TLDR: HealthProcessAI is a new framework that simplifies complex healthcare process mining by integrating Large Language Models (LLMs) for automated interpretation and report generation. It acts as a wrapper for existing process mining libraries, making the technology more accessible to clinicians and researchers. Validated with sepsis data, the framework demonstrated robust performance, with LLMs like Claude Sonnet-4 and Gemini 2.5 Pro showing high consistency in generating understandable reports, and also highlighted cost-effective AI integration.
Process mining has emerged as a powerful tool for understanding complex workflows, especially in healthcare. However, its widespread adoption faces significant hurdles, including technical complexity, a lack of standardized methods, and limited training resources. Healthcare professionals and data scientists often find existing tools challenging to use, and interpreting the outputs requires a deep understanding of both algorithmic principles and clinical contexts.
Addressing these challenges, researchers have introduced HealthProcessAI, a new framework designed to simplify process mining applications in healthcare and epidemiology. This innovative system acts as a comprehensive wrapper around existing Python (PM4PY) and R (bupaR) libraries, making advanced analytical techniques more accessible. A key feature of HealthProcessAI is its integration of multiple Large Language Models (LLMs) for automated interpretation of process maps and report generation. This helps translate complex technical analyses into outputs that a diverse range of users, including clinicians, data scientists, and researchers, can easily understand.
The HealthProcessAI framework is built on a modular architecture comprising six main components: data loading and preparation, process mining analysis, LLM integration for interpretation, advanced analytics, multi-model report orchestration, and a validation framework. This design ensures a structured approach from raw data to actionable insights. The framework was validated using sepsis progression data as a proof-of-concept, and the outputs of five state-of-the-art LLM models were compared through the OpenRouter platform.
The system successfully processed sepsis data across four proof-of-concept scenarios, demonstrating robust technical performance and its ability to generate reports through automated LLM analysis. An evaluation using five independent LLMs as automated assessors revealed distinct strengths among the models. Claude Sonnet-4 and Gemini 2.5 Pro achieved the highest consistency scores, indicating their strong performance in interpreting and generating reports.
One of the significant advantages of HealthProcessAI is its ability to reduce technical and training barriers in healthcare process mining while maintaining scientific objectivity. By integrating LLMs for automated interpretation and report generation, it tackles the common unfamiliarity with process mining outputs, making them more user-friendly. This combination of structured analytics and AI-driven interpretation represents a novel advancement in translating complex process mining results into potentially actionable insights for healthcare applications.
The framework also incorporates an economic analysis through its OpenRouter integration, highlighting the cost-effectiveness of different LLMs. DeepSeek R1, for instance, demonstrated the highest performance-to-cost ratio, making it an efficient choice for certain applications. The OpenRouter platform itself provided substantial operational and financial benefits, significantly reducing costs compared to direct API pricing.
The research paper, available at HealthProcessAI: A Technical Framework and Proof-of-Concept for LLM-Enhanced Healthcare Process Mining, details how this framework can transform healthcare process data into standardized event log formats with AI-enhanced interpretation. It provides accessible interpretation for clinical stakeholders through integrated educational components, addressing the limitations of traditional manual analysis which often lacks scalability and consistency.
Also Read:
- AI Clinical Teams: A New Approach to Diagnosing Patient Problems from Medical Notes
- MEDLEY: Harnessing AI’s Imperfections for Smarter Medical Decisions
While the current work focuses on technical development and initial validation using synthetic and retrospective data, future plans include direct clinical validation with healthcare practitioners using real-time data. This will involve systematic evaluation by clinicians and prospective deployment in clinical settings to validate actionable insights against actual patient outcomes and process improvements. HealthProcessAI lays a strong foundation for advancing healthcare process mining through AI integration, promising a more accessible and scalable approach to clinical process intelligence as data-driven healthcare continues to evolve.
