95% confidence intervals of most meta-analyses (systolic BP 65%; diastolic BP 92%). This suggests substantial heterogeneity across studies, which was often not acknowledged by authors' conclusions (78%). Consequently, downgrading the certainty of the available evidence may be justified alone due to heterogeneity across studies. Finally, we illustrate areas for improving I2 interpretation and provide practical recommendations on how to address statistical heterogeneity across all stages of a SRMA.