confounds

A confound is a variable that influences both the exposure and the outcome, creating a spurious association or masking a real one. Confounding can bias effect estimates and complicate causal interpretation in both experimental and observational studies. It is not simply a correlated factor; it is related to both the presumed cause and the effect in a way that can distort conclusions.

Examples: In a trial comparing a new drug to standard care, age or comorbidity burden may confound effects if older or sicker patients are more likely to receive the drug and also have worse outcomes. In observational research on education and health, socioeconomic status can confound the relationship if it affects both education level and health.

Common sources and related concepts: selection bias, measurement error, and time-varying confounders can introduce confounding. Confounding is distinct from mediation (a variable on the causal pathway) and from colliders, which can bias results when conditioned on them.

Control strategies include randomization, restriction, matching, and statistical adjustment (multivariable regression, stratification, or propensity scores). Other methods include instrumental variables and study designs that reduce bias (placebo, blinding, crossover). Using causal diagrams (DAGs) helps identify potential confounders and plan appropriate adjustments.

Limitations: residual confounding may persist after adjustment, especially in observational data with imperfect measurements. Transparent reporting and sensitivity analyses help assess robustness of conclusions.