Generative artificial intelligence for outcome prediction in critical care: the future is now?

Area of Science:

• Critical care medicine
• Artificial intelligence
• Machine learning

Background:

• Artificial intelligence (AI) is poised to revolutionize critical care.
• Generative AI, a recent advancement, can process and generate data sequences for outcome prediction.
• The field is rapidly evolving with potential to enhance prognostic accuracy.

Purpose of the Study:

• To review the emerging applications of generative AI in critical care outcome prediction.
• To compare different generative AI approaches for forecasting patient trajectories.
• To assess the current readiness of generative AI for clinical implementation.

Main Methods:

• Review of current literature on generative AI in critical care.
• Comparison of text-based (Large Language Models - LLM) and event-based (Generative Trajectory Models - GTM) generative AI approaches.
• Evaluation of model performance based on data type and training scale.

Main Results:

• Limited publications highlight significant challenges in generative AI for critical care.
• Generative Trajectory Models (GTM) demonstrate superior performance over Large Language Models (LLM) in outcome prediction.
• GTMs show enhanced accuracy when trained on large, diverse healthcare datasets.

Conclusions:

• Generative AI offers a potential shift from static risk scores to dynamic, sequential outcome simulation in critical care.
• Current evidence is preliminary and insufficient for integrating generative AI into bedside decision-making.
• Further research and validation are required before clinical adoption.