Confounding

The textbook example is the link between coffee and lung cancer that vanishes once smoking is taken into account. Smokers drink more coffee and also smoke more, so smoking confounds the coffee-cancer relationship. A true confounder must be associated with the exposure, independently affect the outcome, and not lie on the causal pathway between them.

Study design and analysis both offer tools. Randomization, when feasible, distributes confounders evenly across groups and is why randomized trials support causal claims more strongly than observational work. When randomization is impossible, researchers adjust through multivariable regression, stratification, matching, or propensity scores, though these handle only measured confounders. Unmeasured confounding always remains a threat.

Reviewers should ask which confounders a study measured and adjusted for, whether important ones were omitted, and whether the authors overclaim causation from an adjusted association. Residual confounding is a standard limitation that observational papers should acknowledge explicitly.