Monte-Carlo Simulation Based on Patient-Individual Distributions for Supporting Intensive Care Occupancy Management

Date
2022-01-04
Authors
Witteborg, Arne
Borgstedt, Rainer
Günther, Markus
Jansen, Gerrit
Römer, Michael
Contributor
Advisor
Department
Instructor
Depositor
Speaker
Researcher
Consultant
Interviewer
Annotator
Journal Title
Journal ISSN
Volume Title
Publisher
Volume
Number/Issue
Starting Page
Ending Page
Alternative Title
Abstract
Managing Intensive Care Units (ICUs) in hospitals is a highly challenging endeavor. In particular, decisions such as admitting elective patients and discharging patients from the ICU have to be taken under a high level of uncertainty since the occupancy of ICUs does not only depend on these decisions but also on unknown parameters such emergency patient arrivals and lengths of stay of the patients in the ICU. In this paper, we develop a framework for supporting ICU occupation management by quantifying the impact of admission and discharge decisions on the probability of reaching critical ICU occupancy levels in a given planning horizon. A key component of this framework is the use of data-driven approaches for obtaining probability distributions for the parameters affected by uncertainty. In particular, we use standardized treatment and patient health state data to create patient-specific length-of-stay distributions with a Machine Learning approach. These patient-individual distributions are then validated and/or adjusted by medical experts. The validated distributions form the input to a Monte-Carlo Simulation that is used to approximate the probability distributions of the daily ICU occupancy levels resulting from ICU admission and discharge decisions. We experimentally evaluate our framework in a counterfactual simulation based on one year of historical data from 2019 from a medium-sized ICU in a German hospital. In that evaluation, we use a simple ICU management policy based on the probabilistic occupancy forecasts aiming at reducing the risk of running out of ICU capacity. The results show that following this policy would have avoided hitting critical occupancy levels by around 70% and would have had a smoothing effect on ICU occupancy levels.
Description
Keywords
Decision Support for Healthcare Processes and Services, decision support, health care, intensive care unit, machine learning, simulation
Citation
Extent
10 pages
Format
Geographic Location
Time Period
Related To
Proceedings of the 55th Hawaii International Conference on System Sciences
Table of Contents
Rights
Attribution-NonCommercial-NoDerivatives 4.0 International
Rights Holder
Local Contexts
Email libraryada-l@lists.hawaii.edu if you need this content in ADA-compliant format.