Regarding confounding variables in epidemiological studies, all of the following are methods to control for confounding EXCEPT:

Explanation

## Methods to Control for Confounding ### Definition of a Confounder **Key Point:** A confounder is a variable that: 1. Is associated with the exposure 2. Is independently associated with the outcome 3. Is NOT on the causal pathway between exposure and outcome ### Valid Methods to Control for Confounding | Method | Stage | Mechanism | Strength | |--------|-------|-----------|----------| | **Stratification** | Analysis | Separate analysis by confounder levels, then combine results | Good; allows effect modification detection | | **Matching** | Design | Select comparison groups with same confounder distribution | Excellent; prevents confounding at baseline | | **Randomization** | Design | Random allocation distributes all confounders equally | Best; controls for known and unknown confounders | | **Adjustment (Regression)** | Analysis | Statistical control in multivariate models | Good; quantifies independent effects | ### Why Option 4 is Incorrect **High-Yield:** Increasing sample size does NOT control for confounding. It reduces random error (increases precision) but does NOT eliminate systematic bias from confounding. **Warning:** This is a common misconception. Students often confuse: - **Sample size ↑** → reduces random error, improves precision - **Confounding control** → eliminates systematic bias These are independent problems. A large study with confounding is still biased; a small study without confounding is unbiased but imprecise. **Clinical Pearl:** If a confounder is present and not controlled, increasing the sample size will give you a more precise estimate of the WRONG answer. **Mnemonic:** **SMART** — Stratification, Matching, Adjustment (regression), Randomization, and Thinking about causality are valid methods. Size is NOT. ### Flowchart: Confounding Control Strategies ```mermaid flowchart TD A[Confounder identified]:::outcome --> B{When to control?}:::decision B -->|At design stage| C[Matching]:::action B -->|At design stage| D[Randomization]:::action B -->|At analysis stage| E[Stratification]:::action B -->|At analysis stage| F[Regression adjustment]:::action G[Increase sample size]:::urgent --> H[Reduces random error only]:::outcome H --> I[Does NOT control confounding]:::urgent ``` ### Why Each Other Option is Correct - **Option 1 (Stratification):** Valid analysis method. Dividing into strata and analyzing separately allows confounding to be controlled and effect modification to be detected. - **Option 2 (Matching):** Valid design method. Ensures comparison groups are similar with respect to the confounder, preventing confounding at baseline. - **Option 3 (Randomization):** Valid design method. Random allocation distributes both known and unknown confounders equally between groups.