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: Professor Tomas Wester, Karolinska Institutet, Sweden
- Second opponent: Associate Professor Ishita Barua, University of Agder, Norway
- Third member and chair of the evaluation committee: Associate Professor Hans Christian Erichsen Landsverk, University of Oslo
Chair of the Defence
Associate Professor Kine Pedersen, University of Oslo
Principal Supervisor
Senior Research Investigator Rune Ougland, Vestre Viken Hospital Trust
Summary
Clinical prediction underpins daily diagnostic, prognostic and treatment decisions, making it central to the care of children with diseases requiring pediatric surgery and to counseling their families. Reliable decision support tools can strengthen shared decision-making by improving the quality of clinical advice and enhancing parental health competence.
This thesis aims to improve diagnostic risk stratification in children with constipation undergoing evaluation for Hirschsprung’s disease (HD), and prognostic guidance for parents of boys with posterior urethral valves (PUV). Prediction is particularly challenging when studying such rare disease, where small cohorts limit robust evidence generation and often result in uncertain recommendations in clinical guidelines. We therefore sought to develop prediction models that could support more confident clinical decision-making in this setting.
Paper I evaluated the diagnostic value of guideline-recommended clinical features in 225 children with constipation. Only 22% of the children who underwent rectal biopsy were diagnosed with HD, and the diagnostic yield was only 10% among girls, with decreasing prevalence with increasing age. Building on these findings, Paper II demonstrated that a recall-optimized machine learning (ML) framework could have reduced the number of rectal biopsies by 35% among the 178 children older than one month, without missing a single HD diagnosis. Paper III applied a recursive feature elimination ML framework to identify early predictors of long-term renal dysfunction in 83 boys with PUV. Together, Papers II and III demonstrate that ML approaches can generate robust and clinically meaningful predictions from small and imbalanced datasets.
Overall, this thesis demonstrates that interpretable ML-based clinical decision-support tools have the potential to improve diagnostic and prognostic evaluations in rare pediatric diseases. If implemented in clinical practice, such tools could reduce unnecessary diagnostic procedures, support more personalized counselling and follow-up, and promote more efficient use of healthcare resources.
Additional information
Contact the research support staff.