Due to copyright issues, an electronic copy of the thesis must be ordered from the faculty. For the faculty to have time to process the order, the order must be received by the faculty at the latest 2 days before the public defence. Orders received later than 2 days before the defence will not be processed. After the public defence, please address any inquiries regarding the thesis to the candidate.
Trial Lecture – time and place
See Trial Lecture.
Adjudication committee
- First opponent: Researcher Nina Vibeche Skei, Nord-Trøndelag Hospital Trust
- Second opponent: Head of Classical Hematology Armin Piehler, Munich Leukemia Laboratory
- Third member and chair of the evaluation committee: Professor II Torsten Eken, University of Oslo
Chair of the Defence
Associate Professor Jørgen Vildershøj Bjørnholt, University of Oslo
Principal Supervisor
MD, PhD Erik Koldberg Amundsen, Oslo University Hospital
Summary
In this thesis we used supervised and unsupervised models on patients with infectious diseases, with a focus on laboratory variables from patients admitted to the emergency department (ED). These models are used for diagnosis, prognosis, and subgrouping of patients.
In the first part of the thesis, we evaluated four published supervised models for prediction of outcomes in COVID‑19 patients. Using cohorts from Oslo and London, we found that all models performed worse and with greater variability than in their original studies. Recalibration improved performance and the results highlight the importance of validating prediction models across different populations.
Next, we developed a supervised model to predict infection using Sysmex cell population (CP) data from patients with suspected sepsis. These research parameters are produced by hematology instruments when analyzing white blood cells and reflect the cells morphology and degree of activation. Among the parameters, monocyte complexity (MO‑X) and neutrophil fluorescence activity (NEUT‑RI) showed the strongest individual predictive value. Using LASSO and backward selection, we created a four‑variable model that performed slightly below C‑reactive protein (CRP) alone. The multilayer perceptron model demonstrated the best calibration, particularly when CRP was added. Further standardization and external validation are required before clinical use.
Finally, we explored unsupervised clustering to identify subgroups of sepsis patients. Unlike the SENECA study, which reported four phenotypes, our analyses did not reproduce these subgroups. A large multicenter replication study with cohorts from Oslo, Karolinska, and Oxford also failed to reproduce the SENECA subgroups, suggesting that these phenotypes may not generalize across populations. In the Oslo cohort, seven clusters were more stable and clinically distinct. Further validation of the seven clusters should be conducted.
Additional information
Contact the research support staff.