Contributions to estimation of lifetime risk and of biomarker trajectory before a right-censored event
Embargo Date
2027-03-27
OA Version
Citation
Abstract
In this dissertation, we addressed three common yet often overlooked issues in statistical and biomedical research studies and proposed novel approaches with the goal of advancing estimation and inference techniques. In simulation studies aiming at neutral comparisons of competing statistical tests, comparing test performance based on power alone may be misleading, particularly if the size of the tests being compared are either different from one another or from the nominal size. We proposed to use relative likelihood ratios to factor in both power and size in the comparison of multiple tests. We derived sample size formulas for a comparative simulation study. We illustrated the proposed method by comparing multiple statistical tests for small-study effects in meta-analyses of randomized controlled trials. We demonstrated that the proposed likelihood ratio approach enables an accurate comparison of the trade-off between power and size between competing tests. We proposed to use the inverse probability of censoring weighting technique to address potential immortal time bias in studies assessing the association between polygenic risk and incident disease. We combined the proposed method with the pseudo-observation framework. We compared the performance of the proposed approach with alternative statistical methods in simulation studies. We applied our proposed method to lifetime risk of atrial fibrillation (AF) with data from the Framingham Heart Study (FHS). Our simulation results demonstrated scenarios where the proposed method is recommended to account for immortal time bias. In recent literature, there is an increasing interest in studying the evolution of a biomarker over time and its association with the occurrence of an event of interest subject to right censoring. We proposed an extension of the "backward looking and estimation" method, in which we compare the backward trajectory to a "baseline" trajectory. We use the difference curve and its confidence band to identify signature biomarker trajectories leading up to the event. In a small simulation study, we examined the performance of the proposed approach. We illustrated the approach with data from FHS for AF. We discussed the advantages and limitations of the proposed test and the "backward looking and estimation" method.
Description
2024