Understanding heterogeneity in prevalence meta-analyses: from structural incompatibility to statistical variation

Résumé fourni par la source

Heterogeneity is a defining and expected feature of prevalence studies and their systematic reviews, yet it is frequently interpreted using conceptual and statistical frameworks developed for intervention research. Unlike treatment effects, prevalence estimates do not represent a single underlying biological parameter but are inherently determined by case definitions, population characteristics, measurement strategies, and contextual factors. Consequently, heterogeneity in prevalence meta-analyses often reflects structural incompatibility between studies rather than random statistical variation around a common effect. This article explains why conventional heterogeneity metrics, particularly the I 2 statistic, are poorly suited to prevalence data and may be misleading when used as decision rules for pooling. We clarify the conceptual distinction between structural heterogeneity and statistical heterogeneity and demonstrate why only the latter can be meaningfully addressed through quantitative synthesis. Building on epidemiological principles and current methodological guidance, we propose a pragmatic, clinically oriented framework to support the interpretation of heterogeneity and to guide decisions regarding pooling, stratification, or alternative synthesis approaches. The aim is to promote more rigorous, transparent, and conceptually coherent synthesis of prevalence evidence, improving its interpretability and usefulness for clinicians, epidemiologists, statisticians, and health policy researchers.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.