Which approaches address confounding in observational clinical research?

• A. Randomization only
• B. Matching and stratification only
• C. Multivariable adjustment and propensity scores only
• D. Both design strategies (randomization, matching, stratification) and analysis strategies (multivariable adjustment, propensity scores)

Answer: (D)

Confounding happens when a third variable is related to both the exposure and the outcome, making the exposure seem to have an effect when it might not. The strongest way to address this in observational studies is to use a combination of design and analysis approaches.

Design strategies aim to make the groups being compared as similar as possible before any analysis. Matching involves pairing or grouping individuals who share key confounding characteristics so the exposed and unexposed groups are comparable. Stratification splits the data into levels of a confounder and examines the exposure effect within each level, keeping that confounder constant across comparisons. While randomization is ideal for eliminating confounding, it’s typically not feasible in observational work; the idea is to apply these design ideas to reduce differences between groups from the start.

Analysis strategies then adjust for any remaining differences. Multivariable adjustment includes confounders in statistical models to isolate the exposure’s effect. Propensity scores summarize the probability of exposure given the confounders and can be used to match, stratify, or weight analyses to balance groups on observed characteristics.

Putting design and analysis together provides the most robust control of confounding, which is why the best approach combines both.