ORAHS2025
Abstract Submission

28. Simulating the impact of errors on length-of-stay predictions and rescheduling policies on elective surgery planning

Invited abstract in session FA-2: Surgery scheduling 2, stream Sessions.

Friday, 9:00-10:30
Room: NTNU, Realfagbygget R8

Authors (first author is the speaker)

1. Martina Doneda
DEIB, Politecnico di Milano
2. Pieter Smet
Computer Science, KU Leuven
3. Ettore Lanzarone
DIGIP, University of Bergamo
4. Giuliana Carello
Elettronica, Informazione e Bioingegneria, Politecnico di Milano

Abstract

When planning the admission of elective surgical patients, it is crucial to ensure that the proposed plan takes into account the availability of downstream resources, with post-surgery recovery beds being among the most scarce ones. Therefore, it is sensible for managers to incorporate an estimation of patients' lengths-of-stay (LOSs) into their admission planning, so to account for the number of bed-days each patient will consume.

However, LOSs observed in practice may deviate from the estimation used during the admission scheduling phase, potentially causing inefficiencies, or even making the schedule unfeasible. To address this issue, online rescheduling strategies can be implemented iteratively, leveraging operational flexibility to adjust schedule violations. Among these real-time modifications, it is possible to include postponing admissions, reallocating patients to different wards, or transferring those already hospitalized. A proactive approach to minimize the need for adjustments is to enhance the accuracy of LOSs estimations by predicting them using machine learning (ML) tools. However, these models can be costly to train and inherently imperfect and subject to error.

Building on prior research that explored simulated ML for preliminary evaluations of data-driven strategies, this study examines the relationship between LOS predictive error and rescheduling flexibility under various corrective policies.

Keywords

Status: accepted

Back to the list of papers