Survival analysis
Descriptive statistics gave some indications of predictors for unscheduled readmissions. The patterns of patient readmissions, however, were quite complex and often involved multiple admissions and overlapping 30-day readmission periods. Logistic regression analysis of the retrospective data was attempted, but it oversimplified the data. The results were not included in this paper for that reason. To better estimate the risk of readmission given several explanatory variables, and to account for multiple readmissions per patient across the 1-year time frame, survival modeling was more successful. Table 3 summarizes the survival model variables or interaction terms found to be significant in at least one model.