A public health researcher in Delhi conducts a study to investigate the association between coffee consumption and myocardial infarction (MI) in adults aged 40–65 years. The study recruits 500 cases (patients with MI) from a tertiary hospital and 500 controls (age-matched, hospital visitors without MI). Cases report significantly higher coffee consumption (mean 4 cups/day) compared to controls (mean 1.5 cups/day), with an odds ratio of 2.8 (95% CI: 2.1–3.7). However, when the researchers stratify the analysis by smoking status, the association disappears in both smokers and non-smokers separately. What type of bias or confounding is most likely responsible for the observed association?

Explanation

## Identification of Confounding **Key Point:** When stratification by a variable eliminates or substantially reduces an association that was present in the crude analysis, that variable is a confounder, not a source of bias. ### Why Smoking is the Confounder Here Smokingacts as a **classical confounder** because it: 1. Is associated with the exposure (coffee consumption) — smokers tend to drink more coffee 2. Is independently associated with the outcome (MI) — smoking is a major risk factor for MI 3. Is not on the causal pathway between coffee and MI When the analysis is stratified by smoking status, the association between coffee and MI disappears in both strata. This is the hallmark of confounding: the apparent association was spurious and due entirely to the confounding variable. ### Stratification as a Confounder Control Method ```mermaid flowchart TD A[Smoking]:::action --> B[↑ Coffee consumption]:::outcome A --> C[↑ MI risk]:::outcome B -.->|Apparent association| C D[True causal pathway] -.-> B D -.-> C ``` **High-Yield:** When crude association disappears on stratification, suspect **confounding**. When association persists across strata, the original association is likely causal or due to effect modification. ### Contrast with Bias | Feature | Confounding | Bias | |---------|-------------|------| | **Definition** | A third variable distorts the association | Systematic error in measurement or selection | | **Controlled by** | Stratification, matching, multivariate analysis | Study design (blinding, standardized protocols) | | **Effect on stratified analysis** | Disappears or reverses | Persists across strata | | **Nature** | Real variable affecting both exposure and outcome | Methodological flaw | **Clinical Pearl:** In observational studies, always consider whether an observed association could be explained by a confounder. Smoking is one of the most common confounders in cardiovascular epidemiology. [cite:Park 26e Ch 9]