Predicting the patient’s remaining length of stay for real-time decision-making: a process mining approach

Purpose This research aims to develop a structured methodology to estimate the remaining length of stay (RLoS) throughout a patient’s care journey, thereby providing decision-makers with timely information to support real-time operations. Design/methodology/approach We developed a structured methodology for predictive process monitoring based on advanced process mining techniques. To test our method, we conducted a case study at a major Italian transplant hospital, extracting a dataset of 31,521 real events from 4,809 surgical patients using several hospitals' information systems. Findings Our results show that RLoS estimates become more accurate as the care process unfolds, offering valuable insights for real-time decision-making. Research limitations/implications Findings derive from single-procedure surgical pathways at one specialized center, with lower accuracy for prolonged stays that often end in transfer rather than discharge. Future research should address multicentre validation, competing terminal events and concept drift. Practical implications The proposed structured methodology transforms hospital event logs into real-time operational insights for the management of surgical inpatients, whose care unfolds through a defined pre-operative, operative, and post-operative sequence. Within this scope, it supports more proactive bed management, patient flow coordination, and resource allocation. Originality/value The study advances real-time decision-making in hospital operations by bringing predictive process monitoring to bear on RLoS estimation. It reframes RLoS prediction as a process-aware task that draws on the heterogeneous, sequentially unfolding events of the care pathway, generating multi-stage estimates along the care pathway.