A case-control study examining the association between pesticide exposure and lung cancer in agricultural workers reports an unadjusted odds ratio of 3.2 (95% CI: 2.1–4.8). After adjustment for smoking using multivariate logistic regression, the adjusted OR becomes 2.8 (95% CI: 1.9–4.1). The researchers suspect smoking may be a confounder. Which investigation would be most appropriate to formally assess whether smoking meets the statistical criteria for confounding?

Explanation

## Formal Assessment of Confounding in Case-Control Studies **Key Point:** To formally confirm that a variable meets statistical criteria for confounding, stratified analysis using the Mantel-Haenszel test is the gold standard investigation. It compares crude and adjusted estimates to quantify the degree of confounding. ### Criteria for Confounding A variable is a confounder if: 1. It is associated with the exposure (pesticide use) 2. It is associated with the outcome (lung cancer) 3. It is NOT on the causal pathway (not a mediator) 4. **Adjustment changes the estimate by ≥10%** (statistical criterion) **High-Yield:** Change in OR = $\frac{\text{Crude OR} - \text{Adjusted OR}}{\text{Crude OR}} \times 100$ Here: $\frac{3.2 - 2.8}{3.2} \times 100 = 12.5\%$ → **Smoking is a confounder** ### Why Mantel-Haenszel Stratified Analysis? The MH test: - **Stratifies** the sample by smoking status (smokers vs. non-smokers) - **Calculates** the association between pesticide exposure and lung cancer within each stratum - **Compares** crude OR (3.2) with stratified ORs - **Formally tests** whether the association persists after controlling for smoking - Provides **statistical evidence** (p-value, confidence intervals) of confounding | Aspect | Crude OR | Stratified OR (MH) | Interpretation | |--------|----------|-------------------|----------------| | **Pesticide → Lung cancer** | 3.2 | ~2.8 (pooled) | Smoking explains ~12.5% of association | | **Smoking association** | Not shown | Shown separately | Smoking independently increases lung cancer risk | | **Confounding status** | Unclear | **Confirmed** | MH test formally validates confounding | **Clinical Pearl:** In case-control studies, smoking is a notorious confounder in lung cancer research because it is both a strong risk factor for lung cancer AND differentially distributed between exposed and unexposed groups. ### Why Not Other Options? - **Sensitivity analysis (excluding smokers)**: Tests robustness but does NOT formally assess confounding criteria; it's a supplementary method, not the primary investigation - **Attributable risk percentage**: Measures disease burden attributable to smoking, not whether smoking is a confounder of the pesticide-lung cancer association - **Propensity score matching**: Balances covariates but is a **matching/adjustment technique**, not an investigation to **detect and confirm** confounding **Mnemonic:** **MH = Confounder Validator** — Mantel-Haenszel formally confirms confounding by stratification.