Application of Logistic Regression in Health and Allied Research: Part II – Binary Logistic Regression

Abstract

Background: Binary logistic regression is a widely used multivariable method in health and allied research when the outcome is dichotomous. However, its appropriate application requires careful consideration of variable coding, model specification, assumptions, diagnostics, and interpretation. This article is Part II of a series on logistic regression and provides practical guidance for non-statisticians, clinical professionals, and early-career researchers.

Methods: This methodological tutorial describes the application of unconditional and conditional binary logistic regression using Stata. It covers coding of dichotomous outcomes and categorical predictors, selection of reference categories, model fitting, assessment of multicollinearity and model performance, estimation of predicted probabilities, variable selection, and interaction terms. Interpretation of regression coefficients adjusted odds ratios, 95% confidence intervals, and post-estimation results is illustrated using practical examples.

Results: The tutorial demonstrates how binary logistic regression can be used to estimate adjusted associations and predict dichotomous outcomes. Practical examples illustrate the interpretation of adjusted odds ratios for categorical and continuous predictors, assessment of multicollinearity, evaluation of model calibration and discrimination, estimation of predicted probabilities, and examination of interaction terms. Important methodological considerations, including model diagnostics, overfitting, and variable selection, are highlighted.

Conclusions: Binary logistic regression is a flexible method for analyzing dichotomous outcomes in health and allied research. Appropriate model specification, diagnostics, and interpretation are essential for obtaining valid and meaningful results. This tutorial provides practical guidance for applying and reporting binary logistic regression using Stata.