Maximizing Exploratory Predictive Models

One of the more common submissions to the Journal of Cardiac Failure describes a retrospective analysis of secondary data (i.e., data collected for a previous study for a different purpose) that examines associations between a novel clinical or laboratory parameter and clinical outcomes (e.g., hospital readmission, death) using either logistic regression (static event) or Cox proportional hazards regression (time-to-event). At their best, these analyses can be effective in identifying potentially important, but heretofore unconsidered, prognostic factors regarding morbidity/mortality. The use of existing data to explore new associations, estimate effect sizes, generate promising hypotheses, and create a foundation for future studies is fundamental to the inquiry process. That being said, there are a number of issues that require consideration, at least to some extent, to maximize the contribution of papers that report these analyses.