Design of an interface to communicate artificial intelligence-based prognosis for patients with advanced solid...

Area of Science:

• Oncology
• Medical Informatics
• Human-Computer Interaction

Background:

• Communicating machine learning (ML)-based prognosis for advanced solid tumors presents challenges for oncologists.
• Existing tools may not adequately address the need for clear, interpretable prognostic information in clinical practice.

Purpose of the Study:

• To design and refine an interface that supports oncologists in communicating ML-based prognosis.
• To incorporate oncologists' needs and feedback throughout the design process using a user-centered approach.

Main Methods:

• Employed an interdisciplinary, user-centered design approach with 5 iterative design rounds.
• Conducted expert reviews, incorporated feedback from a color-blind adult, and performed 13 semi-structured interviews with oncologists.
• Utilized patient vignettes and interfaces with representative data to gather feedback on predicted survival for treatment decision points.

Main Results:

• Developed a 7-section interface enabling oncologists to "tell a story" about prognosis.
• Iterative enhancements addressed user-focused questions, improved communication of ML-based prognosis, and exposed design assumptions.
• Clinicians prioritized interpretability over explainability, requesting specific usability and trust enhancements (e.g., using months instead of days, addressing prior treatments).

Conclusions:

• User-centered design and ongoing clinical input are crucial for developing effective AI-enabled decision support tools.
• Visualizations that clearly communicate ML-related outcomes are essential, especially for conveying prognosis risk.
• The designed interface aims to enhance trust and understanding in ML-driven prognostic communication between oncologists and patients.