Statistical Modeling of Medication and Placebo Effects

描述(由申请人提供)：该项目的目标是开发一种统计工具，以阐明药物、安慰剂和其他相关因素如何影响治疗结果的时间进程。它阐述了RFA的目标，即“开发和应用先进的纵向统计技术，以描述安慰剂反应在酒精临床试验过程中的变化，并探索这些反应轨迹中异质性的来源。”标准的统计学方法，如线性模型，不太适合研究在治疗期间和治疗后影响结果的非常动态的过程。基于关于药物和安慰剂效应的时间进程的理论上合理的假设，我们提出了一个动态的非线性统计模型。该模型使用服药依从性和其他时变信息来预测治疗期间和治疗后的饮酒情况。该模型被应用于来自Project Combine的随机半样本数据，这是一项大型的多部位酒精治疗研究。基于综合主要结果分析的线性模型解释了饮酒随时间变化的6%；非线性模型能够解释87%。非线性模型需要进一步发展，以提高其对结果的时间进程的适合性，并纳入协变量。然后将在合并数据的另一半样本上进行验证。此外，还将使用项目预测的数据再次验证这一点，这是一项与在德国进行的联合项目平行的研究。模拟结果的时间进程虽然在技术上具有挑战性，但它使区分药物、安慰剂和其他对结果的重要影响成为可能。这反过来又可以更清楚地确定药物在哪里成功，在哪里失败。这项研究中开发的工具提供的信息将帮助我们设计更好的干预方案，以改善成瘾障碍患者的健康。 公共卫生相关性：这项研究使用先进的统计方法来梳理药物、安慰剂效果和其他因素如何在治疗期间和治疗后影响饮酒。这有助于研究人员药物开始起作用的速度，以及药物效果逐渐消失的速度，这将帮助我们设计更好的干预组合，以帮助患有成瘾障碍的人。

DESCRIPTION (provided by applicant): The goal of this project is to develop a statistical tool to shed light on how the time course of treatment outcome is affected by medications, placebos, and other relevant factors. It addresses the RFA goal to "Develop and apply advanced longitudinal statistical techniques to describe how the placebo response changes over the course of alcohol clinical trials and explore sources of heterogeneity in these response trajectories." Standard statistical methods such as linear modeling are not well suited for studying the very dynamic processes that affect outcome during and after treatment. Based on theoretically-justified assumptions about the time course of both medication and placebo effects, we propose a dynamic nonlinear statistical model. This model uses medication compliance and other time varying information to predict drinking during and after treatment. This model was applied to a random half-sample of data from Project COMBINE, a large multi-site alcohol treatment study. A linear model based on the COMBINE primary outcome analyses accounted for 6 percent of the variance of drinking over time; the nonlinear model was able to account for 87 percent. The nonlinear model requires further development to improve its fit to the time course of outcome, and to incorporate covariates. It will then be validated on the other half-sample of the COMBINE data. Furthermore, it will be validated again using data from Project Predict, a study parallel to Project COMBINE that was conducted in Germany. Modeling the time course of outcome, while technically challenging, makes it possible to differentiate between medication, placebo, and other important effects on outcome. This in turn makes it possible to ascertain more clearly where medications succeed as well as where they falter. The information from the tool developed in this study will help us to engineer better intervention packages to improve the health of persons with addictive disorders. PUBLIC HEALTH RELEVANCE: This study uses advanced statistical methods to tease apart how medications, placebo effects, and other factors affect drinking during and after treatment. This helps researchers how quickly medications begin to help, as well as how quickly their effects taper off, which will help us engineer better combinations of interventions to help people with addictive disorders.