Patient Perspectives on AI-Driven Predictions of Schizophrenia Relapses: Understanding Concerns and Opportunities for Self-Care and Treatment

Title of the Paper

Patient Perspectives on AI-Driven Predictions of Schizophrenia Relapses: Understanding Concerns and Opportunities for Self-Care and Treatment

Paper Information

• Subject Area: Intersection of artificial intelligence and mental health, with a particular focus on patient feedback regarding schizophrenia relapse prediction
• Keywords: Artificial intelligence, mental health, schizophrenia relapse, patient perspectives, social media data, self-reflection, support systems, privacy, ethical issues, human-computer interaction

Research Background and Problem

• Identified Challenges: Current AI models for predicting schizophrenia relapses often rely on self-reported data, social media, or other sources. However, their practical effectiveness, patient perspectives, and ethical risks have not been thoroughly explored.
• Significance of the Research: Schizophrenia affects approximately 24 million people worldwide, with over 80% experiencing relapses that severely disrupt their quality of life. Early detection and intervention are critical to mitigating symptom deterioration.
• Motivation and Related Work: AI models have the potential to predict relapse signals in schizophrenia, utilizing data such as electronic health records, smartphone usage, or social media activity. However, there are significant gaps in aligning these models with patients' lived experiences and addressing ethical concerns, including privacy, data security, and potential harm to mental health. This paper aims to explore patient perspectives to address these gaps in existing AI research in mental health.

Proposed Solution

• Proposed Approach:
  • Conduct a three-phase study: semi-structured interviews to gather patient feedback; prototype development based on feedback; and user data collection to evaluate the prototype.
  • Design patient-friendly language in the prototype to replace terms like "relapse," allow users to view the raw social media data used by the AI, and facilitate dialogue with support systems.
• Innovative Aspects:
  • Emphasis on "patient-human-AI" multi-relational interaction design.
  • Propose that the role of AI models should extend beyond prediction to fostering effective communication between patients and support systems.
• Implementation Steps and Key Techniques:
  • Interview Phase: Recruit 28 patients through social media support groups for one-on-one interviews, using reflexive thematic analysis to extract key concerns.
  • Prototype Development Phase:
    • Modify language to avoid "triggering" terms such as "relapse" or "hospitalization."
    • Add options for patients to share prediction results with their support systems.
    • Display AI-analyzed social media posts to enhance transparency and reduce algorithmic distrust.
  • Feedback Collection Phase: Use participants' actual social media data to simulate AI prediction processes and gather user feedback on prediction results and feature designs.

Research Outcomes

• Specific Findings:
  • Identified a significant disconnect between patients' definitions of schizophrenia relapse and the clinical assumptions embedded in current AI models, which often overlook social or behavioral signals.
  • Designed prototype models that not only focus on relapse prediction but also encourage collaboration between patients and support systems, enhancing self-management and self-reflection.
• Advantages Over Existing Solutions:
  • Compared to traditional models, this approach is more aligned with patients' lived experiences and emphasizes transparency and ethical considerations.
  • Developed interpretable prediction information displays, reducing patient anxiety.
• Experimental or Evaluation Results:
  • Participants widely appreciated the feature of sharing prediction results with support systems, viewing it as a way to foster collaboration.
  • The "review social media content" feature helped users gain self-insights into their mental state, which was perceived as a positive influence on mental health.
  • Most participants indicated that even if the AI model could not perfectly predict relapses, the tools provided were still valuable for reminding them to stay connected with their support systems.
• Limitations and Future Directions:
  • The current study has a limited sample size and lacks diversity, focusing primarily on patients in the United States without encompassing all cultural and socioeconomic backgrounds.
  • The effectiveness of the tool for patients with more severe symptoms requires further validation.
  • Future research could explore direct collaboration between AI models and support systems and develop specialized technologies for patients lacking access to support systems.

Summary: This paper reexamines the feasibility of AI in mental health from the patient perspective and proposes patient-centered system designs to enhance autonomy and collaboration, ultimately advancing the practical application of AI in predicting and intervening in schizophrenia relapses.