Emergency Room Frequent Users: Applying Behavioral and Predictive Analytic Frameworks to Create Operationally Effective Predictive Models

Document Type

Presentation

Loading...

Media is loading
 

Publication Date

5-8-2023

Comments

Presentation: 50:20

Abstract

In recent years, rising emergency department (ED) demand and crowding issues have adversely affected US healthcare cost and outcomes. The ED is not an optimal setting to support the continuity of care required for many patients: care coordination is best managed by primary care and specialty settings. ED frequent users represent a sizable proportion of all ED visits. Frequent users are a more clinically & socially complex patient population compared to non-frequent users, with multiple comorbid conditions, higher rates of hospital admissions and mortality, and complex social needs. To improve health outcomes and reduce the cost of care, frequent ED users require nuanced interventions that improve continuity of care outside of the ED setting. Contending with over-utilization, excess capacity, crowding, and the increasing pressure of cost containment, EDs need strategies and tools that can ease this significant burden, collaboratively support cost-effective decision-making, and respond to patient needs quickly and effectively.

Current ED frequent utilization research lacks: 1) a comprehensive and commonly referenced listing of predictor factors associated to repeated ED use; 2) study of outcome measures such as outcome measures of diagnostic or dispersion utilization patterns; 3) predictive models derived from broader patient populations at-risk; and 4) care management strategies informed by predictive models that accommodate larger patient populations at-risk and inform the use of additional less-intensive intervention strategies.

This study was conducted as a retrospective, multi-hospital cross-sectional methodological study using routinely collected EMR and publicly available data to develop four predictive models. Study participants were adult patients, 18 years or older, discharged from Main Line Health’s four Emergency Departments (Bryn Mawr Hospital, Lankenau Medical Center, Paoli Hospital, Riddle Hospital) from calendar year 2022. A one-year look back period from the last discharge date of each ED patient was included in the study to determine ED visit volume and frequent use categories. Predictor and outcome variables were selected using a categorization method informed by Andersen’s Behavioral Model of Health Services Use (BMHSU). Four predictive models were derived and tested for consistent and acceptable sensitivity and specificity results. The first three domains of the Electronic Health Predictive Analytics (e-HPA) framework, Data Barriers, Transparency, and Ethics & Privacy were addressed in this study through: 1) successful extraction and statistical analysis of EMR and publicly available data; 2) development of a Tableau interactive report with key metric, trends, and interactive patient exploration tools to increase stakeholder engagement; and 3) development of a design plan and proposed timeline for the integration of the four predictive models into Main Line Health’s care management workflow.

The extensive and innovative scope of this study contributes significantly to current research in ED frequent utilization. The use of BMHSU in the context of predictive analytics to define health care utilization, or in the context of this study, ED utilization, in a more holistic and broad manner is novel. In addition, the utility of frameworks to govern and manage predictive analytic work has not been widely done. The e-HPA framework is new and while the framework itself has been published in a few research articles, the application of e-HPA in predictive research has not been conducted. This study is the first to attempt use of the framework in adapting predictive analytics in care management settings.

Language

English

This document is currently not available here.

Share

COinS