## Cohort Study Analysis & Causal Inference The researcher has conducted a **prospective cohort study** and calculated both **relative risk (RR)** and **odds ratio (OR)**. The RR and OR are close (2.5 vs 2.3), which is expected when disease incidence is low. However, before translating this into occupational health policy, critical quality checks are required. ### Why Verify Incidence, Loss to Follow-Up, and Confounding? **Key Point:** In a cohort study, the **relative risk is the direct and preferred measure** of association. However, the validity of the RR depends on: 1. **Complete follow-up** — loss to follow-up can introduce bias 2. **Competing mortality** — workers may die from other causes before developing lung cancer, artificially reducing observed incidence 3. **Confounding** — smoking is a major confounder in lung cancer studies and must be adjusted for 4. **Effect modification** — the RR may differ by age group or smoking status **High-Yield:** The standard epidemiological next step after calculating RR in a cohort study is: - Assess data quality (completeness of follow-up, missing data) - Perform stratified analysis by known confounders (age, smoking) - Calculate stratum-specific RRs to detect effect modification - Perform sensitivity analysis (e.g., exclude workers lost to follow-up after year 5) - Only then draw causal conclusions ### Why RR ≈ OR in This Study? $$\text{RR} \approx \text{OR when} \quad \frac{\text{Incidence in exposed}}{\text{Incidence in unexposed}} \text{ is small}$$ Here, incidence in exposed = 120/5000 = 2.4%, and in unexposed = 20/5000 = 0.4%. The rare disease assumption is reasonably met, so RR and OR are close. **However, this does not mean we can skip confounder assessment.** ### Why Not the Other Options? | Option | Problem | |--------|----------| | **Report both RR and OR, recommend cessation based on RR** | While the RR is the appropriate measure in cohort studies, recommending immediate policy action without assessing confounding (especially smoking) is premature. Asbestos is a known carcinogen, but the magnitude of effect must be adjusted for confounders. | | **Discard OR as unreliable** | OR is not "unreliable" in cohort studies—it is simply not the primary measure. OR can be useful for secondary analyses, and the fact that RR ≈ OR here actually validates the rare disease assumption. | | **Conduct a nested case-control study** | This is not a "next step"—it would require additional data collection. Cohort studies *do* establish causality if confounding is ruled out and the temporal sequence is clear (which it is here: exposure precedes outcome). | ### Clinical Pearl **Confounding by smoking** is the classic pitfall in occupational lung cancer studies. If exposed workers are more likely to smoke, the observed RR may overestimate the true causal effect of asbestos. Conversely, if unexposed workers smoke more, the RR may underestimate asbestos risk. Stratification by smoking status is mandatory. **Mnemonic: GRADE** — Assessing causality in observational studies: - **G**radient (dose–response) — does higher exposure → higher risk? - **R**eversibility — does removing exposure reduce risk? - **A**nalogy — are there similar exposures with similar effects? - **D**ose–response & **D**uration — temporal and dose relationships - **E**xperimental evidence — do animal or in vitro studies support causality? **Tip:** In NEET PG, whenever you see a cohort study result (RR calculated), the next step is almost always "assess for confounding and loss to follow-up," not "implement the finding immediately."
Sign up free to access AI-powered MCQ practice with detailed explanations and adaptive learning.