MIND: Empowering Mental Health Clinicians with Multimodal Data Insights through a Narrative Dashboard

Paper Title

MIND: Empowering Mental Health Clinicians with Multimodal Data Insights through a Narrative Dashboard

Publication Info

• Topic area: Mental health data visualization and clinical decision support systems
• Keywords: Narrative dashboards, multimodal data, mental health, clinical decision-making, large language models, co-design, patient-generated data, clinical notes, visualization, healthcare informatics

Background and Problem

• Problem / challenge: Existing dashboards for mental health clinicians often present multimodal data in fragmented formats, requiring clinicians to manually synthesize insights, which is time-consuming and cognitively demanding. Machine learning models for mental health prediction lack generalizability and fail to provide actionable insights for clinical decision-making.
• Significance: Efficiently integrating and presenting multimodal data can improve clinicians' ability to understand patient conditions, make informed decisions, and optimize treatment outcomes, especially given the time-sensitive nature of clinical workflows.
• Motivation and related work: Prior research has explored multimodal data modalities (e.g., clinical notes, active sensing, passive sensing) and clinical dashboards, but these systems often lack narrative coherence and integration, leaving clinicians to navigate large volumes of data without sufficient contextual explanation. This paper builds on the concept of narrative dashboards to address these gaps.

Solution

• Proposed approach: MIND (Multimodal data Integrated Narrative Dashboard), a large language model-powered system that presents multimodal data insights through narrative text and interactive visualizations.
• Novelty:
  1. Development of a narrative dashboard that integrates multimodal data into coherent, clinically relevant stories.
  2. Implementation of a hybrid computational pipeline combining large language models and rule-based methods for insight generation.
  3. Co-design methodology involving five mental health experts to refine design goals and align the system with real-world clinical needs.
  4. Evaluation of MIND's effectiveness through a user study with 16 licensed mental health clinicians, demonstrating its advantages over baseline dashboards.
• Procedure and key techniques:
  • Co-design process with domain experts to identify design goals (intuitive, text-first, adaptive, insightful).
  • Creation of a two-level interface: Level 1 for narrative insights and Level 2 for drill-down data exploration.
  • Hybrid computational pipeline with modules for analyzing raw data, synthesizing insights, and narrating results.
  • User study comparing MIND with a baseline data collection dashboard (FACT) using simulated patient cases.

Results

• Concrete findings:
  • MIND significantly outperformed the baseline dashboard FACT in helping clinicians discover hidden data insights (p<.001) and supporting clinical decision-making (p=.004).
  • MIND was rated higher in multimodal data integration (p<.001), narrative cohesiveness (p=.004), efficiency (p=.006), and time-saving potential (p<.001).
  • Clinicians reported high trustworthiness and clinical relevance of MIND's insights, comparable to FACT.
• Advantage over baselines: MIND's narrative and integrative design improved clinicians' ability to synthesize multimodal data and apply insights directly to decision-making, while maintaining usability and workload levels comparable to the baseline.
• Experiments / evaluation:
  • Mixed-method within-subject study with 16 licensed mental health clinicians.
  • Comparison of MIND and FACT using simulated patient cases and measures such as SUS, NASA-TLX, and Likert-scale surveys.
  • Qualitative feedback highlighting MIND's summarization capability, intuitive navigation, and transparency.
• Limitations and future work:
  • Limited evaluation on homogeneous simulated patient cases; real-world heterogeneity may affect performance.
  • Participant sampling may over-represent clinicians with positive attitudes toward AI.
  • Controlled lab study lacks ecological validity; future work should explore real-world deployment and longitudinal evaluations.

Summary

MIND introduces a narrative dashboard that integrates multimodal mental health data into coherent, clinically relevant insights, addressing challenges in fragmented data presentation. Through a hybrid computational pipeline and co-design methodology, MIND balances narrative storytelling with interactive exploration, enabling clinicians to efficiently review patient information and make informed decisions. A user study demonstrated MIND's superiority over traditional dashboards in revealing hidden insights, supporting decision-making, and integrating multimodal data. Future work should focus on real-world evaluations, tailoring the system for diverse clinical workflows, and addressing ethical considerations in AI-powered healthcare systems.