Precision Pharmacogenomic Perioperative Prediction

背景：退伍军人外科质量改进计划(VASQIP)预测重要的 术后结果和分享来自高性能部位和低性能部位的过程改进 绩效网站，以不断改善手术结果。VASQIP非常成功，以至于 在私营部门实施，今天仍在继续。拟议的研究将增加药物基因组学 从百万退伍军人计划(MVP)到VASQIP的数据。此外，VASQIP正在与 退伍军人事务部国家人工智能研究所(NAII)将从其他退伍军人数据库中添加更多表型数据 包括VASQIP、集中交互表现学资源(密码)、退伍军人信息学和计算 基础设施(芬奇)和企业数据仓库。此表型数据也将添加到 VASQIP和机器学习/人工智能将在一个单独的项目中用于更新VASQIP 这将是同时进行的。 意义：药物基因组学检查个人影响药物新陈代谢的基因 靶向、药物转运或药物免疫反应以及对药物不良事件和治疗的影响 有效性。药物基因组学可以解释在治疗反应中常见的变化 临床实践。药物基因组学一直与更差和更好的结果和成本有关 在许多临床环境中的有效性。药物基因组数据包括在499个FDA药品标签上。 尽管药物基因组测试的好处得到了承认，但这种测试并不是常规的 一般在退伍军人事务部内完成，而不是专门用于手术。 创新与影响：对于拟议的研究，有几种创新的方法。施药 药物基因组数据到手术结果，使用机器学习和人工智能来添加 VASQIP计划的表型数据，目标是将结果快速实施到患者护理中 优化以患者为中心的决策，结果都是创新的。 具体目标：1)在接受药物治疗的个体中确定药物基因组风险与结果的关联 过去10年血管外科和心脏外科已建立的(第1级和第2级)药物/基因集。2) 在接受血管手术的个体中确定药物基因组风险与结果的相关性 过去10年心脏手术中未确定的(第3级)药物/基因组合。3)评估学习频率 用于未来研究的药物使用和药物基因组基因的存在。4)识别高风险 可能受益于药物基因组测试的亚组。 方法：这是一项回顾性队列研究，将使用标准的VASQIP变量和结果。 基线分析将使用线性回归或Cox比例风险模型，并将对患者进行控制 基线特征和手术因素，使用匹配或反向加权的倾向性分数。 机器学习方法，如人工神经网络、分类和回归树或集成 学习将被用来改进预测，并解释下列因素之间的非线性关系和相互作用 潜在的大量药物基因组学特征。 下一步/实施：拟议研究的结果将用于更新所有VASQIP 手术，然后可以进行现场实施，以提供实时临床决策支持。高危患者 将确定将受益于术前药物基因组测试的亚组，并进一步 干预研究。

Background: The VA Surgical Quality Improvement Program (VASQIP) predicts risk for important postoperative outcomes and shares process improvements from high performance sites with lower performance sites to continuously improve surgical outcomes. The VASQIP was so successful it was implemented in the private sector and continues today. The proposed research will add pharmacogenomic data from the Million Veterans Program (MVP) to the VASQIP. In addition, the VASQIP is collaborating with the VA National Artificial Intelligence Institute (NAII) to add more phenotype data from other VA databases including VASQIP, Centralized Interactive Phenomics Resource (CIPHER), VA Informatics and Computing Infrastructure (VINCI), and the Corporate Data Warehouse. This phenotype data will also be added to the VASQIP and machine learning/ artificial intelligence will be used to update the VASQIP in a separate project that will be done in parallel. Significance: Pharmacogenomics examines an individual person’s genes that affect drug metabolism, drug target, drug transport, or drug immune response and the impact on adverse drug events and treatment effectiveness. Pharmacogenomics can explain the variation in treatment response that is commonly seen in clinical practice. Pharmacogenomics has been associated with both worse and improved outcomes and cost effectiveness in a number of clinical settings. Pharmacogenomic data is included on 499 FDA drug labels. Despite this acknowledgement of the benefits of Pharmacogenomic testing, such testing is not routinely completed within the VA in general, and not for surgery specifically. Innovation & Impact: There are several innovative approaches to the proposed research. Applying pharmacogenomic data to surgical outcomes, using machine learning and artificial intelligence to add phenotypic data to the VASQIP program with the goal of rapidly implementing the results into patient care to optimize patient centered decision making and outcomes are all innovative. Specific Aims: 1) Identify pharmacogenomic risk associations with outcomes among individuals receiving vascular surgery and cardiac surgery the past 10 years for established (tier 1 and 2) drug/ gene sets. 2) Identify pharmacogenomic risk associations with outcomes among individuals receiving vascular surgery and cardiac surgery the past 10 years for non-established (tier 3) drug/ gene sets. 3) Assess frequency of study drug usage and presence of pharmacogenomic genes for power modeling future studies. 4) Identify high-risk subgroups that may benefit from pharmacogenomic testing. Methodology: This is a retrospective cohort study that will use the standard VASQIP variables and outcomes. Baseline analysis will use linear regression or Cox’s proportional hazards model and will control for patient baseline characteristics and surgical factors using propensity scores with matching or inverse weighting. Machine learning methods such as artificial neural networks, classification and regression trees, or ensemble learning will be used to improve predictions and account for nonlinear relationships and interactions among the potentially large set of pharmacogenomic features. Next Steps/Implementation: The results of the proposed research will be used to update all VASQIP surgeries, then field implementation can occur for real time clinical decision support. High risk patient subgroups will be identified that would benefit from preoperative pharmacogenomic testing and further intervention studies.