A cross-sectional survey in a rural Indian district investigates the association between well water consumption and fluorosis. The crude prevalence ratio (PR) for fluorosis among well water users is 2.8 compared to non-users. However, when the data are stratified by age group, the PR in children (5–12 years) is 2.5 and in adolescents/adults (≥13 years) is 2.6. The crude PR of 2.8 lies outside the range of stratum-specific PRs. Which investigation should be performed to quantify the magnitude of confounding by age?

Explanation

## Quantifying Confounding: Mantel-Haenszel Adjustment ### The Clinical Scenario The crude PR (2.8) lies **outside** the range of stratum-specific PRs (2.5 and 2.6). This is a classic sign of **confounding by age**—the crude association is distorted by the unequal distribution of age between well water users and non-users. ### Confounding Detection Criteria **Key Point:** Age is a confounder if: 1. Age is associated with both exposure (well water use) and outcome (fluorosis) ✓ 2. Age is not on the causal pathway ✓ 3. Stratum-specific associations differ from the crude association ✓ (2.8 outside range of 2.5–2.6) ### Why Mantel-Haenszel Adjusted PR? The **Mantel-Haenszel adjusted prevalence ratio** is the gold-standard investigation to: - **Quantify the magnitude of confounding** by calculating a weighted average of stratum-specific PRs - **Control for confounding** by removing the distortion introduced by age - **Compare with crude PR** to assess how much the crude estimate was biased ### Interpretation | Comparison | Conclusion | |------------|------------| | Adjusted PR ≈ Stratum-specific PRs | Age confounds; crude PR is biased | | Adjusted PR ≈ Crude PR | Age does not confound | | Adjusted PR substantially lower than crude | Positive confounding (crude overestimates) | | Adjusted PR substantially higher than crude | Negative confounding (crude underestimates) | **High-Yield:** In this case: - Crude PR = 2.8 - Stratum-specific PRs = 2.5–2.6 - Adjusted PR will be close to 2.5–2.6, confirming that age confounds the association - The **percent change** = $\frac{\text{Crude} - \text{Adjusted}}{\text{Adjusted}} \times 100\%$ quantifies the magnitude of confounding ### Mnemonic **MH-ADJUST** = **M**antel-**H**aenszel **ADJUST**ed estimate to quantify confounding - Weights stratum-specific estimates by the distribution of the confounder - Removes bias introduced by unequal distribution of confounder - Allows direct comparison with crude estimate