During a cross-sectional survey on the association between smoking and chronic obstructive pulmonary disease (COPD) in a factory workforce, the investigator finds a weak association (PR = 1.3, 95% CI 0.95–1.78). The investigator suspects that occupational dust exposure may be confounding the relationship. The prevalence of dust exposure is 60% among smokers and 20% among non-smokers. What is the most appropriate next step to evaluate and control for this confounding?

Explanation

## Evaluating Confounding in Cross-Sectional Studies ### Understanding the Scenario **Key Point:** The investigator suspects **confounding** because: - Dust exposure is associated with both smoking (60% vs. 20%) and likely with COPD. - Dust exposure is not in the causal pathway (smoking → COPD); it is an independent risk factor. - The weak association (PR = 1.3, CI crosses 1.0) may be due to confounding by dust exposure. ### Why Stratified Analysis Is the Best Next Step **High-Yield:** **Stratified analysis** (also called **stratification** or **subgroup analysis**) is the gold-standard method to: 1. **Detect** confounding by comparing the crude and stratum-specific associations. 2. **Control** for confounding by presenting separate estimates for each stratum. 3. **Assess effect modification** (interaction) if associations differ between strata. ### How Stratified Analysis Works ```mermaid flowchart TD A[Crude association: Smoking-COPD]:::outcome A --> B[Stratify by Dust Exposure]:::action B --> C[Stratum 1: Dust-Exposed]:::outcome B --> D[Stratum 2: Dust-Unexposed]:::outcome C --> E{Stratum-specific PRs similar to crude?}:::decision D --> E E -->|Yes, similar| F[No confounding]:::outcome E -->|No, different| G[Confounding present]:::outcome G --> H[Report stratum-specific estimates]:::action ``` ### Interpretation of Results If stratified analysis reveals: - **Stratum-specific associations** are similar to each other but **different from the crude association** → **confounding is present**. - **Stratum-specific associations** are similar to the crude association → **no confounding**. - **Stratum-specific associations** differ from each other → **effect modification** (interaction) is present. ### Comparison of Approaches | Approach | Pros | Cons | When to Use | | --- | --- | --- | --- | | **Stratified Analysis** | Detects and controls confounding; reveals effect modification; easy to interpret | Limited to few strata; loses power with small strata | First-line for suspected confounding | | **Exclusion** | Simple | Removes data; may introduce selection bias; reduces generalizability | Never for confounding control | | **Multivariate Regression** | Handles multiple confounders; maintains sample size | Assumes linear relationships; harder to detect effect modification | After stratified analysis confirms confounding | | **Increase Sample Size** | Improves precision | Does not address confounding; expensive | For underpowered studies without bias | **Warning:** Excluding dust-exposed workers is **not** confounding control — it is **selection bias**. The remaining sample no longer represents the source population. **Clinical Pearl:** In occupational epidemiology, stratification by exposure (dust) is standard practice to disentangle multiple workplace hazards. [cite:Park 26e Ch 9]