Confounding

Confounding is a distortion of the estimated association between an exposure and an outcome caused by a third variable — the confounder — that is independently associated with both. Classic criteria require the confounder to be a cause (or proxy) of the outcome and to be unequally distributed between exposed and unexposed groups, without lying on the causal pathway. In longevity research, the healthy-user bias exemplifies this: people who adopt a preventive intervention (e.g., statin therapy, caloric restriction, exercise) tend to have healthier baseline behaviors and socioeconomic profiles, so apparent survival benefits in observational cohorts may reflect those unmeasured advantages rather than the intervention itself. Adjustment methods — multivariable regression, propensity-score matching, inverse-probability weighting, and Mendelian randomization — can reduce but rarely eliminate confounding, because unmeasured or poorly measured variables leave residual confounding; Shrank et al. (2011) showed that even extensive covariate adjustment failed to fully remove healthy-user bias in Medicare pharmacoepidemiology studies. The E-value, introduced by VanderWeele and Ding (2017), quantifies how strong an unmeasured confounder would need to be — on the risk-ratio scale — to fully explain an observed association, giving readers a practical yardstick for the robustness of any observational longevity finding.