Predicting Healthcare Utilization Outcomes With Artificial Intelligence: A Large Scoping Review

Area of Science:

• Artificial Intelligence in Healthcare
• Health Services Research
• Predictive Analytics

Background:

• AI is increasingly used for healthcare utilization prediction.
• Existing research has gaps in data, methods, and reporting standards.
• A comprehensive overview is needed to guide future AI applications in healthcare.

Purpose of the Study:

• To map the research landscape of AI-based healthcare utilization prediction.
• Identify trends, gaps, and opportunities in datasets, methodologies, outcomes, and reporting.
• Provide insights for enhancing AI's reliability and impact in healthcare planning.

Main Methods:

• Scoping review following Joanna Briggs Institute methodology.
• Searched three major databases (inception to January 2025).
• Extracted data on datasets, AI methods, predicted outcomes, and reporting adherence (TRIPOD+AI).

Main Results:

• 121 studies included; most from the US (62%).
• Limited data variable inclusion; electronic health records (60%) and claims (28%) were primary sources.
• Ensemble models (66.9%) dominated; deep learning was less common (16.5%).
• Focus on predicting future events (90.1%), mainly hospitalizations (57.9%).
• Moderate general reporting, but limited TRIPOD+AI compliance.

Conclusions:

• Broaden predicted outcomes to include process- and logistics-oriented events.
• Expand AI applications beyond prediction to cohort selection and matching.
• Explore underused AI methods like distance-based algorithms and deep neural networks.
• Enhance adherence to TRIPOD+AI reporting guidelines for improved reliability and impact.