A public health researcher in Mumbai conducts a case–control study on the association between oral contraceptive use and myocardial infarction (MI) in women aged 30–50 years. Cases are recruited from the cardiology ward of a tertiary hospital, and controls are selected from the outpatient department of the same hospital. The study finds a strong positive association between oral contraceptive use and MI (OR = 3.2, 95% CI 2.1–4.8). However, when the researchers stratify by smoking status, the association becomes much weaker in both smokers and non-smokers. What type of bias or confounding is most likely operating in this study?

Explanation

## Analysis of the Study Design The study demonstrates a classic example of **Simpson's paradox**, where a strong association observed in the overall population reverses or weakens substantially when the data are stratified by a third variable (smoking status). ### What Happened Here **Key Point:** Simpson's paradox occurs when confounding by a third variable creates a spurious association in the crude analysis, but the association disappears or reverses when stratified by the confounder. In this case: - Oral contraceptive use is associated with smoking (smokers may be more likely to use OCP or vice versa) - Smoking is a strong independent risk factor for MI - When stratified by smoking, the OCP–MI association weakens because smoking was doing most of the causal work ### Why This Is Confounding, Not Bias | Feature | Confounding | Bias | |---------|-------------|------| | **Definition** | A third variable distorts the true association | Systematic error in data collection or selection | | **Stratification effect** | Association changes when stratified | Association does not change with stratification | | **Causality** | Confounder is associated with both exposure and outcome | Bias affects measurement or selection independently | | **Example here** | Smoking is associated with OCP use AND MI risk | — | **High-Yield:** Simpson's paradox is a form of **confounding**, not bias. The key diagnostic feature is that the crude association is explained away (or substantially reduced) by stratification on the confounder. ### Clinical Pearl This finding suggests that smoking, not oral contraceptives per se, is the primary driver of MI risk in this population. The apparent OCP–MI association was **confounded** by smoking status. ## Why Stratification Reveals the Truth 1. Among smokers: OCP use shows weak/no association with MI (smoking is the dominant risk factor) 2. Among non-smokers: OCP use shows weak/no association with MI (baseline MI risk is low) 3. The crude association was inflated because OCP users were enriched for smokers **Mnemonic:** **CONFOUNDING = CRUDE ≠ STRATIFIED** — When crude association disappears on stratification, think confounding.