LLMs for Healthcare Process Orchestration: Promises and Challenges

Area of Science:

• Healthcare Management
• Artificial Intelligence in Medicine
• Process Optimization

Background:

• Healthcare processes involve both stable workflows and unpredictable, case-dependent tasks.
• Integrating AI, including large language models (LLMs), requires careful orchestration rather than just technical feasibility.
• Existing evaluation metrics for healthcare processes may not adequately capture the complexities of human-AI collaboration.

Purpose of the Study:

• To develop a framework for distinguishing between structured and case-dependent tasks in healthcare.
• To guide the appropriate application of LLMs and autonomous AI systems within healthcare workflows.
• To identify key challenges and propose new evaluation metrics for human-AI integrated healthcare processes.

Main Methods:

• Conceptual review of literature on AI in healthcare processes.
• Development of a Business Process Model and Notation (BPMN)-Case Management Model and Notation (CMMN) lens.
• Analysis of task characteristics influencing AI autonomy and process structure.

Main Results:

• LLMs are best suited for structured tasks like documentation, summarization, and communication.
• More autonomous AI systems are relevant for case-dependent segments requiring context, exceptions, and judgment.
• A significant challenge lies in evaluating processes distributed across humans and AI.

Conclusions:

• Orchestration is key to effectively integrating LLMs and AI in healthcare.
• New evaluation measures are needed beyond traditional KPIs to assess human-AI collaboration quality.
• Healthcare organizations must consider supervision burden and outcome stability in AI implementation.