Using large language model (LLM) to identify high-burden informal caregivers in long-term care

Area of Science:

• Artificial Intelligence in Healthcare
• Natural Language Processing
• Gerontology

Background:

• Increasing global elderly population necessitates improved caregiving assessment.
• Traditional methods may not fully capture informal caregiver challenges.
• Informal caregivers are crucial in Long-term Care (LTC).

Purpose of the Study:

• To evaluate Large Language Models (LLMs) for detecting overburdened informal caregivers.
• To compare LLM efficacy against rule-based and machine learning approaches.
• To explore LLM interpretability using attention mechanism visualization.

Main Methods:

• Utilized textual case summary reports from 1,791 informal caregivers in Southern Taiwan.
• Employed structured questionnaire data for traditional machine learning models.
• Leveraged LLM attention mechanism visualization for interpretative analysis.

Main Results:

• LLM achieved an AUROC of 0.84 and AUPRC of 0.70.
• Demonstrated an 8% and 14% improvement over traditional methods.
• Attention visualization aligned with expert evaluations, focusing on high-burden indicators.

Conclusions:

• LLMs show significant capability in identifying high-burden informal caregivers in LTC.
• LLM offers a promising avenue for future LTC research and policy development.
• AI-driven insights can enhance support systems for informal caregivers.