MERID-EQ: Multimodal Early Risk Identification for Home Healthcare Equity in Infection Risks

Other Titles

A Resource-Conscious Approach Using Small LLMs for Structured Infection Indicator Extraction from Home Healthcare Clinical Notes [Detailed Abstract Title]

Abstract

Annually, three million people in the U.S. receive home healthcare (HHC). Aging-related deterioration, surgical history, and suboptimal home-disinfection practices increase infection risk, which accounts for over 40% of HHC-related hospital or emergency transfers. Critical indicators—such as clinical signs, procedures, and home-hygiene factors—are documented in narrative HHC notes. Converting these unstructured texts into actionable cues requires efficient automation that fits agencies’ limited computing capacity and evolving patient needs. Prior work shows clinicians value context-rich indicators (e.g., “wound deterioration: right thigh with large exudate drainage”) (1). While large language models excel at such tasks, they require extensive resources. Smaller LLMs may offer a resource-conscious alternative, but their effectiveness in domain-specific clinical extraction is unclear. In this study, we compare an instruction-tuned small LLM (LLaMA-3.1-8B) to another small model (Flan-T5-XL, 3B) and a mid-sized model (Qwen-14B) via fine-tuning, and to a large model (LLaMA-3.3-70B) via prompting, for structured infection-indicator extraction under realistic resource constraints.

Authors

Zidu Xu, Columbia UniversityFollow
Jiyoun Song
Shuang Zhou
Danielle Scharp
Mollie Hobensack
Jingjing Shang
Maxim Topaz

Author Details

Zidu Xu, MMed, MPhil¹; Jiyoun Song, PhD, RN²; Shuang Zhou, PhD³; Danielle Scharp, PhD, APRN, FNP-BC⁴; Mollie Hobensack, PhD, RN⁵; Jingjing Shang, PhD, RN¹; Maxim Topaz, PhD, RN¹

1. Columbia University School of Nursing, New York, NY;

2. Department of Biobehavioral Sciences, University of Pennsylvania School of Nursing, Philadelphia, PA;

3. Department of Surgery, University of Minnesota, Minneapolis, MN;

4. Division of General Internal Medicine, Icahn School of Medicine at Mount Sinai, New York, NY;

5. Department of Geriatrics and Palliative Care, Icahn School of Medicine at Mount Sinai, New York, NY

Sigma Membership

Non-member

Type

Report

Format Type

Text-based Document

Study Design/Type

Secondary Analysis

Research Approach

Quantitative Research

Keywords:

Home Healthcare Patients, HHC, Infection Risk, Infection Risk Prediction, Social Risk Factors

Recommended Citation

Xu, Zidu; Song, Jiyoun; Zhou, Shuang; Scharp, Danielle; Hobensack, Mollie; Shang, Jingjing; and Topaz, Maxim, "MERID-EQ: Multimodal Early Risk Identification for Home Healthcare Equity in Infection Risks" (2025). Sigma Foundation for Nursing Research Grant Reports. 186.
https://www.sigmarepository.org/grant_reports/186

Rights Holder

All rights reserved by the author(s) and/or publisher(s) listed in this item record unless relinquished in whole or part by a rights notation or a Creative Commons License present in this item record.

Review Type

None: Sigma Grant Recipient Report

Acquisition

Proxy-submission

Full Text of Presentation

wf_yes