Statistical methods for regression modeling of global percentile outcome in neurological diseases

项目摘要 神经学和卒中研究的设计和分析由于缺乏单一的主要指标而受到挑战。 结果可以全面评估与之相关的多维度损伤和症状 这种疾病例如，已知帕金森氏病（PD）的个体结果测量，甚至是帕金森氏病（PD）的个体结果测量。 MDS-MRS不能全面捕获PD体征和症状的全部谱。全球每- 百分位结果提供了一种有效和稳定的方式来整合多个单独的结果，提供了一个单一的 全球疾病严重程度的指标。奥布莱恩的全球秩和检验允许两个或K组比较， 全球百分位数的结果，并已成功地应用于许多临床试验，包括神经保护， 帕金森病（NET-PD）长期研究1（LS-1）和FS-ZONE的保护性探索性试验。然而，在这方面， 一直缺乏严格的统计工具来对全球百分位数结果进行回归建模， 系统探索全球疾病负担和全球疾病进展的风险因素。出于 这些挑战和机遇，（目标1）我们提出了一个新的和严格的回归框架，明确 将全球百分位数结果与多个风险因素联系起来， 链接函数和误差分布。我们的估计程序利用信息的行列，以实现 稳健的估计，产生与全球疾病严重程度最一致的风险评分。接下来，（目标 2)我们将开发一个合理的回归框架，用于探索全球百分位数结果的时间趋势 通过纵向数据，专门检测导致全球排名加速上升的风险因素。 我们进一步扩展了我们的方法，以适应常见的辍学机制完全失踪， 随机和随机缺失。此外，（目标3）我们将应用所提出的方法系统地分析- 分析LS-1研究和FS-ZONE中总体疾病严重程度和总体疾病进展的风险因素 study.我们的方法对神经系统疾病和许多其他疾病的研究人员具有实质性的实用价值。 菲尔德。我们将向一般研究界提供所有统计工具的用户友好软件。

PROJECT SUMMARY Design and analysis of neurological and stroke studies have been challenged by the lack of a single primary outcome that can comprehensively assess the multidimensional impairments and symptoms associated with the disease. For example, it is known that individual outcome measures for Parkinson's disease (PD), even the MDS-UPDRS, cannot comprehensively capture the full spectrum of PD signs and symptoms. The global per- centile outcome offers an efficient and stable way to integrate multiple individual outcomes, providing a single metric of the global disease severity. The O'Brien's global rank-sum test allows two or K-group comparisons for the global percentile outcome and has been successfully applied in many clinical trials, including the Neuropro- tection Exploratory Trials in Parkinson's Disease (NET-PD) Long-term Study 1 (LS-1) and FS-ZONE. However, rigorous statistical tools have been lacking for regression modeling of the global percentile outcome, preventing systematic explorations of risk factors for global disease burden and global disease progression. Motivated by these challenges and opportunities, (Aim 1) we propose a novel and rigorous regression framework to explicitly link the global percentile outcome to multiple risk factors, under minimal modeling assumptions regarding the link function and the error distribution. Our estimation procedure exploits information in the ranks to achieve robust estimation, yielding a risk score that is in maximum concordance with global disease severity. Next, (Aim 2) we will develop a sensible regression framework for exploring the time-trend of the global percentile outcome with longitudinal data, to specifically detect risk factors that lead to accelerated progression in global ranks. We further extend our methods to accommodate the common dropout mechanisms of missing completely at random and missing at random. Furthermore, (Aim 3) we will apply the proposed methods to systematically an- alyze risk factors of global disease severity and global disease progression in the LS-1 study and the FS-ZONE study. Our methods bear substantial practical utility for researchers in neurological diseases and many other fields. We will provide user-friendly software for all statistical tools to the general research community.