تحلیل و پیش بینی فرایندهای درمانی: رویکرد مبتنی بر علم داده فرایندگرا

Background:
Healthcare processes are inherently complex and demand data-driven strategies for effective health information management. This study integrates process mining with deep learning to predict subsequent activities in infectious disease emergency care based on event logs, aiming to support more efficient management.
Methods:
The event log comprised 1,050 patients admitted to the emergency department with infection symptoms and suspected sepsis, for whom specific treatment pathways were expected. First, patient flows were mapped using process mining tools to visualize and clarify care trajectories. Next, process data were analyzed, and the proposed model—built on deep learning with a transformer-based architecture—was employed to predict each patient’s next step. Model performance was evaluated using metrics such as accuracy.
Results:
Process mining identified bottlenecks, including delays in diagnostic tests and congestion in blood test units. The transformer model achieved an average accuracy of 83.5% in next-activity prediction, representing a 13% improvement over existing approaches. It also reduced mean absolute error (MAE) to 0.93 days for next-event time and 5.43 days for remaining time.
Conclusion:
The proposed model offers a robust tool for analyzing and improving clinical processes. This approach can reduce patient waiting times, optimize resource allocation, and enhance data-driven decision-making in healthcare. The findings contribute to advancing health information management systems and improving treatment services.