Machine Learning for Adverse Event Analysis in Healthcare - Identifying Opportunities and Conducting a Classification Study - Master’s thesis

Aug 4, 2026

Adverse events, unintended harms to patients, serve as a critical indicator of hospital patient safety, while research suggests many such events are preventable through systematic analysis and intervention. The thesis studies how machine learning can improve adverse event reporting and analysis at St. Olav’s Hospital, using 46,087 reports from 2015–2022. It identifies improvement opportunities via clinician interviews, then validates feasibility and reliability for automatic NOKUP-based classification using Naive Bayes and SVM.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages138

Reading Time5 h 48 min

AccessFree to download

Abstract

Objectives and data

Interview-based process insights

Automated NOKUP-based classification study

Experimental evaluation (Naive Bayes, SVM)

Conclusions and future directions

Master’s thesis NT NU Norwegian Un iversity of Science and Technology Faculty of Information Techno logy and Electrical Engineering Department of Computer Science Ragnhild Kleiven and Solveig Hergot Langås Machine Learning for Adverse Event Analysis in Healthcare Identifying Opportunities and Conducting a Classification Study Master’s thesis in Computer Science Supervisor: Øystein Nytrø Co-supervisor: Melissa Yan June 2023 Ragnhild Kleiven and Solveig Hergot Langås Machine Learning for Adverse Event Analysis in Healthcare Identifying Opportunities and Conducting a Classification Study Master’s thesis in Computer Science Supervisor: Øystein Nytrø Co-supervisor: Melissa Yan June 2023 Norwegian University of Science and Technology Faculty of Information Technology and Electrical Engineering Department of Computer Science i Abstract Adverse events, which refer to unintended harm caused to patients, serve as a crucial indicator of patient safety within hospitals. While some are unavoidable, research has revealed that many adverse events are preventable. Thus it is crucial for hospitals to have a systematic approach to address adverse events, enabling analysis and insights for preventive measures to enhance patient safety. This thesis aims to research how applications of machine learning can aid in improving the adverse event reporting and analysis process at St. Olav’s Hospital. With access to a dataset of 46,087 adverse event reports spanning from 2015 to 2022, the thesis explores how the textual data in the reports can be utilized with machine learning to enhance the current process. The thesis has two main objectives: firstly, identifying potential areas within the adverse event reporting and analysis process where machine learning can contribute to improvement, and secondly, selecting a specific area to conduct a machine learning study in order to validate its feasibility and reliability. To gain insights into the current process, interviews were conducted with clinicians and experts. These interviews provided valuable insights into the existing process and helped identify areas that could benefit from enhancements. One area identified from the interviews where the automatic classification of adverse events according to the National guidelines for classification of patientrelated adverse events (NOKUP) . This application can potentially improve the current process by addressing issues such as misclassification and inconsistent categorization that hinder accurate analysis and identification of preventive measures. With these benefits in mind and confirming the interest of the clinicians, this application was selected for the machine learning study. The machine learning study aimed to verify the feasibility of the classification of adverse events into predefined categories. It explored the potential of two distinct classification techniques, N¨aive Bayes (NB) and Support Vector Machine (SVM) . Several experiments were conducted to optimize the performance of both classifiers. The resulting macro F1-score for NB was 0.7182 and 0.7165 for SVM. Although these results could be considered reliable in other domains, further improvement is needed before the models could be implemented in a healthcare context. However, they demonstrate the potential of automatic classification of adverse events. Thus, this thesis has provided a foundation for the automatic classification of adverse events, demonstrating its significance and potential. Moreover, in collaboration with clinicians, two additional machine learning applications have been identified, providing valuable insights for future research directions. ii Sammendrag Uønskede hendelser er en nøkkelindikator for pasientsikkerhet p˚a sykehus. Slike hendelser innebærer utilsiktede pasientskader, og selv om noen av disse er uunng˚aelige, viser forskning at mange uønskede hendelser kan forhindres. For˚a bedre pasientsikkerheten er det viktig

FAQs

What problem does the thesis address in healthcare?

The thesis addresses how hospitals can systematically analyze adverse events to support prevention and improve patient safety.

What dataset is used for the machine learning study?

It uses 46,087 adverse event reports collected between 2015 and 2022.

Which classification techniques were evaluated and what were the results?

Naive Bayes and Support Vector Machine were evaluated and optimized; Naive Bayes achieved a macro F1-score of 0.7182 and SVM 0.7165.

Aug 4, 2026

Upload by Theodore