View Abstract

A4980 - An Adaptive Predictive ""APP"" Utilizing an Artificial Neural Network (ANN) for Early Identification of Hospital Readmission Risks in COPD and Pneumonia Patients
Author Block: A. Roberts1, L. Wang2, R. Ogunti3, M. Puppala2, S. Chen2, T. He2, X. Yu2, A. Nezamabadi4, S. Wong2, A. Frost5, R. E. Jackson4; 1Ob-Gyn, Houston Methodist Hospital, Houston, TX, United States, 2Systems Medicine and Bioengineering, Houston Methodist Hospital Research Institute, Houston, TX, United States, 3Systems Medicine and Bioengineering, Houston Methodist Hospital, Houston, TX, United States, 4Medicine, Houston Methodist Hospital, Houston, TX, United States, 5Smith Tower, Suite 1001, Houston Methodist Lung Center, Houston, TX, United States.
Introduction: Rising healthcare costs and the mandate to deliver quality care have motivated regulatory agencies to tie reimbursement to the rate of hospital readmissions stimulating efforts to identify those at risk for readmission. Particular focus has been paid to the primary diagnoses of COPD and Pneumonia -frequent causes of 30-day re-admissions. Development and validation of a simple tool deploying an artificial neural network (ANN) for early identification of patients with high readmission risk was the objective of this study. Material and Methods: Multiple potential factors predictive of readmission were evaluated. Using the electronic medical record (EMR) feeding directly to the ANN from 6 hospitals within our hospital system, four variables were identified and validated. The predictive algorithm was then implemented in a smartphone app called Re-Admit, using a proprietary risk assessment model that incorporates: the number of inpatient visits within the past 6 months, the number of unique medications started on hospital day one, Insurance status, and the Rothman Index, all parameters obtainable on day one of admission. The Rothman Index is an in-hospital mortality predictive index never before utilized in predicting readmission risk. Data from 1075 patients with pneumonia from Jan 2010 to Nov 2015 and 332 patients with COPD from Jan 2013 to June 2014 were used retrospectively to train the application's risk prediction algorithm. The predictive results of the Re-Admit app were then successfully validated prospectively in 229 pneumonia patients admitted from December 2015 to May 2016, and 172 COPD patients admitted after July 2015. Results: The receiver operating characteristics (ROC) of the pneumonia validation data was 0.70 for logistic regression and 0.73 for the ANN; the COPD validation ROC was 0.70 for logistic regression and 0.76 for the ANN. The ANN demonstrated better prediction power by ROC in both patient groups. Conclusion: By identifying patients vulnerable to readmission early (day 1 of index admission) inpatient efforts to mobilize outpatient support can be started early. This Re-Admit App is a simple automated tool using mobile technology and the existing day 1 EMR for data input. It also uniquely utilizes the Rothman Index as a readmission risk predictor. In our patient population this Readmit App appears to simply and reliably predict readmission risk and allows the hospital to focus on organizing community resources in high risk patients to prevent readmission.