Innovative Analytic Methods of Person-Centered Data and Adaptive Designs for Alco

描述(由申请人提供)：本申请涉及广泛的挑战领域(05)：比较有效性研究和特定挑战主题：05-AA-102酒精治疗研究的适应性设计和以人为中心的数据分析。正如挑战主题所暗示的那样，在许多酒精依赖的临床研究中，使用以变量为中心的方法(例如，均值比较)的统计分析是不够的。例如，统计假设(例如正态分布)经常被违反。此外，这种方法不能充分解释饮酒结果的可变性。同样，比较治疗和安慰剂的简单试验往往不能回答对临床医生特别重要的问题，他们必须根据最初和随后治疗的反应在同一患者身上做出一系列决定。最近，人们对使用各种药理制剂作为社会心理治疗的有希望的辅助手段以减少酒精消费的兴趣和研究大大增加。其中许多研究使用回溯法收集以个人为中心的饮酒数据，例如时间线回溯法来回忆和记录过去一周的日常饮酒结果。然后对这些日常饮酒记录进行总结和分析。例如，约翰逊等人。(2003)在治疗评估期浓缩了每日饮酒记录，而Johnson等人。(2007)以每周的形式浓缩了它们。然而，这种浓缩的结果并不像最初的每日饮酒记录那样提供信息。此外，这些结果通常被认为是正常的，这可能会被违反。第三，在这些分析中不能完全捕捉饮酒结果的轨迹。在这项拨款提案中，我们将开发新的统计方法来分析酒精治疗研究中以人为中心的数据。首先，我们将使用原始的每日饮酒量作为响应变量，因此我们的方法比使用浓缩结果的方法更有效。其次，我们通过两部分模型(下午2点)来处理日常饮酒结果的非正态分布，以分别描述每天饮酒为零的几率和饮酒一天中的实际饮酒量。我们还提出了新的方法来解决正的日常饮酒水平中的偏态和可能的异方差。第三，我们对饮酒结果随时间变化的轨迹感兴趣，以更好地捕捉结果的差异。我们将使用参数和半参数方法(例如，样条线)来描述这样的时间模式。在许多简单的试验中，我们比较两个手臂，通常是治疗手臂和对照手臂，以确定干预的效果。然而，当我们对从一系列剂量中确定最佳剂量感兴趣时，这样的研究是不够的。例如，约翰逊等人。(2003,2007)在托吡酯试验中使用了剂量递增方案(从25毫克增加到300毫克)。在这些概念验证试验中，他们确定了托吡酯治疗在改善饮酒结果方面的整体效果。然而，托吡酯在不同剂量水平的效果仍有待确定，以便我们能够确定具有满意疗效的最佳剂量，同时将不良事件的发生率降至最低。适应性设计可以提供一种潜在的解决方案。适应性设计背后的动机是结合序贯设计的统计优势和伦理要求，即根据先验知识和当前积累的数据，以被判断为最佳的剂量治疗尽可能多的患者。在此背景下，持续重新评估方法(CRM)为使用具有某些最佳特性的工作统计模型开辟了新的领域(O‘Quigley，Pepe和Fisher，1990)。将提出寻找最成功剂量(MSD)的新方法，以经济有效的方式确定最佳剂量。 公共卫生相关性(由申请者提供)：在这项拨款提案中，我们将提出并应用几个创新模型来分析酒精治疗研究中以人为中心的数据。我们还将引入新的适应性设计，以经济高效的方式确定最佳剂量。我们预计，这项研究的完成将加快比较酒精依赖研究中不同治疗方法的有效性的进程。

DESCRIPTION (provided by applicant): This application addresses broad Challenge Area (05): Comparative Effectiveness Research and specific Challenge Topic: 05-AA-102 Adaptive Designs and Person-Centered Data Analysis for Alcohol Treatment Research. As the challenge topic implies, statistical analyses using variable-centered approaches (e.g., comparison of means) are insufficient in many clinical studies of alcohol dependence. For example, statistical assumptions (e.g., normality) are routinely violated. Also, such methods can not adequately account for variability in drinking outcome. Similarly, simple trials comparing a treatment and a placebo often do not answer questions of particular import to clinicians, who have to make a series of decisions in the same patient based upon response to initial and subsequent treatment. Recently, there has been substantially increased interest in - and research on - the use of various pharmacological agents as promising adjuncts to psychosocial treatment to reduce alcohol consumption. Many of these studies collected person-centered drinking data using retrospective method, e.g., the timeline follow-back method to recall and record the daily drinking outcome for the past week. These daily drinking records were then summarized and analyzed. For example, Johnson et al. (2003) condensed the daily drinking records in the treatment assessment periods, while Johnson et al. (2007) condensed them in the weekly format. However, such condensed outcomes are not as informative as the original daily drinking record. Also, normality is often assumed for these outcomes, which could be violated. Third, the trajectory of the drinking outcome can not be fully captured in these analyses. In this grant proposal we will develop new statistical methods to analyze person-centered data for alcohol treatment research. First, we will use the original daily drinking level as the response variable, thus our method is more efficient than those using the condensed outcomes. Second, we tackle the non-normality of the daily drinking outcome by two- part models (2PM) to separately describe the odds of daily drinking being zero and the actual number of drinks in a drinking day. We also propose new methods to tackle the skewness and possible heteroscedasticity in the positive daily drinking level. Third, we are interested in the trajectory of the drinking outcome over time to better capture differences in outcomes. We will use both parametric and semiparametric methods (e.g., splines) to describe such temporal patterns. In many simple trials, we compare two arms, often a treatment arm and a control arm, to determine the efficacy of the intervention. However, such studies are insufficient when we are interested in determining the optimal dose from a range of doses. For example, Johnson et al. (2003, 2007) used a dose escalation scheme (from 25 mg to 300 mg) in the topiramate trials. In these proof of concept trials, they established the overall topiramate treatment effect at improving drinking outcomes. However, the topiramate effect at different dose levels remains to be ascertained so that we can identify the best dose which has the satisfactory efficacy while minimizing the rate of adverse events. Adaptive designs can offer a potential solution. The motivation behind adaptive designs is to bring together the statistical advantages of a sequential design with the ethical imperative of treating as many patients as possible at a dose judged to be the best, in the light of prior knowledge and the current accumulated data. In this context, the continual reassessment method (CRM) opened up the field to the use of working statistical models which have some optimal characteristics (O'Quigley, Pepe and Fisher, 1990). New methods to find the most successful dose (MSD) will be proposed to identify the optimal dose in a cost-effective way. PUBLIC HEALTH RELEVANCE (provided by applicant): In this grant proposal we will propose and apply several innovative models to analyze the person-centered data for alcohol treatment research. We will also introduce new adaptive designs to identify optimal dose in a cost-effective way. We expect that the completion of this study will speed the process of comparing effectiveness of different treatments in alcohol dependence studies.