An epidemiologist in Mumbai is conducting a case–control study on the association between oral contraceptive use and venous thromboembolism (VTE) in women aged 25–45 years. During data analysis, she observes that cases (women with VTE) are significantly older than controls (mean age 38 vs. 32 years). Age is known to be independently associated with VTE risk. What is the most appropriate next step?

Explanation

## Controlling Confounding in Case–Control Studies: Design vs. Analysis Phase **Key Point:** Once data collection is complete and confounding is identified, analytical methods (stratification or multivariate adjustment) are appropriate. Design-based methods (matching) cannot be retroactively applied. ### Why Multivariate Regression Is Optimal Here In a case–control study where: - Data collection is already complete - Confounding by age has been identified post-hoc - Multiple confounders may exist (smoking, BMI, family history) Multivariate logistic regression is the **most efficient and flexible** analytical approach because: 1. **Simultaneous adjustment:** Adjusts for age and any other confounders in one model 2. **Preserves sample size:** No loss of data due to stratification (which requires adequate numbers in each stratum) 3. **Quantifies adjusted effect:** Provides an adjusted odds ratio (aOR) with confidence intervals 4. **Tests for effect modification:** Can test whether the OC–VTE association differs by age ### Comparison of Confounder Control Methods in Case–Control Studies | Method | Timing | Feasibility | Efficiency | Use Case | |--------|--------|-------------|-----------|----------| | **Matching** | Design phase | Requires restart | Moderate | Prevents confounding at recruitment | | **Stratification** | Analysis phase | Post-hoc possible | Lower (loses power) | Few confounders; simple interpretation needed | | **Multivariate Regression** | Analysis phase | Post-hoc possible | **Highest** | **Multiple confounders; post-hoc adjustment** | | **Restriction** | Design phase | Requires restart | Moderate | Narrows applicability | **High-Yield:** Multivariate logistic regression is the **gold standard** for post-hoc confounder adjustment in case–control studies because it: - Preserves all data - Handles multiple confounders simultaneously - Provides adjusted odds ratios with precision estimates ### Why Stratification Is Suboptimal Here While stratification is valid, it has drawbacks in this scenario: - Age is continuous; creating strata (e.g., 25–34, 35–45) loses information - Smaller numbers in each stratum reduce precision - Comparing multiple stratum-specific ORs is less elegant than a single adjusted estimate **Mnemonic: SMART Adjustment** - **S**tratification — few confounders, simple interpretation - **M**ultivariate — multiple confounders, post-hoc adjustment ← **USE HERE** - **A**nalysis — post-hoc control - **R**egression — flexible, efficient - **T**echnique — preserves power and data **Clinical Pearl:** In published case–control studies, multivariate logistic regression is the standard method for reporting adjusted odds ratios. Reviewers expect to see confounders controlled via regression, not stratification alone.