Abstract:Despite the proliferation of wearable health trackers and the importance of sleep and exercise to health, deriving actionable personalized insights from wearable data remains a challenge because doing so requires non-trivial open-ended analysis of these data. The recent rise of large language model (LLM) agents, which can use tools to reason about and interact with the world, presents a promising opportunity to enable such personalized analysis at scale. Yet, the application of LLM agents in analyzing personal health is still largely untapped. In this paper, we introduce the Personal Health Insights Agent (PHIA), an agent system that leverages state-of-the-art code generation and information retrieval tools to analyze and interpret behavioral health data from wearables. We curate two benchmark question-answering datasets of over 4000 health insights questions. Based on 650 hours of human and expert evaluation we find that PHIA can accurately address over 84% of factual numerical questions and more than 83% of crowd-sourced open-ended questions. This work has implications for advancing behavioral health across the population, potentially enabling individuals to interpret their own wearable data, and paving the way for a new era of accessible, personalized wellness regimens that are informed by data-driven insights.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to use large language model (LLM) agents to extract personalized health insights from wearable device data. Despite the high popularity of wearable health trackers and the fact that sleep and exercise are crucial for health, it is still challenging to extract actionable personalized insights from these data because it requires complex open - ended analysis of these data. Currently, most wearable devices only provide general summaries of health behaviors, such as daily step counts or estimated sleep quality, and are unable to provide interactive health insights tailored to individual needs and interests. The paper proposes a new system - the Personal Health Insight Agent (PHIA), which solves this problem by means of advanced code generation, web - search integration, and the ReAct agent framework to conduct in - depth analysis and respond to thousands of real - world health queries. Specifically, the main contributions of the paper include: - Introducing the first LLM - based health agent framework, which utilizes multi - step iterative reasoning, code generation, and web - search tools to conduct in - depth analysis of thousands of health insight questions. - Conducting 650 hours of human evaluation, involving more than 6,000 model responses, and performing automatic evaluation of 12,000 model responses to demonstrate the superior ability of LLM agents in time - series behavioral health data analysis and in - depth health insight interpretation. Compared with non - agent LLM baseline methods, these agents perform better in code generation and text - number reasoning. - Releasing a set of high - fidelity synthetic wearable data, which are sampled from a large amount of anonymous production data. - Releasing a personal health insight evaluation dataset, which contains more than 4,000 closed and open - ended questions covering multiple domains and is suitable for automated and manual evaluation. Through these contributions, the paper aims to promote progress in the field of behavioral health, enabling individuals to interpret their own wearable device data and paving the way for data - driven personalized health solutions.

Transforming Wearable Data into Health Insights using Large Language Model Agents

From Classification to Clinical Insights: Towards Analyzing and Reasoning About Mobile and Behavioral Health Data With Large Language Models

PhysioLLM: Supporting Personalized Health Insights with Wearables and Large Language Models

Towards a Personal Health Large Language Model

Large Language Models for Wearable Sensor-Based Human Activity Recognition, Health Monitoring, and Behavioral Modeling: A Survey of Early Trends, Datasets, and Challenges

Health-LLM: Large Language Models for Health Prediction via Wearable Sensor Data

ALPHA: AnomaLous Physiological Health Assessment Using Large Language Models

Graph-Augmented LLMs for Personalized Health Insights: A Case Study in Sleep Analysis

Large Language Model Agents for Improving Engagement with Behavior Change Interventions: Application to Digital Mindfulness

Efficient and Personalized Mobile Health Event Prediction via Small Language Models

Redefining Digital Health Interfaces with Large Language Models

Critical Care Studies Using Large Language Models Based on Electronic Healthcare Records: A Technical Note

Leveraging Large Language Models for Patient Engagement: The Power of Conversational AI in Digital Health

LLM-CGM: A Benchmark for Large Language Model-Enabled Querying of Continuous Glucose Monitoring Data for Conversational Diabetes Management.

AWARE Narrator and the Utilization of Large Language Models to Extract Behavioral Insights from Smartphone Sensing Data

Supporting Physical Activity Behavior Change with LLM-Based Conversational Agents

Large Language Models as Agents in the Clinic

A Survey of Large Language Models for Healthcare: from Data, Technology, and Applications to Accountability and Ethics

Large Language Models Illuminate a Progressive Pathway to Artificial Intelligent Healthcare Assistant

Conversational Health Agents: A Personalized LLM-Powered Agent Framework

MedAgents: Large Language Models as Collaborators for Zero-shot Medical Reasoning