An introduction to statistical methods used in binary outcome modeling.

Logistic regression is a cornerstone of epidemiology and the method of choice for risk adjustment models in cardiac surgery and critical care. Although linear regression methods may be satisfactory to evaluate relationships between independent (predictor) variables and a outcome that is continuous, a more complex mathematical approach is required when the outcome is binary (yes/no; alive/dead). Although the odds are 4 to 1 that finding an antilogarithm may sound intimidating, once you get past the initial equations and terminology, we go on to discuss how to select variables for a model, how to deal with collinearity and interaction terms, how to use diagnostic tests to ensure the model is not adversely affected by a small number of observations, and how to assess a model's discrimination and calibration. A full understanding of how logistic models are developed is useful when assessing the medical literature on risk assessment.