Which finding best distinguishes confounding from effect modification in a stratified analysis of a cohort study examining the association between smoking and lung cancer?

Explanation

## Core Distinction **Key Point:** Confounding and effect modification are opposite phenomena: confounding is a **bias to be removed**, while effect modification is a **true biological phenomenon to be reported**. ### Confounding - A **third variable** distorts the true association between exposure and outcome - The confounder is: - Associated with the exposure - Associated with the outcome **independently** - Not on the causal pathway - **Stratum-specific estimates are similar to each other** (homogeneous) - The **crude (pooled) estimate is biased** away from the true effect - **Solution:** Stratify and report adjusted estimate ### Effect Modification (Interaction) - The **strength or direction of the association varies** by levels of a third variable - The third variable **modifies** the effect of the exposure on the outcome - The third variable **IS on the causal pathway** or represents a true biological difference - **Stratum-specific estimates differ meaningfully** (heterogeneous) - **No single pooled estimate is appropriate** — report stratum-specific estimates separately - **Solution:** Report stratified results; do NOT pool ## Comparison Table | Feature | Confounding | Effect Modification | | --- | --- | --- | | **Nature** | Bias (unwanted) | True biological phenomenon (wanted) | | **Stratum-specific estimates** | Similar across strata | Different across strata | | **Crude vs. adjusted** | Crude is biased; adjusted is true | Both are true; report separately | | **Causal pathway** | Confounder NOT on pathway | Modifier IS on pathway | | **Statistical test** | Mantel-Haenszel, adjusted OR | Test for homogeneity (Breslow-Day) | | **Action** | Adjust/control | Stratify and report | ## Stratified Analysis Algorithm ```mermaid flowchart TD A[Stratify by potential third variable]:::action --> B{Are stratum-specific estimates similar?}:::decision B -->|Yes, homogeneous| C[Likely confounding]:::outcome C --> D[Report adjusted/pooled estimate]:::action B -->|No, heterogeneous| E[Likely effect modification]:::outcome E --> F[Report stratum-specific estimates separately]:::action B -->|Borderline| G[Test homogeneity: Breslow-Day test]:::action G --> H{p < 0.05?}:::decision H -->|Yes| I[Effect modification present]:::outcome H -->|No| J[Confounding likely]:::outcome ``` **High-Yield:** **Homogeneous strata = confounding; heterogeneous strata = effect modification.** This is the practical discriminator in any stratified analysis. **Mnemonic:** **COME** — **C**onfounding = **O**ne estimate (pooled/adjusted); **M**odification = **M**ultiple estimates (stratum-specific). **Clinical Pearl:** In a smoking–lung cancer study: - If the association is strong in both males and females (similar ORs), but the crude OR differs from the sex-adjusted OR, **confounding by sex** is present. - If the association is strong in males but weak in females (different ORs), **sex is an effect modifier** — report male and female results separately. [cite:Park 26e Ch 9]