SEMIPARAMETRIC BAYESIAN METHODS FOR SURVIVAL DATA

生存数据的半参数贝叶斯方法

• 批准号: 2113306
• 负责人: DEBAJYOTI SINHA
• 金额: $ 6.86万
• 依托单位: UNIVERSITY OF NEW HAMPSHIRE
• 依托单位国家: 美国
• 财政年份: 1995
• 资助国家: 美国
• 起止时间: 1995-08-15 至 2000-07-31
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/2113306
• 关键词: epidemiology; human data; mathematical model; model design /development; statistics /biometry

DESCRIPTION (Adapted from Applicant's Abstract): The proposed research is in biostatistical methods associated with event time data (data on survival times, detection times of carcinogenic growths, recurrence times of symptoms, etc.). The objective is to develop additional statistical models for such data, along with associated methods of statistical analysis. The focus will be on semiparametric Bayesian models and methods. The semiparametric nature allows considerable generality and applicability but enough structure for useful physical interpretation and understanding for particular applications in medical research. Recent advances in Bayesian theory and computations make the study of complex models and data structures feasible. Four categories of event time data will be considered: univariate survival data (survival times for a group of unrelated patients), multiple event time data (successive repeated events in each of a group of unrelated patients), multivariate survival data (survival times for a group of patients who are related to each other either genetically or environmentally), two independent events in tandem to every patient (such as infection and onset of disease). Each model considered has been subjected to some kind of censoring mechanism, such as right censoring, grouping (due to inexact measurement or periodic followup), interval censoring (due to missed followups). Monte Carlo algorithms, including data augmentation and Gibbs sampling, will be used to deal with the complexity of the model as well as the censoring or grouping present in the data. The research method includes mathematical modeling, mathematical developments of statistical methods, writing of computer algorithms, and exemplification by reanalysis of published data sets from cancer and other medical studies and animal experiments with carcinogens. Models and methods developed in this research should lead to an improved understanding of event time data occurring in clinical research in cancer and other diseases--including the relation of event times to various risk factors, and quantification of group (familial) dependencies.

描述（改编自申请人摘要）：拟议的研究 是在与事件时间数据（关于 存活时间、致癌生长的检测时间、复发 症状等）。 目标是开发更多的 这些数据的统计模型，沿着相关的方法， 统计分析 重点将放在半参数贝叶斯 模型和方法。 半参数性质允许相当大的 但对于有用的物理结构来说， 医学中特殊应用的解释和理解 research. 贝叶斯理论和计算的最新进展使得 研究复杂的模型和数据结构是可行的。 四类 将考虑事件时间数据：单变量生存数据 （一组无关患者的生存时间），多事件时间 数据（在一组不相关的 患者），多变量生存数据（一组患者的生存时间， 彼此有血缘关系的患者， 环境），两个独立的事件串联到每一个病人 (such如感染和发病）。 每一个被考虑的模型都受到某种审查 机制，如右删失、分组（由于测量不准确 或定期随访）、间隔删失（由于错过随访）。 蒙特卡罗算法，包括数据扩充和吉布斯抽样， 将用于处理模型的复杂性以及 数据中存在的删失或分组。 研究方法包括数学建模、数学建模、 统计方法的发展，计算机算法的编写， 通过重新分析已发表的癌症数据集， 其他医学研究和致癌物的动物实验。 模型 在这项研究中开发的方法应该会导致一个改进的 了解临床研究中发生的事件时间数据， 癌症和其他疾病--包括事件时间与 各种风险因素和组（家族）的量化 个依赖项