Background: Readmission following an index admission for heart failure (HF) remains the leading cause of hospital readmission in the U.S. Risk prediction tools at the time of hospital discharge (d/c) provide better allocation of resources towards higher-risk populations. We use modeling and additional patient data to improve risk assessment in HF patients at the time of d/c.
Methods: We collected data at d/c from 1,319 consecutive patients admitted with a primary diagnosis of HF to Ochsner Medical Center. Patients were later labeled as readmitted if they were hospitalized within 30 days of d/c. 41 patients who died were excluded from the analysis. To create a readmission model, we fit a logistic ridge regression model to the data and used doubly nested cross validation to tune the model’s regularization parameter and estimate model performance on unseen data. We compared our model with the CORE Readmission Risk (CRR) using area under the ROC curve (AUC).
Results: Patients had mean age 69 ± 15, were 44% female and 25% were readmitted. Our model achieved an AUC of 0.654 (95% CI [0.632–0.675]), significantly better than the CRR with an AUC of 0.599 (95% CI [0.578–0.619]). The model identified variables having a significant association with readmission rates even when controlling for CRR.
Conclusions: Additional data collected at hospital d/c that is not specific to HF provide significant improvement in readmission prediction for patients discharged with HF.