An observational study examining the relationship between oral contraceptive use and venous thromboembolism (VTE) in 2000 Indian women reports an unadjusted odds ratio of 3.2 (95% CI 2.1–4.8). After adjustment for age, smoking, and obesity, the odds ratio becomes 2.8 (95% CI 1.9–4.1). Which feature best distinguishes selection bias from confounding in this context?

Question

Accepted Answer

B. Selection bias arises from how subjects are chosen into the study, whereas confounding arises from an extraneous variable associated with both exposure and outcome. ## Selection Bias vs. Confounding: Operational Distinction

**Key Point:** Selection bias and confounding are fundamentally different sources of error. Selection bias is a **structural flaw in how subjects enter the study**; confounding is a **variable imbalance between exposure groups**.

### Core Distinction

| Dimension | Selection Bias | Confounding |
|-----------|----------------|-------------|
| **Origin** | **How subjects are selected/recruited** | **An extraneous variable** |
| **Timing** | Occurs at **study design/recruitment** | Occurs at **analysis** |
| **Association** | Affects probability of being selected | Associated with both exposure AND outcome |
| **Adjustment** | **Cannot be fixed by statistical control** | **Can be reduced/eliminated by adjustment** |
| **Example** | Healthy worker effect; volunteer bias | Smoking confounding coffee–MI link |

### In This Scenario

The question presents:
- **Crude OR:** 3.2 (95% CI 2.1–4.8)
- **Adjusted OR:** 2.8 (95% CI 1.9–4.1)
- **Interpretation:** Modest reduction after adjustment

This pattern suggests **confounding** (not selection bias) because:
1. The association **persists but weakens** after adjustment
2. Age, smoking, and obesity are variables that can be measured and controlled
3. Selection bias would not respond predictably to adjustment

**High-Yield:** The **defining feature of selection bias is that it cannot be corrected by statistical adjustment** — it is baked into the study design. Confounding, by contrast, can be partially or fully removed by controlling for the confounder.

**Mnemonic:** **SECO** — **S**election bias = **E**ntry into study; **C**onfounding = **C**ovariate imbalance; **O**ne cannot be fixed, **O**ther can.

[cite:Park 26e Ch 8]

Answer

A. Selection bias increases the confidence interval width, whereas confounding narrows it

Answer

C. Selection bias affects only the exposed group, whereas confounding affects both exposed and unexposed equally

Answer

D. Selection bias is eliminated by stratification, whereas confounding persists despite adjustment

Explanation

Practice similar questions