Integrating Sensor Data with Large Language Models for Enhanced Elderly Care: A Methodological Framework

The global aged population is expected to exceed 2.1 billion, representing 21.65% of the total population by 2050. This demographic shift underscores an urgent need for efficient elderly care, particularly in home settings. AI advancements have made sensor technology, including wearable biosensors, environmental monitors, and biochemical sensors, essential for elderly care by enabling the collection of physiological and activity data. Current systems overwhelm caregivers with complex data analysis and personalized recommendations. Large language models (LLMs) address this by offering insights through natural language interfaces, using extensive medical data. While some studies have integrated sensor data with LLMs for health monitoring applications, a comprehensive framework for seamlessly combining diverse sensor data with LLMs in elderly care is still missing. In this study, we propose a novel methodological framework that addresses the challenges of integrating heterogeneous sensor data with LLMs to provide real-time healthcare insights for caregivers of the elderly using sensor technologies. Our framework employs few-shot learning on Generative Pre-trained Transformer (GPT-4) and GPT-3.5 to process structured sensor data from wearable and environmental devices. The LLM-powered application then generates insightful responses based on the user’s input, providing actionable and personalized recommendations. The GPT-4 model outperformed GPT-3.5 in Structured Query Language (SQL) query generation for sensor data retrieval and processing, achieving a semantic similarity score of 0.95, precision of 88.5%, recall of 98.92%, and an F1-score of 93.40%. In this study, we explore how integrating sensor data with LLMs enhances usability and reduces complexity in health monitoring systems. Our framework sets a new benchmark for advancing elderly care through innovative LLM-powered applications and sensor technology.