Seminar by Tatiana Krikella

Please Note: This seminar will be held in-person.

Personalized Predictive Model that Jointly Optimizes Discrimination and Calibration

Precision medicine is accelerating rapidly in the field of health research. This includes fitting predictive models for individual patients based on patient similarity in an attempt to improve model performance. We propose an algorithm which fits a personalized predictive model (PPM) using an optimal size of a similar subpopulation that jointly optimizes both model discrimination and calibration, as it is criticized that calibration is not assessed nearly as often as discrimination despite poorly calibrated models resulting in potentially misleading clinical predictions. We empirically show that as the size of subpopulation increases, the calibration of the PPM improves. This is contrary to the relationship between the size of subpopulation and discrimination, which deteriorates as the size of subpopulation increases. This motivates our newly proposed mixture loss function, which is an extension of the Brier Score and is used to tune the size of subpopulation in our algorithm. We also investigate the effect of patient weighting on performance, and conclude, as expected, that the choice of the size of subpopulation has a larger effect on the predictive accuracy of the PPM compared to the choice of weight function. A comprehensive set of simulation studies will be presented to demonstrate our methods and findings.