A CRITICAL REVIEW AND COMPARISON OF METHODS TO PREDICT MORTALITY IN PEOPLE WITH CYSTIC FIBROSIS

Abstract

Cystic fibrosis (CF) is a rare genetic disease that predominantly affects the lungs. Lung damage is the primary cause of complications and mortality in individuals with CF. Lung transplantation is one of the few treatment options available for those with end-stage lung damage. Due to the limited availability of donor lungs, selecting transplant candidates carefully is crucial to maximizing long-term survival. The timing of the transplant plays a vital role in survival; thus, accurately predicting individuals with an increased risk of mortality is essential for prioritizing candidates for transplantation. Several models have been proposed to predict the risk of mortality in CF patients. However, only one of these models is currently in clinical use. This thesis aims to (1) review previously proposed models and (2) compare these models using a single dataset. Nine models were developed in the Canadian Cystic Fibrosis Registry. Three regression models (logistic, Cox and landmark regression) and three variable selection methods (forward, backward, and lasso) were paired to each other so that there were 3 × 3 = 9 models. Logistic and Cox regression showed better performance than landmark, and forward and backward selection generally performed better than lasso. However, while these models performed well, one did not perform much better than the others. They also did not address the problem of rare events. Therefore, more investigations are still needed to address the limitations of the models.