Risk assessment of post-discharge mortality among recently hospitalized Medicare heart failure patients with reduced or preserved ejection fraction

Abstract

Objective: Targeted care management for hospitalized patients with acute decompensated heart failure (ADHF) with reduced or preserved ejection fraction (HFrEF/HFpEF) who are at higher risk for post-discharge mortality may mitigate this outcome. However, identification of the most appropriate population for intervention has been challenging. This study developed predictive models to assess risk of 30 day and 1 year post-discharge all-cause mortality among Medicare patients with HFrEF or HFpEF recently hospitalized with ADHF.

Methods: A retrospective study was conducted using the 100% Centers for Medicare Services fee-for-service sample with complementary Part D files. Eligible patients had an ADHF-related hospitalization and ICD-9-CM diagnosis code for systolic or diastolic heart failure between 1 January 2010 and 31 December 2014. Data partitioned into training (60%), validation (20%) and test sets (20%) were used to evaluate the three model approaches: classification and regression tree, full logistic regression, and stepwise logistic regression. Performance across models was assessed by comparing the receiver operating characteristic (ROC), cumulative lift, misclassification rate, the number of input variables and the order of selection/variable importance.

Results: In the HFrEF (N = 83,000) and HFpEF (N = 123,644) cohorts, 30 day all-cause mortality rates were 6.6% and 5.5%, respectively, and 1 year all-cause mortality rates were 33.6% and 29.5%. The stepwise logistic regression models performed best across both cohorts, having good discrimination (test set ROC of 0.75 for both 30 day mortality models and 0.74 for both 1 year mortality models) and the lowest number of input variables (18–34 variables).

Conclusions: Post-discharge mortality risk models for recently hospitalized Medicare patients with HFrEF or HFpEF were developed and found to have good predictive ability with ROCs of greater than or equal to 0.74 and a reasonable number of input variables. Applying this risk model may help providers and health systems identify hospitalized Medicare patients with HFrEF or HFpEF who may benefit from more targeted care management.

Keywords:

• Heart failure
• ejection fraction
• risk
• mortality
• hospitalization
• machine learning

Transparency

Declaration of funding

This analysis was funded by Novartis Pharmaceutical Corporation.

Declaration of financial/other relationships

M.S. has disclosed that he is an advisor and consultant to Novartis Pharmaceutical Corporation. H.S.F. and P.N. have disclosed that they are employees of DataMed Solutions LLC, which provides contracted services to Novartis Pharmaceutical Corporation. P.R. and E.N.O. have disclosed that they were employees of Novartis Pharmaceutical Corporation at the time of the study. S.P. has disclosed that she is an employee of University of Maryland, Baltimore, providing services to Novartis Pharmaceutical Corporation. CMRO peer reviewers on this manuscript have no relevant financial or other relationships to disclose.

Author contributions

All authors contributed to the conception and design of the analysis. H.S.F. performed the statistical modeling. All authors interpreted the data, critically revised the manuscript for intellectual content, approved the final manuscript for submission and agree to be accountable for all aspects of the work.

Acknowledgements

This study was funded by Novartis Pharmaceutical Corporation. Medical writing and editorial assistance were provided by Erin P. Scott PhD of Scott Medical Communications LLC. This assistance was funded by Novartis Pharmaceutical Corporation.

Previous presentation

Portions of this work were presented at the American College of Cardiology Annual Meeting 2019, New Orleans, LA.

Data availability

The data that support the findings of this study are available from Centers for Medicare and Medicaid Services. Restrictions apply to the availability of these data, which were used under license for this study. Data are available from the authors with the permission of Centers for Medicare and Medicaid Services.