Optimizing treatment decision by accounting for longitudinal biomarker trajectories and competing risks of each individual

点击翻译按钮获取中文摘要

Project Summary/Abstract The goal of this proposal is to develop statistical methods for evaluating treatment strategies at different time points and identifying optimal treatment strategies on the basis of patients' longitudinal biomarker measurements. It is motivated by our research on identifying the best timing for patients with chronic myeloid leukemia (CML) to receive a stem cell transplant (SCT). SCT can cure leukemia, but it is associated with life- threatening risks. For this reason, most patients start with other less-aggressive treatment options that are much safer but cannot cure the disease. Thus the decision-making about optimal timing of SCT depends on a patient's disease progression. However, it is infeasible to conduct a randomized controlled trial to weigh the risks and benefits of SCT at various times. To optimize this decision-making process, sophisticated and comprehensive statistical models are needed to provide an accurate estimation of the benefits and risks (and their trade-offs) over time for patients under different SCT timing options. However, these have not yet been developed, due to the challenges elaborated below. First, the question of an optimal decision on SCT cannot be answered by a single statistical model, it requires assembling information from a series of models and analyses. Second, there most likely is not a uniform solution for this question, as the optimal timing of SCT depends on each individual's disease progression status. Consequently, physicians must use patients' longitudinal biomarker trajectories to monitor their health status and make treatment decision in a dynamic fashion. Third, the treatment decision for each individual must account for their competing risks, including death by treatment-related complications and other causes (e.g., heart diseases and diabetes). Finally, it is impossible to implement optimal decision-making without an easy-to-use software. The following specific aims are proposed to solve these problems. Aim 1: Use functional component principal component analysis (FPCA) techniques to fully capture the dominant patterns from patients' longitudinal biomarker trajectories, and use them as predictors of patients’ risk of disease progression. Aim 2: Estimate dynamic competing risks based on baseline covariates and longitudinal biomarker trajectories using multi-state models. Aim 3: Use analytic and microsimulation approaches to estimate and compare the mean survival times under different SCT timing options. Aim 4: Conduct validation studies, develop software, and broaden application. Three CML studies will be used to cross-validate each other regarding the optimal timing of SCT. Software programs with user-friendly interfaces will be made publicly available. The proposed statistical and software programs will be adapted and applied to a study of kidney disease to test their broad application.