Drug and Comorbid Trajectories: Jointly-Modeled, Differently-Distributed Outcomes

说明(由申请人提供)：鉴于充分的研究表明多种药物的并存率极高，了解多种药物成瘾之间的相互关系以及同时发生的药物使用和精神障碍之间的相互关系是NIDA的一个关键优先事项。现有分析方法的局限性阻碍了在这些领域的进一步进展，因为描述药物使用和共生疾病的轨迹随着时间的推移是复杂的--通常是非线性的，并且基于具有不同分布的结果。时间线跟踪(TLFB)是确定药物使用结果的最广泛和被接受的衡量标准，它收集特定时间间隔内多种药物的日常使用报告(例如，使用大麻，是/否；每天的关节)。但TLFB数据并未按收集的方式进行分析(例如，重复的二项式和泊松变量)。相反，数据通常被压缩成试验期间的总使用天数(或禁欲天数)之类的摘要，或者在较小的间隔内(例如每月使用的天数总和)折叠，以试图创建正态分布的变量。这样的综合分数牺牲了信息，失去了效率，往往是不正常的。存在大量先前收集的具有不同分布的结果的纵向数据，例如来自TLFB的结果，但无法如此分析它们阻碍了令人满意地解决与科学和临床相关的问题，例如：使用两种或两种以上药物之间的时间关系是什么？药物使用的变化和共生疾病的变化之间的时间关系是什么？共同出现的精神症状是否随着药物使用的减少而缓解，或者精神症状的减少是否先于药物使用的减少？在什么情况下，一项指标相对于另一项指标会出现下降？该项目的前两个目标提出了严格的理论推导和模拟，以开发多变量非线性混合模型(MvNLMIXED)，以同时估计和比较药物使用和共病结果的非线性轨迹，这些结果随着时间的推移和组间的不同分布(目标1)，并通过估计它们之间的关联(目标2a)和一个相对于另一个的变化的顺序和时间滞后来评估这些联合建模的轨迹之间的相互关系(目标2b)。AIMS 3和4将应用MvNLMIXED方法回答两项针对同时接受药物使用认知行为治疗的青少年注意力缺陷多动障碍(ADHD)的药物治疗试验中的重要问题：一项是托莫西汀试验，另一项是NIDA临床试验网络中的一项渗透释放哌酸甲酯的多点试验。这些新颖的分析有望确定不同药物相互使用(目标3)和ADHD(目标4)之间的重要相互关系，潜在地告知治疗可能是如何进行的。除了它们对评估这两个试验中的相互关系以及一般药物使用和共病之间的重要性之外，这些新方法还广泛适用于评估任何联合建模的纵向变量之间的时间关系，有利于新的研究和提高现有数据库的科学价值。这些方法的用户友好软件将在多个平台上提供。

DESCRIPTION (provided by applicant): Understanding the inter-relationships among addiction to multiple drugs and between co-occurring substance use and psychiatric disorders is a key priority of NIDA's, given ample research showing extremely high rates of their co-morbidity. Limitations of existing analytic methods impede further progress in these areas, because trajectories describing drug use and co-occurring disorders over time are complex--often nonlinear and based on outcomes with different distributions. Timeline Follow back (TLFB), the most widely utilized and accepted measure for determining drug use outcomes, collects reports of daily use (e.g. used marijuana, yes/no; joints per day) on multiple drugs over specified intervals. But TLFB data are not analyzed as collected (e.g. repeated binomial and Poisson variables). Rather, data are typically collapsed into summaries like total days of use (or abstinence) during a trial, or collapsed over smaller intervals (e.g. monthly sums of days used) in an attempt to create normally distributed variables. Such composite scores sacrifice information, lose efficiency, and are often not normal. Abundant previously collected longitudinal data on outcomes with different distributions like those from TLFB exist, but an inability to analyze them as such prevents satisfactorily addressing scientifically and clinically relevant questions such as: What is the temporal relationship between use of two or more drugs? What is the temporal relationship between change in drug use and change in co-occurring disorders? Do co- occurring psychiatric symptoms remit with reductions in drug use, or do reductions in psychiatric symptoms precede reductions in drug use? At what point does reduction in one occur relative to the other? This project's first two aims propose rigorous theoretical derivation and simulation to develop a multivariate nonlinear mixed model (MvNLMIXED) to simultaneously estimate and compare nonlinear trajectories of substance use and comorbidity outcomes with different distributions over time and between groups (Aim 1), and to evaluate inter- relationships among those jointly modeled trajectories by estimating their association (Aim 2a) and the order of, and time-lag among, change in one relative to the other (Aim 2b). Aims 3 and 4 will apply MvNLMIXED methods to answer important questions in two pharmacotherapy trials for co-occurring Attention-Deficit Hyperactivity Disorder (ADHD) in adolescents who also received cognitive behavioral therapy for drug use: a trial of atomoxetine and a multi-site trial o Osmotic-Release Methylphenidate in NIDA's Clinical Trials Network. These novel analyses are expected to identify important inter-relationships among use of different drugs with each other (Aim 3) and with ADHD (Aim 4), potentially informing how treatment may be operating. Beyond their importance for evaluating inter-relationships in these two trials and among drug use and comorbidity in general, these novel methods are widely applicable to evaluating temporal relationships among any jointly modeled longitudinal variables, benefiting new research and enhancing the scientific value of existing databases. User-friendly software for these methods will be made available on multiple platforms.