Development and validation of a predictive model for all-cause hospital readmissions in Winnipeg,Canada

Abstract

Objective

A number of predictive models have been developed to identify patients at risk of hospital readmission. Most of these have focused on readmission within 30 days of discharge. We used population-based health administrative data to develop a predictive model for hospital readmission within 12 months of discharge in Winnipeg, Canada.

Methods

This was a retrospective cohort study with derivation and validation data sets. Multivariable logistic regression analyses were performed and factors significantly associated with readmission were selected to construct a risk scoring tool.

Results

Several variables were identified that predicted readmission (i.e. older age, male, at least one hospital admission in the previous two years, an emergent (index) hospital admission, Charlson comorbidity score >0 and length of stay). Discrimination power was acceptable (C statistic =0.701). At a median risk score threshold, the sensitivity, specificity, positive and negative predictive values were 45.5%, 79%, 68.8% and 58.6%.

Conclusions

This predictive model demonstrated that hospital readmission within 12 months of discharge can be reasonably well predicted based on administrative data. It will help health care providers target interventions to prevent unnecessary hospital readmissions.