Hierarchical logistic regression models for clustered binary outcomes in studies of IVF-ET

Objective: To describe a hierarchical logistic regression model for clustered binary data, apply it to data from a study on the effect of hydrosalpinx on embryo implantation, and compare the results with analyses that do not account for clustering. Design: Observational study. Setting: Academic research environment. Patient(s): Women undergoing IVF-ET for tubal disease. Main Outcome Measure(s): Odds of per embryo implantation. Result(s): Although regression estimates are largely similar between the models, the hierarchical model properly reflects the added variation due to clustering. Standard errors are higher, confidence intervals are wider, and P values indicate fewer “statistically significant” effects. Conclusion(s): Ignoring important sources of variation in any analysis can lead to incorrect confidence intervals and P values. In studies of IVF-ET, where clustered data are common, unexplained heterogeneity can be substantial. In this setting, hierarchical logistic regression is an appropriate alternative to standard logistic regression.