Readmission after hospitalization for heart failure among patients with chronic kidney disease: a prediction model.

AIMS

30-day readmission rates after hospitalization for heart failure (HF) approach 25%, and patients with chronic kidney disease (CKD) are disproportionately represented. A retrospective cohort study was conducted to develop a prediction tool for 30-day readmission after hospitalization for HF among those with non-dialysis dependent CKD.

METHODS

Geisinger primary care patients with Stage 3 - 5 CKD hospitalized with a primary discharge diagnosis of HF during the period July 1, 2004 through February 28, 2010 were eligible. Multivariate logistic regression was employed to build models from predictors of 30-day readmission, drawn from demographic, clinical, laboratory, and pharmaceutical variables in the electronic health record. Variables were manually removed to achieve a model with satisfactory goodness-of-fit and parsimony while maximizing area under the receiver operating characteristic curve (AUC). Internal validation was performed using the bootstrap resampling method (1,000 samples) to provide a bias-corrected AUC.

RESULTS

607 patients with CKD were admitted for HF during the study period; 116 (19.1%) were readmitted within 30 days. A model incorporating 23 variables across domains of medical history, active outpatient pharmaceuticals, vital signs, laboratory tests, and recent inpatient and outpatient resource utilization yielded an AUC (95% CI) of 0.792 (0.746 - 0.838). The bias-corrected AUC was 0.743. At an estimated readmission probability of 20%, the model correctly classified readmission status for 73% of the population, with a sensitivity of 69% and a specificity of 73%.

CONCLUSION

A robust electronic health record may facilitate the identification of CKD patients at risk for readmission after hospitalization for HF.