Abstract 21098: Variables From the CMS Heart Failure Readmission Model Poorly Predict 30-Day Rehospitalization Risk in Heart Failure Patients From a Large Academic Hospital System

Abstract

Introduction: Predicting 30 day readmission risk in patients with acute heart failure (HF) remains challenging. As part of the Hospital Readmissions Reduction Program, the Centers for Medicare and Medicaid Services (CMS) has implemented a claims-based 30 day readmission risk model to stratify US hospitals for its HF penalty algorithm, with a reported c-statistic of 0.61. The variables used in this model may be useful in hospitals for predicting readmission risk and informing population health efforts. Methods: We evaluated the discriminatory power of variables from the CMS HF readmission risk model to predict 30 day readmission in patients part of the Cleveland Clinic Heart Failure Registry (n=29,749 hospitalizations), which contains data from all admissions from 2010 to present with a primary diagnosis of HF to all Cleveland Clinic Health System hospitals. The CMS model contains 37 variables including history of HF, ACS, COPD, renal failure (RF) and drug/alcohol abuse. Results: Twenty one percent of our cohort experienced all cause 30 day readmission (n=5109). The median age was 74 (IQR 63-84) and 48% were female. Based on billing codes from the year prior to a HF admission, there was no significant difference in the rates of ACS, RF, active cancer, CVA, dementia and COPD between patients who did and did not experience a 30 day readmission (p>0.05). Using 2010-13 admissions (n=19,904) and 37 variables from the CMS HF readmission model, we derived a 30 day all cause readmission predictive model. We validated this model using 2014-15 admissions (n=9,845). This model had poor discriminatory power in the derivation and validation cohorts (c-statistic=0.51 for both). In a parallel analysis, a bootstrapped model using the entire cohort was created, with a c-statistic of 0.53. The discriminatory power did not improve when the cohort was limited to patients ≥ 65 years (validation c-statistic=0.51). Conclusion: A prediction model derived using claims-based variables from the CMS HF readmission model poorly predicts 30 day all cause readmission in a large hospital system-based cohort, with a validated c-statistic of 0.51. The applicability of CMS’ readmission risk stratification approach to clinical care and local population management efforts in HF is likely limited.