On inferring cognitive impairment from a battery of tests and predicting an event-of-interest using longitudinal and time-to-event data.
Access changed 1/8/24.
In this dissertation we first investigated two commonly used methods’ and a recently proposed method’s ability to predict conditional survival probabilities based on longitudinal biomarker measurements. Then, in preparation for the application of such dynamic prediction methods, we proposed the use of Bayesian multivariate mixture model accounting for censoring to infer cognitive impairment status at baseline from a battery of cognitive tests in the presence of censoring. The currently used methods for inferring cognitive impairment from a battery of cognitive tests and the proposed method were applied to the task of inferring cognitive impairment from a battery of cognitive tests administered to pediatric Multiple Sclerosis patients. The impact of censoring on the inferred cognitive impairment status was examined, as well as the predictive accuracy of the proposed and currently used methods via simulation. Finally, in order to infer cognitive impairment based on results from a battery of cognitive tests obtained during follow-up in the presence of a practice effect, we proposed the use of a Bayesian continuous-time mixed hidden Markov model. To account for the practice effect in the hidden Markov model, we proposed incorporating an adapted form of the half-life regression equation presented by Settles and Meeder. We simulated two example datasets based on the results from the analysis of the battery of cognitive tests at baseline due to the lack of longitudinal cognitive testing data from pediatric Multiple Sclerosis patients and applied the hidden Markov model to infer cognitive impairment. We examined the ability of the hidden Markov model to correctly infer cognitive impairment at each time point during follow-up for the simulated patients as well as the accuracy of parameter estimates. The predictive accuracy of the proposed methods in simulated and real data obtained from pediatric Multiple Sclerosis patients enable us to efficiently design clinical studies and trials aimed at improving the understanding and treatment of cognitive impairment in this patient population, as well as other diseases in which cognitive impairment is a known effect.