Aligning Artificial Intelligence Prediction Targets with Clinical Workflows Using Human Centered Design Methods.

Area of Science:

• Healthcare Informatics
• Clinical Decision Support
• Human-Computer Interaction

Background:

• Artificial intelligence (AI) models in healthcare often show strong predictive accuracy but fail to improve patient outcomes due to poor integration with clinical workflows.
• A gap exists between the technical performance of AI models and their real-world clinical utility.

Purpose of the Study:

• To demonstrate a human-centered design approach for developing healthcare AI models.
• To ensure AI prediction targets align with actionable clinical interventions and improve patient outcomes.

Main Methods:

• A case study of pediatric acute kidney injury was used.
• A multidisciplinary working group employed user stories, People, Environment, Technology, and Tasks (PETT) Scan, and process mapping.
• Sociotechnical factors and workflow leverage points were analyzed before AI model development.

Main Results:

• Distinct prediction targets were identified for different clinical roles (hospitalists, nephrologists, intensivists).
• Key barriers included inadequate monitoring, poor visibility of at-risk patients, and unclear injury progression.
• High-impact prediction targets were defined to support actionable interventions.

Conclusions:

• Integrating clinical context and human-centered design before AI development is crucial.
• This approach bridges the gap between AI model performance and clinical utility.
• The methodology can enhance the effectiveness of AI in improving patient care.