Understanding CNS Stimulant Use and Safety in Veterans with TBI

项目总结/摘要 背景：使用中枢神经系统（CNS）兴奋剂，如安非他明、苯丙胺、 哌甲酯、阿莫达非尼和莫达非尼不鼓励用于创伤性脑损伤（TBI）患者，因为它们 没有证实的好处，并进行了食品和药物管理局黑盒警告的风险更高的发展物质的使用 疾病（SUD）。 意义：创伤性脑损伤是退伍军人发病率和死亡率的主要来源，也是HA/ORD/HSR&D的首要任务。 我们的初步数据表明，截至2021年5月，全国范围内有728，065名退伍军人在他们的工作中被诊断为TBI。 电子病历TBI登记处估计，81%的退伍军人患有轻度TBI。患有TBI的退伍军人更有可能 接受中枢神经系统兴奋剂的人比没有TBI的人多。我们的初步数据表明，5.8%（42，437/728，065）的 TBI的退伍军人接受了中枢神经系统兴奋剂的处方，这是10倍以上， 无TBI的退伍军人（0.56%）。我们的初步研究结果还表明，与非使用者相比， 中枢神经系统兴奋剂，用户有较高的风险SUD。目前，没有循证疗法用于治疗 轻度TBI和VA轻度TBI指南不鼓励使用药物来改善神经认知功能 症状然而，许多患有TBI的退伍军人接受CNS兴奋剂的处方，但对 这些药物在TBI退伍军人中的安全性。 创新与影响：据我们所知，研究问题以前从未得到过回答。 这项研究的关键创新在于填补了科学知识的空白，潜在的临床意义。 调查结果的影响，以及与退伍军人群体的相关性。我们的方法创新将 包括深度机器学习方法的使用，包括所开发的影响和交互评分 来量化深度学习的结果。 具体目的：1）描述轻度TBI退伍军人的兴奋剂处方模式; 2）测试 假设开始兴奋剂治疗与SUD事件、住院治疗 和患有轻度TBI的退伍军人的死亡率;和3）开发可解释的机器（深度）学习风险预测 该模型将允许更准确和精确地评估兴奋剂的临床获益与风险， 个别退伍军人。 方法：这些目标将通过使用VA TBI登记和EHR数据来实现。对于目标1和3，我们 将使用所有退伍军人与创伤性脑损伤诊断和任何使用兴奋剂。对于目标2，我们将模拟 RCT，使用无SUD的TBI退伍军人和轻度TBI后CNS兴奋剂的新处方 诊断.然后，我们将在轻度TBI的退伍军人亚组中进行敏感性分析， 全面TBI评估（CTBIE）表。倾向评分匹配将用于结局设盲 根据测量的协变量平衡的队列集合，敏感性分析将用于估计影响 无法测量的混杂因素。医疗保险和医疗补助服务中心（CMS）数据将用于验证 我们的预测模型的可推广性。 下一步/实施：除了通过介绍和 出版物，我们将分享我们的软件工具与VA AI中心的传播，并与我们的 业务合作伙伴将拟议项目的结果纳入临床医生教育材料。

Project Summary/Abstract Background: The use of central nervous system (CNS) stimulants such as amphetamine, dextroamphetamine, methylphenidate, armodafinil and modafinil is discouraged in patients with traumatic brain injury (TBI) as they have no proven benefits and carries the FDA black-box warning of a higher risk for developing substance use disorders (SUD). Significance: TBI is a major source of morbidity and mortality for Veterans, and a top HA/ORD/HSR&D priority. Our preliminary data suggest that as of May 2021, nationwide 728,065 Veterans had a diagnosis of TBI in their EHR. The TBI registry estimates that 81% of the Veterans have mild TBI. Veterans with TBI are more likely to receive CNS stimulants than those without TBI. Our preliminary data suggests that 5.8% (42,437/728,065) of the Veterans with TBI received prescriptions for CNS stimulants, which is over 10 times higher than that in Veterans without TBI (0.56%). Findings of our preliminary study also suggest that compared to non-users of CNS stimulants, users have a higher risk of SUD. Currently, there is no evidence-based therapy for treatment of mild TBI and the VA mild TBI guidelines discourages the use of medications to ameliorate neurocognitive symptoms. However, many Veterans with TBI receive prescriptions for CNS stimulants but less is known about the safety of these drugs in Veterans with TBI. Innovation & Impact: To the best of our knowledge, the study questions have never been answered before. The key innovation of the proposed study is in the filling of the scientific knowledge gap, the potential clinical implications of the findings, and the relevance to the Veteran population. Our methodological innovation will include the use of deep machine learning approaches including the impact and the interaction scores developed by our team to quantify the results of deep learning. Specific Aims: 1) To characterize stimulant prescription pattern in Veterans with mild TBI; 2) To test the hypothesis that initiation of stimulant therapy is associated with a higher risk of incident SUD, hospitalization, and mortality in Veterans with mild TBI; and 3) To develop an explainable machine (deep) learning risk prediction model that will allow a more accurate and precise assessment of clinical benefits vs. risk of stimulants in individual Veterans. Methodology: These aims will be achieved by using the VA TBI registry and EHR data. For Aims 1 and 3, we will use all Veterans with a TBI diagnosis and any use of stimulants. For Aim 2, we will emulate the design of an RCT, using Veterans with TBI free of prevalent SUD and new prescriptions of CNS stimulants after mild TBI diagnosis. We will then conduct sensitivity analysis in the subset of Veterans with mild TBI using the Comprehensive TBI Evaluation (CTBIE) tables. Propensity score matching will be used for outcome-blinded assembly of cohorts balanced on measured covariates, and sensitivity analyses will be used to estimate impact of unmeasured confounders. Centers for Medicare & Medicaid Services (CMS) data will be used to validate the generalizability of our prediction model. Next Steps/Implementation: In additions to the traditional dissemination approaches through presentations and publications, we will share our software tools with the VA AI center for dissemination and work with our operational partners to incorporate the findings of the proposed project into clinician education materials.