The difference between effect modification and covariate confounding

Fadl et al.1 describe a case–control study evaluating whether gestational diabetes mellitus (GDM) is associated with increased risk of future cardiovascular disease (CVD) and this differs for different subgroups women. One subgroup they looked at was weight status (normal or overweight). It concluded that GDM indeed useful marker CVD risk, the increase in indicated by higher women who are overweight. was, therefore, an effect modifier. Effect modification from confounding. Confounding occurs when variable to some extent accounts observed factor. In our example, would mean association between could be explained fact more likely overweight whereas without normal weight. Under scenario, actually representing CVD. This can modelled including (overweight weight) as independent regression model. Such model assume having increases same degree If, however, than then For value predictor might stronger compared investigated informally performing linear on separate subgroups. we perform two regressions characteristic CVD, one ‘overweight’ second ‘normal weight’ (our subgroups). formal approach, tested multivariable ‘interaction GDM’ independant (an interaction test). Then P test indicates sizes significantly different, i.e. estimates usually presented within (or stratified subgroups), example odds ratio factor separately Analysis models accounting modifiers allow researcher identify population groups or, clinical trial setting, participants whom intervention effective. wish evaluate labetalol less effective treating hypertension ethnic Subgroups interest should specified priori, clinically plausible limited, thereby avoiding spurious results due chance. Completed disclosure interests form available view online supporting information. Please note: The publisher not responsible content functionality any information supplied authors. Any queries (other missing content) directed corresponding author article.