Mantel Haenszel Estimator
- Produces a weighted average of subgroup (stratum-specific) odds ratios to estimate a common odds ratio.
- Controls for a single confounding factor by stratifying the data (e.g., age or gender).
- Assumes the exposure–outcome relationship is the same in every stratum and handles one confounder at a time.
Definition
Section titled “Definition”The Mantel-Haenszel estimator is a statistical method used to estimate the common odds ratio of a binary outcome, such as the likelihood of developing a certain disease, in different populations or subgroups.
Explanation
Section titled “Explanation”The method controls for confounding by stratifying data according to a confounding factor (for example, age or gender). Within each stratum the odds ratio for the outcome of interest is calculated. Each stratum-specific odds ratio is then weighted—according to the number of individuals in that stratum—and the weighted average of these odds ratios is reported as the Mantel-Haenszel estimator. The estimator provides a single summary odds ratio that reflects the combined evidence across strata under the assumption that the exposure–outcome relationship is homogeneous across those strata.
Examples
Section titled “Examples”Smoking and lung cancer
Section titled “Smoking and lung cancer”Researchers may use the Mantel-Haenszel estimator to estimate the odds ratio of developing lung cancer in male and female smokers, as well as in different age groups, then compare these estimates to assess whether certain subgroups, such as younger females, are at higher or lower risk of developing lung cancer.
Medication and blood pressure
Section titled “Medication and blood pressure”Researchers may use the Mantel-Haenszel estimator to estimate the odds ratio of achieving a desired blood pressure level in patients with different preexisting medical conditions, such as diabetes or hypertension, to determine whether the medication is more effective in certain subgroups.
Use cases
Section titled “Use cases”- Comparing the effects of different factors (for example, age or gender) on a binary outcome by controlling for a single confounding variable.
Notes or pitfalls
Section titled “Notes or pitfalls”- Assumes the relationship between exposure and outcome is the same in each subgroup (homogeneity assumption), which may not hold in practice.
- Accounts for only one confounding factor at a time, so it may be inappropriate when multiple confounders must be controlled simultaneously.
Related terms
Section titled “Related terms”- Odds ratio
- Confounding factor
- Stratification