Data-Driven Decision Support for Post-Hospitalization Diabetes Case Management Referral: Protocol for Predictive Analysis

Abstract

Abstract Background While diabetes cases become more complex with increasing age and comorbidity, social determinants of health (SDoH), including food security, medication availability, and transportation, act as a significant source of disparities in diabetes risk and outcomes. However, with the existing supply-demand mismatch in diabetes case management, current case management referrals are primarily based on the most apparent clinical information. Data-driven decision support that learns from large-scale electronic health records (EHRs) encompassing the SDoH is a promising approach to helping prioritize demand and alleviate disparities through the identification of patients at highest risk. Methods This protocol is for a predictive analysis study to develop a proactive risk assessment decision support (PRADS) model incorporating the SDoH data to stratify urgency of needed case management among diabetic patients by identifying patients that are likely to utilize extensive resources including hospitalizations and emergency department (ED) visits. We will collect EHR data spanning from January 2018 to February 2023 from a Level 1 Trauma Center in Southeast (where diabetes and disparities have been prevalent), including demographics, SDoH, comorbidities, laboratory test results, access to care, medications, and the outcome variables (i.e., readmissions and ED visits). The dataset includes 65,821 unique inpatients that account for 132,273 inpatient encounters. 70% of the data observations will be utilized to train a Survival Decision Tree model, and the rest to generate prediction outcomes (probability of readmission or ED visit). The model will be validated in a quantitative manner by evaluating performance evaluation metrics and in a qualitative manner by conducting a three-phase modified Delphi method that involves the case management, clinician, and quality and patient safety teams. Discussion Despite the multifaceted disparities among diabetic patients, e.g., comorbidities and social support, the current case management referral process does not involve a comprehensive assessment of patient information. Data-driven decision support that learns from a wide range of SDoH and clinical information can help better identify high-risk patients for effective utilization of the current workforce. The PRADS model may demonstrate the value of SDoH in estimating diabetes risk factors, potential for system implementation, and improvement in utilizing the case management workforce.