Logistic regression modeling: methodological insights and roadmap

Logistic regression is commonly utilized in clinical and educational research to examine relationships between risk factors and binary outcomes. However, pharmacy researchers may encounter challenges in selecting appropriate predictors, verifying model assumptions, interpreting results, and reporting findings transparently.

This methodology review presents a structured roadmap for conducting logistic regression, covering key steps such as defining the binary outcome, selecting and coding predictors, checking assumptions, fitting the model, and evaluating model diagnostics.

To illustrate the roadmap in practice, we draw on two published studies: the OMICU study, which evaluated opioid use and prescribing outcomes in critically ill patients, and Spivey et al., which identified predictors of academic outcomes in pharmacy students. Additionally, a detailed how-to example using a simulated pharmacy education dataset further demonstrates model construction and interpretation, accompanied by STATA code to support reproducibility. The manuscript also includes a comparison of common software platforms, including STATA, R, and SAS, highlighting their relevance, functionality, and usability in the context of logistic regression.

The manuscript highlights best practices in covariate selection, exploratory data analysis, and model development using advanced techniques such as stepwise and LASSO regression. Guidance is also provided on the interpretation of odds ratios and confidence intervals, handling of sparse events and continuous variables, model performance evaluation, and transparent reporting.