Doctoral Defense in Medical Sciences - Steindór Oddur Ellertsson

On Thursday, February 27, 2025, Steindór Oddur Ellertsson will defend his doctoral dissertation in medical sciences at the Faculty of Medicine, University of Iceland.

The dissertation is titled: "The Use of Artificial Intelligence for Diagnosis and Outcome Prediction in Primary Care."

The opponents are Dr. Ronny Gunnarsson, professor at the University of Gothenburg, and Dr. Hercules Dalianis, professor at Stockholm University.

The supervisor was Emil L. Sigurðsson, professor, with Hrafn Loftsson, lecturer, as co-supervisor. Additionally, the doctoral committee included Jón Steinar Jónsson, general practitioner, Margrét Ólafía Tómasdóttir, lecturer, and Yngvi Björnsson, professor.

Professor Þórarinn Guðjónsson, head of the Department of Medicine, will chair the ceremony, which will take place in the University of Iceland's Hátíðasalur and will begin at 9:00 AM.

Abstract
The aim of this doctoral project was to investigate whether text data from doctors' summary notes could be used to train AI models to predict diagnoses and outcomes for patients in primary care, to examine the models’ internal functionality, and to evaluate their performance. The dissertation is based on three studies: In the first study, an AI model was trained on labeled diagnostic features from doctors’ summary notes to predict primary headache diagnoses in adults in primary care. The results showed that the model performed as well as or slightly better than doctors. An analysis of the SHAP values indicated that the model relied on similar diagnostic features to those used by doctors when generating its outputs. In the second study, an AI model was trained on labeled diagnostic features from doctors’ notes for adult primary care patients with certain respiratory infection codes. The study’s results showed that the model reliably stratified patients into risk groups and correctly distinguished between severe and milder cases. In the third study, the model from Study 2 was evaluated prospectively. The results showed that the model triaged patients with severe outcomes to a high-risk group and those with milder outcomes to a low-risk group. The overall findings of this doctoral project suggest that AI models trained on text data from medical records can predict patients’ prognoses and diagnoses in primary care. Moreover, the models use similar information to that of physicians during inference for primary headache diagnoses.

About the doctoral candidate
Steindór Oddur Ellertsson was born in 1985 in Reykjavík. He graduated from the Natural Sciences Program at Menntaskólinn í Reykjavík in 2005 and completed his medical degree from the University of Iceland in 2012. Steindór began research alongside clinical work in 2019, which led him to pursue a PhD in medical sciences at the University of Iceland in February 2022. The research project received funding from the Science Fund of the Icelandic Association of General Practitioners and RANNÍS. Alongside his doctoral studies, Steindór has worked as a resident doctor in family medicine in Norway and Iceland. Steindór's parents are Ellert Kristján Steindórsson and Rannveig Alma Einarsdóttir. His wife is Olga Hrönn Jónsdóttir, and together they have three children Lísa, Björt, and Styrkár.