Machine Learning-Driven Decision Support for Post-Hospitalization Diabetes Case Management Referral: A Sequential Mixed Methods Study Protocol (Preprint)

Abstract

Diabetes case management provides surveillance of symptoms and care coordination that benefits from considering the patient’s age, comorbidities, and social determinants of health (SDoH). Research finds that SDoH are important to the complexity of diabetes care. However, current referral practices, based mainly on clinical records, lead to unmet diabetes case management needs. While decision support systems have been developed to address the disparities, their effective application is hindered by healthcare professionals' limited understanding of these models' performance and their clinical and operational relevance.

This study proposes the development of a data-driven decision support system that incorporates SDoH to prioritize care and employs a mixed-methods evaluation approach to mitigate disparities in diabetes case management services within a healthcare system.

The proactive risk assessment decision support (PRADS) model for a clinical population with diabetes will use both SDoH and clinical data to prioritize the patient’s urgency of case management need, identifying those most likely to need high-cost healthcare resources, such as the emergency department (ED). It will be developed using data on demographics, SDoH (e.g., food access, transportation, medication availability), comorbidities, hospitalization-related factors, laboratory test results, medications, and outcome variable (i.e., ED visits). We will employ a mixed-methods evaluation approach, combining quantitative validation of the model's performance with qualitative insights from case managers, clinicians, and quality and patient safety experts, employing a modified Delphi method and a semi-structured focus group.

As of December 2023, we gathered data on 174,871 inpatient encounters from January 2018 to September 2023, involving 89,355 unique inpatients meeting our inclusion criteria. All clinical and SDoH data items for these patients and their encounters were fully collected as of December 2023.

The current case management referral process for diabetic patients lacks a comprehensive assessment of patient information, leading to disparities in care. By integrating a critical suite of SDoH with clinical data, a tailored data-driven decision support system like PRADS can more effectively identify patients at elevated risk to use services. By aligning the model with the hospital's specific quality and patient safety considerations through a mixed-methods approach, we aim to enhance the quality of patient care and optimize case management resource allocation.