Quality of life after the ICU: A machine learning approach to one-year post-ICU health predictions
Keywords
Loading...
Authors
Issue Date
2020-08-01
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
PURPOSE - Due to the increasing number of Intensive Care Unit (ICU) survivors, the long term outcomes
of the survival of a serious condition become progressively more important. With predictive modeling, the
consequences of the ICU stay in terms of quality of life (QoL) can be estimated. This study aimed to improve
the statistical model that is currently in use by using machine learning models and additional patient data
from the Electronic Health Records (EHR).
METHODS - Data from the EHR was gathered from adult patients that were admitted to the ICU between
June 2016 and January 2019. Several regression-based machine learning models, among which Random Forest
and Support Vector Regression, were tted on a combination of patient-reported data and 71 expert-selected
EHR variables in order to predict the target value of change in QoL one year after admission. Results were
compared to a baseline model.
RESULTS - Results from all tested regression models show that certain features that can be extracted from
EHR data, such as high body temperatures and low BMIs, can have an in
uence on the regression-based
prediction scores. However, the improvements in target prediction caused by these additions are limited. The
best performing model had a decrease of 0.004 in mean absolute error (MAE) compared to the baseline results.
CONCLUSION - A machine learning-based approach to one-year post-ICU QoL prediction using EHR data
was developed in this study. This approach had a small positive e ect on prediction results when compared to
the established statistical model, but the distinction is too small to put the model into practice without having
to compromise the practicability by physicians.
Keywords Quality of life, Critical care, Survivors, Machine learning, Prediction modeling
Description
Citation
Faculty
Faculteit der Sociale Wetenschappen