On November 4th, at 12:15 Merli Koitmäe will defend her thesis "Modeling time-to-event outcomes in complex data structures".
Supervisors:
Professor Krista Fischer, University of Tartu
Associate professor Märt Möls, University of Tartu
Opponent:
professor Janne Pitkäniemi, Tampere University, Finland
Summary:
Advances in science and technology have enabled the collection and storage of large-scale datasets, highlighting the need for methods that can handle complex data structures. In many studies, the outcome of interest is time to event, such as disease onset, death, or treatment success. One of the most widely used methods for this is the Cox proportional hazards model. Although flexible, its assumptions must be carefully considered to avoid biased results. This dissertation examined the use of the Cox model in complex data settings and the challenges arising from its application. The first part focused on biobank and genetic data. The results showed that the choice of timescale strongly affects outcomes: time from birth to event is most suitable for discovering new associations, whereas age from study entry to event provides more accurate risk estimates. The analyses also showed that combining parental survival data with offspring genetic information may violate key assumptions and lead to biased estimates. The second part used genetic data to improve understanding of type 2 diabetes development. Three biological pathways were identified: glycaemic regulation, insulin resistance and cardiometabolic, and obesity and lifestyle pathway. Genetic risk scores for these pathways helped predict which comorbidities may occur after type II diabetes diagnosis. The obesity and lifestyle pathway was associated with the widest range of comorbidities. The third part described the development of a novel score for assessing gastrointestinal dysfunction in critically ill patients. The score predicted both 28- and 90-day mortality and improved the performance of existing prognostic models even after accounting for disease severity. Overall, the findings show that the Cox model is a versatile tool for both theoretical and applied research. At the same time, they emphasize the importance of understanding its assumptions and limitations when interpreting results from complex data.
