Population Pharmacokinetics/Dynamic: Statistical Issues

描述（由申请人提供）：本提案要求资助继续开展旨在开发统计模型和方法的研究，以帮助医学肿瘤学家找到用药物治疗癌症患者的最佳方法。了解患者间异质性的许多来源是个体化治疗的关键。由于大多数癌症化疗的治疗窗口相对较窄，因此对癌症患者进行个体化治疗尤为重要。我们提出的统计模型将有助于发现患者对治疗反应的变异来源，从而实现个体化抗癌化疗。这项应用的总体假设是，了解抗癌药物的药代动力学（PK）、药效学（PD）和药物遗传学（PGx）对于评估疗效和确定如何在临床上最好地使用这些药物非常重要。新的统计方法将有助于更有效地综合这些信息，以改善癌症患者的治疗效果。我们的建议有四个具体目标。(1)发展决策理论方法，促进剂量个体化。我们的方法结合了患者特定的PK信息和其他患者在一定剂量下的PK数据，以帮助确定当前患者的最佳剂量。(2)建立连贯的统计模型，了解PK摘要与相关snp的依赖性。这些方法将通过药物的PK、代谢物和相关基因型信息的联合模型促进发现。(3)针对重复循环内嵌套的重复数据建立贝叶斯非参数模型。这些数据通常出现在临床研究中，其兴趣可能涉及研究跨周期的患者内部相关性结构或了解影响结果的患者特异性特征，同时考虑重复重复的测量结构。(4)构建同时建模PK和PD数据的框架，将PK和PD响应视为功能响应，而不是专注于少数低维总结。我们将开发一个联合概率模型，该模型将对函数值数据进行回归，将PK作为解释变量，并将纵向PD响应作为结果。潜在的方法论主题是有效地利用在研究过程中可以收集到的所有数据，用适当传播不确定性的模型对未知量进行联合推断，并试图将感兴趣的科学问题作为统计推断问题，并在适用的情况下作为决策问题。

DESCRIPTION (provided by applicant): This proposal requests funding to continue research aimed at developing statistical models and methods to help medical oncologists find the best way to treat cancer patients with pharmaceutical agents. Understanding the many sources of between-patient heterogeneity is key to individualizing therapy. Individualizing therapy for cancer patients is particularly important, because of the relatively narrow therapeutic window of most cancer chemotherapy. Our proposed statistical models will aid the discovery of sources of variation in a patient's response to therapy, allowing individualizing anticancer chemotherapy. The overall hypothesis of this application is that understanding the pharmacokinetics (PK), pharmacodynamics (PD), and pharmacogenetics (PGx) of anticancer agents is important for evaluating efficacy and determining how best to use such agents clinically. New statistical methodology will help synthesize this information more efficiently to improve treatment outcomes for cancer patients. Our proposal has four specific aims. (1) Develop a decision-theoretic approach to facilitate dose individualization. Our approach combines patient-specific PK information with PK data from other patients treated at a range of doses to help determine an optimal dose for the current patient. (2) Build coherent statistical models to learn about the dependence of PK summaries and related SNPs. The methods will foster discovery through joint models for PK of a drug, its metabolites, and related genotype information. (3) Develop Bayesian nonparametric models for repeated data nested within repeating cycles. These data often arise in clinical studies, where interest may concern studying the within-patient correlation structure across cycles or learning about patient-specific characteristics that influence outcomes while accounting for the repeated-repeated measurement structure. (4) Construct a framework to model PK and PD data simultaneously, considering PK and PD responses as functional responses, rather than focusing on a few low-dimensional summaries. We will develop a joint probability model that will implement regression for function-valued data, considering PK as the explanatory variable and the longitudinal PD response as the outcome. Underlying methodological themes are the efficient use of all data that can be collected in the course of a study, joint inference on unknown quantities with models that appropriately propagate uncertainties, and an attempt to cast the scientific questions of interest as statistical inference questions and, where applicable, as decision problems.