ProMind-LLM: Proactive Mental Health Care via Causal Reasoning with Sensor Data

TL;DR

ProMind-LLM tackles robust mental health risk assessment by fusing objective sensor-based behavior data with subjective mental records. It introduces a three-component pipeline: continuous domain-specific pretraining and counterfactual-based supervised fine-tuning to embed mental health knowledge, a self-refine mechanism to preprocess long numerical behavior data for LLMs, and a causal chain-of-thought framework that combines factual and counterfactual reasoning to produce reliable predictions. Evaluations on PMData and Globem demonstrate competitive performance, approaching GPT-4o with considerably smaller models and showing clear gains from the two-stage training, self-refine, and causal CoT components. Overall, the work advances proactive mental health care by delivering a more interpretable, scalable, and robust LLM-based solution that leverages multimodal data while addressing uncertainties in subjective reporting; it also highlights practical deployment considerations and ethical safeguards. For example, the approach uses a binary outcome $G_i \in \{0,1\}$ and causal analysis sets $\mathbf{A}$ and $\mathbf{A}_c$ to reason through the data.

Abstract

Mental health risk is a critical global public health challenge, necessitating innovative and reliable assessment methods. With the development of large language models (LLMs), they stand out to be a promising tool for explainable mental health care applications. Nevertheless, existing approaches predominantly rely on subjective textual mental records, which can be distorted by inherent mental uncertainties, leading to inconsistent and unreliable predictions. To address these limitations, this paper introduces ProMind-LLM. We investigate an innovative approach integrating objective behavior data as complementary information alongside subjective mental records for robust mental health risk assessment. Specifically, ProMind-LLM incorporates a comprehensive pipeline that includes domain-specific pretraining to tailor the LLM for mental health contexts, a self-refine mechanism to optimize the processing of numerical behavioral data, and causal chain-of-thought reasoning to enhance the reliability and interpretability of its predictions. Evaluations of two real-world datasets, PMData and Globem, demonstrate the effectiveness of our proposed methods, achieving substantial improvements over general LLMs. We anticipate that ProMind-LLM will pave the way for more dependable, interpretable, and scalable mental health case solutions.

Figures (19)

• Figure 1: Causal relations among mental status, behavior data, and mental records. In ProMind-LLM, we utilize objective behavior data and subjective mental records to achieve robust mental risk prediction.
• Figure 2: ProMind-LLM framework. The design consists of three components: (1) Construct professional mental health LLM using domain-specific training; (2) Enhance behavior understanding through self-refine mechanism; (3) Improve outcome robustness with causal CoT.
• Figure 3: Ablation study on counterfactual augmentations in SFT.
• Figure 4: Ablation study on behavior data as complementary modality.
• Figure 5: Silhouette scores of the word embedding.