How can confounding be controlled in cohort studies?

• A. Design (randomization if possible, stratification, restriction, matching) and analysis (multivariable regression, stratification, propensity scores, instrumental variables).
• B. By increasing sample size alone.
• C. By using cross-sectional data only.
• D. By ignoring potential confounders.

Answer: (A)

Confounding is a distortion of the true exposure–outcome relationship caused by a third factor that is related to both the exposure and the outcome. In cohort studies you can control confounding through both how you design the study and how you analyze the data. On the design side, you can use randomization if feasible (often in randomized trials, or natural experiments where assignment is effectively random), and you can employ strategies like stratification, restricting the study to specific levels of a potential confounder, or matching exposed and unexposed groups so they have similar distributions of key confounders. On the analysis side, you adjust for confounders in multivariable regression, stratify analyses by confounder levels, use propensity scores to balance measured confounders between exposure groups, or apply instrumental variables when a valid instrument exists to address unmeasured confounding. Increasing sample size helps with precision but does not remove confounding, cross-sectional data lack proper temporality for cohort-style control, and ignoring confounders introduces bias rather than controlling it.