## Identifying Confounding Through Stratification **Key Point:** When crude association is substantially stronger than stratum-specific associations, and the stratum-specific estimates are similar to each other, the third variable is a **confounder**. ### Analysis of the Data | Stratum | Relative Risk | 95% CI | |---------|---------------|--------| | **Crude** | 1.8 | 1.5–2.1 | | **Smokers** | 1.2 | 0.9–1.5 | | **Non-smokers** | 1.1 | 0.8–1.4 | **Observations:** 1. Crude RR (1.8) is **substantially higher** than both stratum-specific RRs (1.2 and 1.1) 2. Stratum-specific RRs are **similar to each other** (homogeneous) 3. Both stratum-specific CIs include 1.0 (no significant association within strata) ### Confounding Criteria Met Smoking meets all three criteria for a confounder: 1. **Associated with exposure:** Shift workers may have higher smoking prevalence (stress, irregular schedules) 2. **Associated with outcome:** Smoking is a well-established risk factor for MI 3. **Not in causal pathway:** Smoking is an independent risk factor, not caused by shift work itself ### Why Confounding Explains the Pattern The crude RR of 1.8 is inflated because: - Shift workers have higher smoking prevalence - Smoking increases MI risk independently - The association between shift work and MI is confounded by smoking - When stratified by smoking status, the true (unconfounded) association emerges: RR ≈ 1.1–1.2 **High-Yield:** **Confounding by stratification:** Crude association > Stratum-specific associations (similar to each other) = Confounding. ### Comparison with Effect Modification | Feature | Confounding | Effect Modification | |---------|-------------|---------------------| | **Stratum-specific RRs** | Similar to each other | **Substantially different** | | **Crude vs. stratum-specific** | Crude is distorted | Crude is weighted average | | **Interpretation** | Third variable distorts true association | Third variable changes the strength of association | | **Action** | Adjust or stratify to remove bias | Report stratum-specific estimates | **Clinical Pearl:** In this case, stratum-specific RRs are nearly identical (1.2 vs. 1.1), indicating **homogeneity** — the hallmark of confounding, not effect modification.
Sign up free to access AI-powered MCQ practice with detailed explanations and adaptive learning.