In a cohort study examining the association between oral contraceptive use and venous thromboembolism, smoking status is found to be associated with both contraceptive use and thromboembolism risk. What is the most appropriate method to control for this confounding variable?

Explanation

## Controlling Confounding in Analytical Studies **Key Point:** Stratification is a powerful analytical method to control for confounding after data collection, allowing assessment of stratum-specific associations and detection of effect modification. ### Why Stratification Is the Best Answer 1. **Separates the confounder's effect:** - Divides the study population into strata (smokers and non-smokers) - Calculates the relative risk (RR) of VTE separately within each stratum - Removes the mixing of effects caused by the confounder 2. **Allows detection of effect modification:** - If stratum-specific RRs differ substantially, smoking may modify the effect of oral contraceptives on VTE - If stratum-specific RRs are similar, a Mantel-Haenszel adjusted RR can be calculated 3. **Maintains internal validity:** - No loss of data (unlike exclusion) - Transparent and easy to interpret ### Methods to Control Confounding | Method | Timing | Mechanism | Pros | Cons | |--------|--------|-----------|------|------| | **Stratification** | Analytical (post-hoc) | Separates confounder effect by subgroup | Transparent, detects effect modification | Reduces sample size per stratum | | **Matching** | Design phase | Ensures confounder balance between groups | Efficient, reduces confounding at entry | Cannot assess confounder's independent effect; may introduce bias | | **Multivariate analysis** | Analytical | Adjusts for confounder statistically | Handles multiple confounders; maintains power | Less transparent; assumes linear relationships | | **Restriction** | Design phase | Limits study to one level of confounder | Simple; eliminates confounding | Reduces generalizability; loss of data | **High-Yield:** **Stratification = Analytical control; Matching = Design-phase control; Restriction = Design-phase control (least preferred).** **Mnemonic:** **SMART** — **S**tratification, **M**atching, **A**djustment (multivariate), **R**estriction, **T**emporal precedence (confounder must precede exposure). **Clinical Pearl:** In observational studies, stratification followed by Mantel-Haenszel adjustment is the gold standard for controlling a single confounder when stratum-specific estimates are homogeneous. [cite:Park 26e Ch 8]