Which approach can reduce confounding in a study?

Which approach can reduce confounding in a study?

• A. Randomization
• B. Stratification
• C. Matching
• D. All of the above

Answer: (D)

Confounding happens when a variable is related to both the exposure and the outcome, distorting the observed association. Randomization reduces confounding by randomly assigning participants to exposure groups, which tends to balance both known and unknown confounders across groups so that differences in outcomes are more likely due to the exposure itself. Stratification reduces confounding by dividing participants into homogeneous subgroups based on the confounder and comparing outcomes within those strata (or by analyzing within a single stratum), which keeps the confounder constant and blocks its influence on the exposure–outcome link. Matching reduces confounding by selecting or pairing participants so that the distribution of confounders is similar across comparison groups, with analysis that accounts for the matched design. Because each of these approaches aims to balance or control for confounding, all of them can reduce confounding in a study. Of course, some residual confounding from unmeasured factors can remain, and overmatching can occur if you match on variables that aren’t true confounders or lie on the causal pathway.

Confounding happens when a variable is related to both the exposure and the outcome, distorting the observed association. Randomization reduces confounding by randomly assigning participants to exposure groups, which tends to balance both known and unknown confounders across groups so that differences in outcomes are more likely due to the exposure itself. Stratification reduces confounding by dividing participants into homogeneous subgroups based on the confounder and comparing outcomes within those strata (or by analyzing within a single stratum), which keeps the confounder constant and blocks its influence on the exposure–outcome link. Matching reduces confounding by selecting or pairing participants so that the distribution of confounders is similar across comparison groups, with analysis that accounts for the matched design.

Because each of these approaches aims to balance or control for confounding, all of them can reduce confounding in a study. Of course, some residual confounding from unmeasured factors can remain, and overmatching can occur if you match on variables that aren’t true confounders or lie on the causal pathway.