From Classification to Clinical Insights: Towards Analyzing and Reasoning About Mobile and Behavioral Health Data With Large Language Models (2311.13063v3)

Abstract: Passively collected behavioral health data from ubiquitous sensors holds significant promise to provide mental health professionals insights from patient's daily lives; however, developing analysis tools to use this data in clinical practice requires addressing challenges of generalization across devices and weak or ambiguous correlations between the measured signals and an individual's mental health. To address these challenges, we take a novel approach that leverages LLMs to synthesize clinically useful insights from multi-sensor data. We develop chain of thought prompting methods that use LLMs to generate reasoning about how trends in data such as step count and sleep relate to conditions like depression and anxiety. We first demonstrate binary depression classification with LLMs achieving accuracies of 61.1% which exceed the state of the art. While it is not robust for clinical use, this leads us to our key finding: even more impactful and valued than classification is a new human-AI collaboration approach in which clinician experts interactively query these tools and combine their domain expertise and context about the patient with AI generated reasoning to support clinical decision-making. We find models like GPT-4 correctly reference numerical data 75% of the time, and clinician participants express strong interest in using this approach to interpret self-tracking data.

From Classification to Clinical Insights: Leveraging LLMs for Mental Health Data Analysis

Introduction

The utilization of passive sensory data from mobile and wearable devices offers a promising avenue for enriching mental health assessments with quantitative insights drawn from individuals' daily lives. However, the integration of this data into clinical practices poses challenges, including device generalization, the ambiguous correlation between sensor data and mental health states, and the interpretation of voluminous sensor data by clinicians. Addressing these challenges, this paper pioneers the use of LLMs to synthesize clinically relevant insights from multi-modal sensor data, moving beyond binary classification towards empowering clinical decisions through a novel human-AI collaborative approach.

Leveraging LLMs for Data Processing

This research marks the initial exploration into processing multi-sensor ubiquitous data with LLMs, setting itself apart from traditional signal processing or standard ML methodologies. Employing models like GPT-4, the paper demonstrates advanced abilities of LLMs in performing binary classification tasks through chain-of-thought prompting, as well as generating insightful reasoning from sensor data. Compared to traditional machine learning baselines, LLMs presented a superior performance, particularly when fine-tuned and employing thoughtful prompting strategies, achieving a maximum accuracy of 61.1% in depression classification.

Shifting Focus to Generative Reasoning

A significant realization from this paper is recognizing the limitations of binary classification in clinical contexts, especially considering the nuanced nature of mental health diagnoses. The greater impact lies in leveraging LLMs' generative capabilities, where the model's reasoning about sensor data trends could provide a more qualitative analysis useful for human clinicians. Through human-AI collaboration, this approach aims to enrich clinical decision-making, combining AI-generated insights with clinicians' expert judgment and patient context.

Clinical Implications and Future Directions

The generative reasoning capability of LLMs, evaluated both in terms of numerical accuracy and clinical relevance, promises a significant shift in how clinicians could interact with patient-generated sensor data. The paper underscores the efficacy of human-AI collaboration, where clinicians find value in LLM-generated analyses to inform therapy discussions, augment patient engagement, and potentially enhance treatment outcomes. This paradigm shift advocates for a more integrative use of AI in mental health care, suggesting future research to focus on improving model accuracies, expanding the range of analysed behaviors, and exploring ethical considerations related to privacy and personalized care.

Conclusion

Exploring the use of LLMs to analyze mobile and behavioral health data unveils a pathway towards enriching mental health assessments with data-driven insights. This paper's approach extends beyond the confines of binary classification, proposing a human-AI collaborative model to synthesize and reason about sensor data. This innovative paradigm fosters a deeper integration of quantitative data analysis within clinical practices, potentially transforming patient care through personalized, data-informed therapeutic interventions.