Artificial Neural Network Modelling of Intensive Care Mortality
(2018) FMS820 20181Mathematical Statistics
- Abstract
- In order to study and evaluate the care provided at the Intensive Care Unit (ICU), an accurate model to assess severity of illness and predict patient mortality based on admission data is required. Since 1983, the Simplified Acute Physiology Score (SAPS) based on logistic regression (LR) has become one of the international standards for this purpose and is used in Sweden for adults. While being a simple, interpretative model, LR results in a decision boundary in the form of a hyperplane, which puts a limit to the classification abilities of something as complex as severe disease. The Artificial Neural Network (ANN) is a machine learning model inspired by the neurons and synapses of the brain and produces non-linear decision boundaries for... (More)
- In order to study and evaluate the care provided at the Intensive Care Unit (ICU), an accurate model to assess severity of illness and predict patient mortality based on admission data is required. Since 1983, the Simplified Acute Physiology Score (SAPS) based on logistic regression (LR) has become one of the international standards for this purpose and is used in Sweden for adults. While being a simple, interpretative model, LR results in a decision boundary in the form of a hyperplane, which puts a limit to the classification abilities of something as complex as severe disease. The Artificial Neural Network (ANN) is a machine learning model inspired by the neurons and synapses of the brain and produces non-linear decision boundaries for classification tasks. A hint of its potential is provided by the universal approximation theorem, stating that an ANN with finite capacity can approximate any continuous function under mild assumptions. In this work, several variants of the ANN in addition to combinations of other machine learning models are developed and assessed for classifying 30-day mortality.
The optimal ANN was trained using batch normalization, dropout, and autoencoder imputation of missing values. The resulting area under the receiver operating curve (AUC) is 0.889 (95%CI: 0.888 − 0.890) on a test set of 22,002 patients, compared to 0.852 obtained by the latest version of SAPS, SAPS3. Even when removing 79% of all predictors (30/38), SAPS3 is still out- performed. The Brier Score (BS), measuring calibration error, is improved from 0.114 to 0.099 (95%CI: 0.0984 − 0.0996), and no large errors are obtained for high risk patients, in contrast to SAPS3. The ANN outperforms both LR, random forest (RF), and ensemble models in all measured regards. Major improvements are obtained especially for patients suffering from complex conditions such as cancer, cardiovascular diseases, and unconsciousness. The results pose interesting questions regarding further applications within the ICU beyond modelling of mortality, and the potential of the ANN model to be used as an international standard of risk adjustment. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/8953311
- author
- Holmgren, Gustav
- supervisor
- organization
- course
- FMS820 20181
- year
- 2018
- type
- H2 - Master's Degree (Two Years)
- subject
- language
- English
- id
- 8953311
- date added to LUP
- 2018-06-27 11:01:30
- date last changed
- 2018-06-27 12:31:46
@misc{8953311,
abstract = {{In order to study and evaluate the care provided at the Intensive Care Unit (ICU), an accurate model to assess severity of illness and predict patient mortality based on admission data is required. Since 1983, the Simplified Acute Physiology Score (SAPS) based on logistic regression (LR) has become one of the international standards for this purpose and is used in Sweden for adults. While being a simple, interpretative model, LR results in a decision boundary in the form of a hyperplane, which puts a limit to the classification abilities of something as complex as severe disease. The Artificial Neural Network (ANN) is a machine learning model inspired by the neurons and synapses of the brain and produces non-linear decision boundaries for classification tasks. A hint of its potential is provided by the universal approximation theorem, stating that an ANN with finite capacity can approximate any continuous function under mild assumptions. In this work, several variants of the ANN in addition to combinations of other machine learning models are developed and assessed for classifying 30-day mortality.
The optimal ANN was trained using batch normalization, dropout, and autoencoder imputation of missing values. The resulting area under the receiver operating curve (AUC) is 0.889 (95%CI: 0.888 − 0.890) on a test set of 22,002 patients, compared to 0.852 obtained by the latest version of SAPS, SAPS3. Even when removing 79% of all predictors (30/38), SAPS3 is still out- performed. The Brier Score (BS), measuring calibration error, is improved from 0.114 to 0.099 (95%CI: 0.0984 − 0.0996), and no large errors are obtained for high risk patients, in contrast to SAPS3. The ANN outperforms both LR, random forest (RF), and ensemble models in all measured regards. Major improvements are obtained especially for patients suffering from complex conditions such as cancer, cardiovascular diseases, and unconsciousness. The results pose interesting questions regarding further applications within the ICU beyond modelling of mortality, and the potential of the ANN model to be used as an international standard of risk adjustment.}},
author = {{Holmgren, Gustav}},
language = {{eng}},
note = {{Student Paper}},
title = {{Artificial Neural Network Modelling of Intensive Care Mortality}},
year = {{2018}},
}