Applying machine learning to electronic health records : a study on two adverse events

• Main Author: Santos, Henrique Dias Pereira dos
• Publication Date: 2021
• Format: Doctoral thesis
• Language: eng
• Source: Biblioteca Digital de Teses e Dissertações da PUC_RS
• Download full: http://tede2.pucrs.br/tede2/handle/tede/9643
• topic: Electronic Health Records; Adverse Events; Machine Learning; Supervised Learning; Unsupervised Learning; Prontuário Eletrônico do Paciente; Eventos Adversos; Aprendizado de Máquina; Aprendizado Supervisionado; Aprendizado Não-Supervisionado; CIENCIA DA COMPUTACAO::TEORIA DA COMPUTACAO

In the hospital environment, the incidence of adverse events (AE) (unforeseen incidents that cause harm to patients) is the primary concern of risk management teams. The use of machine learning techniques could help healthcare professional to identify and mitigate adverse events.This thesis develops experiments to evaluate machine learning approaches to identify two major adverse events in electronic health records (EHR). The first algorithm was created to identify fall events in clinical notes using language models and neural networks. We annotated 1,402 clinical sentences with fall events to train a Token Classifier (TkC) to detect words within the context of falls. The TkC was able to correctly identify 85% of the sentences with fall events. For medication review, we built an unsupervised algorithm based on graph structure to rank outlier prescriptions. In our experiments, the proposed algorithm, the DDC-Outlier, correctly classified 68% (F-measure) of prescribed medications as underdoses and overdoses. Finally, to better understand the performance of our approach in a real-world scenario, we deployed a decision support system for clinical pharmacy in a 1,200-bed hospital. All experiments, source-codes, and the anonymized datasets are publicly available on the GitHub page of our research group.