Exploring and Promoting Diagnostic Transparency and Explainability in Online Symptom Checkers

Title of the Paper

Exploring and Promoting Diagnostic Transparency and Explainability in Online Symptom Checkers

Paper Information

• Research Area: Health Information Technology, Online Symptom Checkers, Explainability in Artificial Intelligence
• Keywords: Symptom Checkers, COVID-19, Explainability, Health, Transparency, User Experience, Trust, Human-Computer Interaction, Medical Decision-Making

Research Background and Issues

• Problems and Challenges:

  • Online Symptom Checkers (OSCs) assist users in self-diagnosis and triage, but their algorithms lack transparency and explainability. This opacity, particularly in high-risk medical domains, can lead to user distrust and potential misguidance.
  • Current research primarily focuses on the diagnostic accuracy of symptom checkers, with limited understanding of actual user needs and their perceptions of the results.
  • Emphasizing diagnostic transparency and enhancing user understanding of system processes are crucial for reducing misunderstandings and optimizing user trust.

• Significance:
  OSCs are widely used in various healthcare scenarios, such as primary care, telehealth, and epidemic management. During the COVID-19 pandemic, OSCs were employed to help users determine whether testing or medical care was necessary. However, doubts persist regarding the validity and transparency of their diagnostic recommendations.

• Research Motivation and Related Work:

  • Numerous studies have highlighted the need for clearer explanations when using OSCs, particularly regarding how diagnostic recommendations are generated. However, there is a lack of user-driven explainability design research specific to the healthcare domain.
  • The authors aim to improve trust, transparency, and interaction experiences with OSCs by providing explanations.

Solution

• Methods and Solutions:

  • Research Methodology: The study consisted of two parts. The first part involved user interviews (N=25) to explore user needs for explanations in existing OSCs. The second part involved designing and developing a COVID-19 symptom checker with three explanation styles and evaluating the impact of these explanations on user experience through an experiment (N=20).
  • Three Explanation Styles:
    1. Rationale-based Explanation: Provides explanations for the purpose of each question or result.
    2. Feature-based Explanation: Outlines the user's input symptoms or information and explains how these factors influence the diagnosis.
    3. Example-based Explanation: Offers cases and data sources similar to the user's input and recommended diagnosis.

• Innovations:

  • Proposed three user-centered explanation styles and validated their design in high-risk medical scenarios.
  • Integrated multidisciplinary explanation models (e.g., XAI and HCI theories) into the transparency design of symptom checkers.

• Implementation Steps and Key Techniques:

  • Conducted thematic analysis of user interview data to identify user needs for explanations.
  • Developed a COVID-19 symptom checker with a chat-based interface, incorporating user needs during the design phase.
  • Analyzed experimental data using a user evaluation framework based on metrics such as trust, transparency, and support for medical decision-making.

Research Findings

• Specific Findings:

  • Enhanced explainability significantly improved user experience, including trust, perceived transparency, and satisfaction with diagnostic quality.
  • The Feature-based and Example-based explanations contributed the most to improving users' perception of diagnostic quality and support for medical decision-making.
  • The explanations also promoted users' awareness and understanding of their health conditions, helping to increase health literacy.

• Comparison with Existing Solutions:

  • Most current academic work focuses solely on the accuracy of diagnostic models, neglecting research on user interaction and explanation design. This study, however, provides practical optimization suggestions from a user experience perspective.
  • The experiments demonstrated the importance of these explanation models in high-risk medical domains.

• Experimental or Evaluation Results:

  • Providing explanations increased user trust in OSCs by more than 1 point (on a 7-point scale).
  • Transparency-related metrics improved by 2-3 points, showing that explanations effectively reduced user doubts about diagnostic results.
  • Users were significantly more likely to follow diagnostic recommendations when explanations were provided compared to the control group.

• Limitations and Future Directions:

  • Limitations:
    • The study focused on a specific disease (COVID-19); future research should validate the system's applicability across various disease contexts.
    • The sample size was limited and primarily consisted of university students; future studies should include a broader population.
    • Only single explanation styles were tested; future research could explore the effects of hybrid explanation styles.
  • Future Directions:
    • Develop generalizable explainable OSCs for multi-disease scenarios.
    • Further investigate the balance between transparency and information overload for users.
    • Study how to dynamically adjust explanation information to suit different user preferences (e.g., advanced users vs. beginners).

Conclusion

This study demonstrates the importance of incorporating explainability into online symptom checkers through user-driven design and experimentation. The findings show that well-designed explanations not only enhance the transparency of symptom checkers but also improve users' trust and understanding of diagnostic recommendations. Future work can focus on expanding the scope of applications and optimizing explanation designs to further enhance user experience and the acceptability of healthcare technologies.