Acute Myocardial Infarction Readmission Risk Prediction Model in Admit and Discharge patients

Résumé fourni par la source

The Hospital Readmission Reduction Program (HRRP) of the Center for Medicare and Medicaid Services (CMS) penalizes acute-care hospitals whose 30-day Acute Myocardial Infarction (AMI) readmission rates are high relative to other facilities. The main aim of this program is to reduce readmissions, thereby steering the hospitals towards value-based care. An AMI readmission risk prediction model can prove beneficial for health-care providers in this scenario aiding reduction in readmissions by enabling targeted interventions. A plethora of work has been done in developing predictive models for readmissions after visits that had AMI principal diagnosis. However, most readmission models were developed to be used after the discharge of the patient and later require follow-up with patients predicted to be readmitted. Further, they do not address the challenges towards real-time integration of the model in clinical setting. We present in this paper implementation of the readmission model which provides predictions during inpatient or observation visits for AMI, so as to aid early targeted interventions. Model also outputs the clinically important features. Cohorts of the population data of different Healthcare Systems with an elevated troponin and a working diagnosis of AMI were used for the study to mimic the real clinical scenario where principal diagnosis is usually not yet unavailable. In this study, we describe and compare the implementation of Generalized Linear Model (GLM) with industry standard methods such as LACE score and Hospital score. The GLM model showed good performance with Area Under the Curve (AUC) of 70.0% at 12-hour after admission and AUC of 70.4% before discharge. The AMI-specific model performance was better than the LACE score and Hospital score by 5-6% and better than all cause readmission prediction model by 2-4% across different healthcare systems.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.