Reducing Emergency Department Readmissions:

~ How Community Health Workers and Social Determinants of Health Make a Difference ~

This study explores the role of Community Health Workers (CHWs) and Social Determinants of Health (SDoH) in reducing 30-day unplanned Emergency Department (ED) readmissions at Sinai Chicago. The research leverages machine learning to assess the impact of integrating CHWs and SDoH data in predictive models. By comparing patient readmission predictions with and without this data, the study demonstrates that when CHWs engage with patients, predictive accuracy improves by 13.0-15.2%. The findings highlight the significance of CHW-related features in classification, reinforcing the value of these programs. The study optimizes predictive models for engaged patients, evaluates various machine learning metrics, and provides recommendations for improving patient care. Ultimately, this work emphasizes the human connection between CHWs and patients as a key factor in reducing avoidable hospital readmissions.

Introduction

Emergency department (ED) readmissions within 30 days remain a persistent challenge for healthcare providers. Not only do frequent readmissions strain hospital resources, but they also signal deeper, unresolved issues in patient care and social support. Many readmissions are linked to Social Determinants of Health (SDoH), such as food security, housing stability, and access to healthcare services. As of 2024, hospitals are required by the Centers for Medicare and Medicaid Services (CMS) to screen patients for key SDoH domains and document these needs in medical records. The goal is to create better interventions and improve patient outcomes.

In our previous study, we demonstrated the positive impact of CHWs and SDoH data on reducing ED readmissions among high-risk patients. Machine learning models were employed to predict the likelihood of readmissions, and it was shown that models incorporating CHW engagement nd SDoH data significantly outperformed models relying solely on demographics and referral data. Specifically, patients who engaged with CHWs had a lower readmission rate of 18.89%, compared to 35.03% for non-engaged patients. However, the scope of the analysis was limited by the relatively small dataset of 1,634 patient records. In this extended study, we started by integrating additional patient data, resulting in a dataset of 2,173 patient records, which increases the robustness of our analysis.

In addition to confirming our previous results, we aimed to perform a comprehensive feature importance analysis, and translate it into useful actionable recommendations for the CHW program at Sinai Health Systems. This extended research starts with observed statistics and employs both supervised and unsupervised machine learning techniques to derive deeper insights into patient characteristics and factors influencing readmissions. Specifically, Random Forest (RF) and Logistic Regression (LR) were used for classification, and K-modes clustering was used for unsupervised analysis. The primary objective of this paper is to evaluate the predictive power of CHW engagement and SDoH data in reducing ED readmissions and to provide CHWs with actionable recommendations based on feature importance. We hypothesize that CHWs, supported by data-driven insights, can significantly lower the risks of ED readmissions for high-risk patients by providing resources that address SDoH data. This research contributes to the growing body of evidence supporting community-based interventions in healthcare and highlights the potential of machine learning to enhance these interventions.

This study takes a deep dive into how Community Health Workers (CHWs) and SDoH data impact ED readmission rates. Using machine learning techniques, we examine how CHWs' engagement with patients contributes to better health outcomes, and we provide actionable recommendations for optimizing their role.

This blog is based on the research published in the International Journal of Semantic Computing:

Citation: Hernandez, N., Karam, K., Baugh, N., Musale, S., Moses, A. P., Raicu, D., Furst, J., McCabe, K., & Tchoua, R. (2023).

"Leveraging Community Health Workers for Predicting Emergency Department Readmissions."

International Journal of Semantic Computing. World Scientific Publishing Company.