Introduction
The Cochran–Mantel–Haenszel (CMH) Test is a powerful statistical method used in biostatistics to evaluate the association between two categorical variables while controlling for a third variable (confounder). It is widely applied in clinical trials, epidemiological studies, and biomedical research where stratified data analysis is required.
For example, when comparing the effectiveness of a drug across multiple hospitals, differences between hospitals (strata) may influence the results. The CMH test helps adjust for such stratification and provides a pooled estimate of association.
Definition
The Cochran–Mantel–Haenszel test is a statistical test used to determine whether there is an association between an exposure and an outcome after controlling for one or more confounding variables through stratification.
It provides:
- A common odds ratio (CMH odds ratio)
- A chi-square test statistic
- A p-value for significance
Concept Explanation
Why CMH Test is Needed
In real-world data, confounding variables can distort the relationship between exposure and outcome. The CMH test removes this bias by analyzing data within strata (e.g., hospitals, age groups).
Key Components
- Exposure variable: Drug vs Control
- Outcome variable: Recovered vs Not Recovered
- Stratification variable: Hospital (H1, H2, H3)
What CMH Does
- Computes odds ratios for each stratum
- Combines them into a weighted average (common odds ratio)
- Tests whether the overall association is statistically significant
Step-by-Step Procedure
Step 1: Arrange Data into Stratified Tables
Each stratum (hospital) has a 2×2 contingency table:
| Outcome | Control | Drug |
|---|---|---|
| Recovered | a | b |
| NotRecovered | c | d |
Repeat for H1, H2, H3.
Step 2: Calculate Odds Ratio for Each Stratum
Odds Ratio (OR) formula:OR=b×ca×d
Step 3: Compute CMH Common Odds Ratio
The CMH method calculates a weighted average of individual ORs.
Step 4: Perform Chi-Square Test
- Null hypothesis (H₀): No association between exposure and outcome
- Alternative hypothesis (H₁): Association exists
Step 5: Check Homogeneity
Use Breslow-Day test to verify whether odds ratios are consistent across strata.
Example Using Your Data
You provided two analyses:
- Drug as exposed group
- Control as exposed group
Case 1: Drug as Exposed Group
Key Results
- CMH Odds Ratio = 0.5077
- 95% CI = 0.09969 to 2.5856
- Chi-square = 0.70352
- p-value = 0.40160
Case 2: Control as Exposed Group
Key Results
- CMH Odds Ratio = 1.9697
- 95% CI = 0.3868 to 10.0316
- Chi-square = 0.70352
- p-value = 0.40160
Result Interpretation
1. Odds Ratio Interpretation
- OR < 1 → Drug reduces risk
- OR > 1 → Drug increases risk
- OR = 1 → No effect
👉 In your case:
- OR = 0.5077 suggests drug may reduce risk, but not significantly
2. Confidence Interval
- If CI includes 1 → Not significant
👉 Your CI includes 1 → No statistically significant association
3. P-value
- p < 0.05 → Significant
- p > 0.05 → Not significant
👉 p = 0.40160 → Not significant
4. Homogeneity Test
| Test | p-value |
|---|---|
| Breslow-Day | 0.05026 |
| Breslow-Day-Tarone | 0.05208 |
👉 Interpretation:
- p ≈ 0.05 → borderline heterogeneity
- Suggests slight variation across hospitals
Result Table Format
Stratified Odds Ratios
| Hospital | Odds Ratio | 95% CI |
|---|---|---|
| H1 | 4.0000 | 0.1342–119.2371 |
| H2 | 1.0000 | 0.03355–29.8093 |
| H3 | 0.02041 | 0.0003089–1.3483 |
CMH Summary Table
| Parameter | Value |
|---|---|
| CMH Odds Ratio | 0.5077 |
| 95% Confidence Interval | 0.09969–2.5856 |
| Chi-square | 0.70352 |
| Degrees of Freedom | 1 |
| p-value | 0.40160 |
Homogeneity Test Table
| Test | Chi-square | DF | p-value |
|---|---|---|---|
| Breslow-Day | 5.9812 | 2 | 0.05026 |
| Breslow-Day-Tarone | 5.9101 | 2 | 0.05208 |
📥 Download Dataset
You can download the dataset used in this analysis here:
👉 Download CMH Test Dataset (Excel file)
Practical Applications
The CMH test is widely used in:
- Clinical trials
- Epidemiology
- Public health studies
- Case-control studies
- Drug effectiveness comparison
Advantages
✔ Controls confounding variables
✔ Provides pooled estimate
✔ Suitable for stratified categorical data
✔ Easy to interpret
Limitations
✖ Assumes homogeneity across strata
✖ Not suitable for continuous variables
✖ Sensitive to small sample sizes
Conclusion
The Cochran–Mantel–Haenszel test is an essential statistical tool for analyzing stratified categorical data while controlling for confounding variables. In your example, although the drug appears to reduce the odds of the outcome, the results are not statistically significant, indicating insufficient evidence to conclude a true association.
The homogeneity test suggests slight variation across hospitals, which should be considered when interpreting results.
Overall, CMH provides a reliable and robust method for analyzing multi-stratum data, especially in medical and biostatistical research.
