University of Twente Student Theses

Login

Incorporating clinical notes in an early sepsis prediction model to improve performance

Siddique, S. (2020) Incorporating clinical notes in an early sepsis prediction model to improve performance.

[img] PDF
1MB
Abstract:Sepsis is a condition caused by the body’s response to an infection that affects an estimated 49 million people globally. It is one of the leading contributors to hospital mortality. Risk-prediction models for sepsis onset are currently used in hospitals to assist with clinical decision-making. Recently, the use of natural language processing (NLP) on clinical notes has aided in improving patient outcome predictions and identifying patient diagnosis. The aim of this study is to improve the prediction of sepsis onset by combining clinical notes to the structured data components of the electronic health records (EHR) data. The proposed model is trained on the MIMIC-III dataset and explores the use of time-series physiological data with clinical note embeddings. To assess the effect of the input features we evaluated logistic regression, multinomial Naïve Bayes, and XGBoost models on the following three configurations: physiological measures, clinical note embeddings, and the combination of physiological and note features. We compared three methods of clinical notes representations: one-hot encoded pointwise mutual information (PMI) vectors, term-frequency inverse-document frequency weighted PMI vectors, and pre-trained note embeddings. Furthermore, we assessed the effect of using different prediction time and look back intervals of time-series physiological data with clinical note embeddings.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Interaction Technology MSc (60030)
Link to this item:https://purl.utwente.nl/essays/83703
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page