Predicting Deterioration of Patients in the Intensive Care Unit
Bouabana, Hatem (2018)
2018
Tietojenkäsittelytieteiden tutkinto-ohjelma - Degree Programme in Computer Sciences
Luonnontieteiden tiedekunta - Faculty of Natural Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2018-05-30
Julkaisun pysyvä osoite on
http://urn.fi/URN:NBN:fi:uta-201806262085
http://urn.fi/URN:NBN:fi:uta-201806262085
Tiivistelmä
The massive influx of data in healthcare encouraged the building of data-driven machine learning models in order to improve medical diagnosis and assess patients health status. Thus, gaining knowledge and actionable insights from this complex, heterogeneous, poorly annotated and generally unstructured data remains a key challenge in transforming healthcare.
In this thesis, we focus on consecutive events in patients Electronic Health Records (EHR) data to predict organ failure in the ICU. We first investigate the ability of both logistic regression and random forest classifiers to achieve this task. Then, we compare these methods along with convolutional neural networks (CNN) for prediction of multiple organ failures: cardiovascular and pulmonary. Our predictions are done using a restricted set of parameters to enable "real-time" usage in the ICU. Predictions are made three hours in the future to support clinically actionable planning while having a 90 minutes window of parameters history. Our results suggest that the CNN has substantial additional predictive value for cardiovascular and pulmonary organ failures among critically ill patients.
In this thesis, we focus on consecutive events in patients Electronic Health Records (EHR) data to predict organ failure in the ICU. We first investigate the ability of both logistic regression and random forest classifiers to achieve this task. Then, we compare these methods along with convolutional neural networks (CNN) for prediction of multiple organ failures: cardiovascular and pulmonary. Our predictions are done using a restricted set of parameters to enable "real-time" usage in the ICU. Predictions are made three hours in the future to support clinically actionable planning while having a 90 minutes window of parameters history. Our results suggest that the CNN has substantial additional predictive value for cardiovascular and pulmonary organ failures among critically ill patients.