Predictive Modeling of Emergency Hospital Transport Using Medical Alert Pattern Data: Retrospective Cohort Study

• DOI: 10.2196/iproc.4772
• Publication Date: 2015-11-25T05:02:21Z
• Authors: Jorn op den Buijs; Tine Smits; Marten Pijl; Mariana Simons; Linda Schertzer
• Keywords: 03 medical and health sciences; 0302 clinical medicine; 3. Good health

Background: In the transition from a fee-for-service to fee-for-value system, health care organizations (HCOs) are under pressure keep patients healthy through preventive services and population management. Predictive analytics based on past behavior of patient can be used predict future risk decline. Objective: The objective this study was develop robust predictive models impending emergency transports hospital enrollment medical alert pattern data subscribers Personal Emergency Response System (PERS) service. This enables targeting clinical programs members that need it most. Methods: De-identified 551,127 PERS service were used. Multivariate logistic regression performed subscriber demographics, self-reported conditions, variables related giver network derived up one year retrospective data. A 10-fold cross-validation scheme transport by in next 30 days. Furthermore, model performance evaluated after retraining using 90 days data, only. Results: 30-day window experienced 2.4% all subscribers. area receiver operator characteristic curve (auROC) 0.75 ± 0.01 validation cohorts. resulted auROC = 0.71 only 0.62 0.01. Conclusions: Our for showed good discriminatory accuracy While yields with best is achieved only, without does not perform as well. We planning prospective algorithm determine value assisting HCOs early interventions avoid department visits hospitalizations. [iproc 2015;1(1):e19]