Machine learning for mortality risk prediction with changing patient demographics

WAINWRIGHT, Richard and SHENFIELD, Alex (2023). Machine learning for mortality risk prediction with changing patient demographics. In: 2023 IEEE Conference on Computational Intelligence in Bioinformatics and Computational Biology (CIBCB). IEEE.

Preview

PDF IEEE_CIBCB_2023___ML_for_mortality_risk_prediction_final_submitted.pdf - Accepted Version Creative Commons Attribution. Download (3MB) | Preview

Abstract

Over the last 25-30 years there has been significant work carried out in producing risk prediction models for patients admitted to intensive care units. The most recent of these models in widespread use is the Intensive Care National Audit and Research Centre (ICNARC) model developed in 2007 which uses data from more than 230,000 admissions to UK intensive care units to develop and validate a UK based model outperforming other approaches. However, as with the majority of risk prediction models, the ICNARC model struggles with changing patient cohort demographics (such as the aging populations seen currently in the western world) and requires periodic recalibration.This paper introduces a machine learning pipeline for developing mortality prediction models and uses it to train a variety of ML models. The top performing of these outperform current commonly used mortality risk prediction models such as APACHE-II, SAPS-II, and the ICNARC model. This machine learning pipeline is then extended to allow continuous retraining via online learning. The results show that it is possible to retrain our model at different intervals to deal with varying patient demographics - improving model performance across a range of different patient cohort scenarios.

Item Type:	Book Section
Identification Number:	https://doi.org/10.1109/CIBCB56990.2023.10264891
SWORD Depositor:	Symplectic Elements
Depositing User:	Symplectic Elements
Date Deposited:	23 Aug 2023 09:05
Last Modified:	11 Oct 2023 11:15
URI:	https://shura.shu.ac.uk/id/eprint/32289

Actions (login required)

View Item

Downloads

Publication Metrics