Analyzing Complex Cancer Studies With Skewed Responses

 描述(由申请者提供)：许多精子癌研究的结果是高度不对称的。此外，数据集中在肿瘤学家、诊所或医院内，结果往往是正确的或间隔审查的，而且有大量感兴趣的预测因素。由于结果的偏斜性，作为协变量函数的结果的中位数和分位数是有意义的。目前可用于处理的文献非常有限，尤其是统计模型和对聚集的偏态响应数据的分析。在这里，为了分析这些数据，我们提出了五个目标的方法，这些方法将对临床和生物统计学科学以及未来的癌症研究产生很大影响。具体地说，这四个目标是：1)高度偏斜的聚类结果(删失和非删失)的量化回归；2)具有对数线性中值的区间删失数据的方法；3)估计高度偏斜混合反应数据(包括零膨胀模型)分位数上的协变量效应；4)当协变量较多时，对偏斜反应的估计和预测。另一个目标是让非统计学家广泛使用新开发的统计/流行病学方法。对于每个目标中描述的方法，我们计划创建可与现有的、广泛使用的统计程序包(例如SAS和R)一起使用的宏和程序。统计宏和程序将在我们的网站上提供，以及关于如何将这些宏应用于结果出版物中分析的例子的文件。我们建议的方法是专门为回答重要的临床问题而开发的，我们的临床合作者需要这些问题来发表未来的临床论文。

 DESCRIPTION (provided by applicant): Many of the outcomes in seminal cancer studies are highly skewed. Moreover, the data are clustered within oncologist, practice, or hospital, and often the outcomes are right or interval censored, and there are a large number of predictors of interest. Because of the skewness in the outcomes, medians and quantiles of the outcome as a function of covariates is of interest. There is very limited current literature available to deal particularly with statistical models and analysis of clustered skewed response data. Here, to analyze such data, we propose methods in five aims that will have a high impact on clinical and biostatistical sciences and future cancer studies. In particular, the four aims are: 1).Quantile regression for highly skewed clustered outcomes (censored and not censored); 2) Methods for interval-censored data with a log-linear median; 3) Estimating covariate effects on quantiles of highly-skewed mixed response data (including zero-inflated type models); 4) Estimation and prediction for skewed responses when there are a large number of covariates. An additional goal is to make the newly developed statistical/epidemiological methodology widely accessible to nonstatisticians. For the methods described in each aim, we plan to create macros and procedures which can be used with existing, widely-used statistical packages (e.g., SAS and R). Statistical macros and procedures will be made available on our website, together with documentation on how to apply these macros to the examples analyzed in the resulting publications. The approaches we propose are specifically developed to answer important clinical questions that our clinical collaborators need to publish future clinical papers.