Design and analysis advances to improve generalizability of clinical trials for treating opioid use disorder

美国的阿片类药物流行是一项突发公共卫生事件，Covid-19大流行加剧了这一情况。阿片类药物使用障碍（mod）的药物治疗——注射纳曲酮、丁丙诺啡和美沙酮——是改善结局和预防OUD患者用药过量的最有效工具，但对mod的使用，尤其是对成功结局通常需要的长期使用，低得令人无法接受。在现实环境中，长期参与率甚至更低——NIDA称之为研究与实践之间的差距。试验与现实世界mod有效性之间的差异可能部分归因于临床试验与现实世界人群特征之间的差异（例如，在精神和物质使用合并症，既往治疗经验，移民身份等方面），如果治疗效果被一些与试验参与相关的特征所改变（增加/减少）。此外，在不了解MOUDs对某些现实世界目标人群的相对有效性的情况下，临床医生、研究人员和政策制定者可能会在有偏见的证据下做出决策。因此，迫切需要提高mod试验的普遍性。如果不能满足这一需求，将进一步加剧研究与实践之间的差距，导致对OUD的整体治疗和关键亚组的治疗不理想。我们建议发展设计和分析方法，我们称之为通线概率性，将mod试验证据与现实世界的人群联系起来。该项目的目标是：在目标1中，识别和描述在美国常规护理机构中寻求OUD治疗的患者的临床意义，可解释的亚组，这些患者在同时基于多个特征的OUD试验中未被代表或代表性不足。这将使我们超越现有的评估代表性的方法，这些方法通常仅限于一次考虑一个个人层面的特征（例如，种族/民族）。我们将把Aim 1第一部分开发的方法应用于试验数据（来自NIDA CTN的3个mod试验）和人口数据（加利福尼亚州和新泽西州的医疗补助申请），以表征代表性不足的亚组。在目标2中，将mod的有效性推广到特定州的成年医疗补助人群，从而估计一个现实的治疗目标，如果治疗保留支持，激励措施和剂量实践得到改进，以与试验相一致。现有的预测广义效应的方法依赖于对非代表性和代表性不足的亚群的外推，这可能导致有偏见和/或信息不足的估计。在目标2的第一部分中开发的方法将进行一些改进，以限制外推和提高效率。在目标3中，在用户友好的软件中实现为目标1和2开发的方法，以方便应用试验人员、研究人员和临床医生的采用。拟议的研究预计将对提高试验参与者的代表性以及了解试验结果如何推广以及向谁推广做出重大贡献。

The opioid epidemic in the US is a public health emergency, exacerbated by the Covid-19 pandemic. Medi- cations for opioid use disorder (MOUD)-injection naltrexone, buprenorphine, and methadone-are the most effective tools for improving outcomes and preventing overdose among persons with OUD, but engagement in MOUD, especially long-term engagement typically required for a successful outcome, is unacceptably low. Long-term engagement rates tend to be even lower in real-world settings-what NIDA has termed the research-to-practice gap. This discrepancy between trial and real-world MOUD effectiveness could be par- tially attributable to differences between clinical trial versus real-world population characteristics (e.g., in terms of psychiatric and substance use comorbidities, previous treatment experience, immigration status, etc.) if treatment effects are modified (increased/decreased) by some of these characteristics that also relate to trial participation. Moreover, without knowing the relative effectiveness of MOUDs for certain real-world target pop- ulations, clinicians, researchers, and policymakers may be tasked with decision-making with biased evidence. Thus, there is a critical need to improve the generalizability of MOUD trials. Failing to meet this need would further ossify the research-to-practice gap, resulting in suboptimal treatment of OUD overall and within key subgroups. We propose to develop design and analytic approaches, what we call a generalizability through- line, to bridge MOUD trial evidence to real-world populations. The objectives of this project are: In Aim 1), to identify and characterize clinically meaningful, interpretable subgroups of persons seeking OUD treatment in US usual-care settings who are not represented or under-represented in MOUD trials based on multiple char- acteristics simultaneously. This will move us beyond existing approaches for assessing representation that have generally been limited to considering one individual-level characteristic at a time (e.g., race/ethnicity). We will apply the approach developed in the first part of Aim 1 to trial data (3 MOUD trials from NIDA CTN) and population data (California and New Jersey Medicaid claims) to characterize under-represented subgroups. In Aim 2), to generalize MOUD effectiveness to state-specific adult Medicaid populations, thereby estimating a realistic treatment goal if treatment retention supports, incentives, and dosing practices were improved to align with those in trials. Existing approaches for predicting generalized effects rely on extrapolation for non- and under-represented subgroups, which can result in biased and/or uninformative estimates. The approach developed in the first part of Aim 2 will make several improvements to limit extrapolation and increase effi- ciency. In Aim 3), to implement the methods developed for Aims 1 and 2 in user-friendly software to facilitate the easy adoption by applied trialists, researchers, and clinicians. The proposed research is expected to make a significant contribution to improving representation among trial participants and to understanding how and to whom trial findings generalize.