Understanding ethnic differences in the comparative effectiveness of antidiabetic medications using high-dimensional propensity scores in electronic

在英国，不同种族的2型糖尿病发病率存在着巨大的不平等。南亚和非洲人后裔患糖尿病的年龄较早，发展成中风和心脏病等并发症的速度也更快。糖尿病结果存在种族差异的潜在原因是，治疗指南主要来自欧洲白人人群。由于临床平衡，英国指南目前推荐了一系列二线糖尿病治疗。目前尚不清楚治疗反应是否因种族而异。药物有效性的差异可能源于生物因素和文化对不同药物的可接受性。在这个项目中，我将比较抗糖尿病药物在预防种族间并发症的有效性与同类的抗糖尿病药物。我还计划使用处方数据中的模式来研究不同种族患者对药物的依从性和不依从性的决定因素。然后，我将使用来自英国生物银行的数据来研究不同种族人群对2型糖尿病的遗传易感性和药物反应。使用电子健康记录来研究治疗效果的一个缺点是无法控制哪些患者被分配到哪些治疗组。在临床试验中，患者通常被随机分配到治疗组，以确保被比较的组是相似的。这使得隔离治疗效果变得更加容易。实际上，治疗组之间存在非随机差异，因为患者和医疗保健提供者会选择最适合的治疗方法。我将使用高维倾向评分来控制治疗组的差异。这种方法使用一种算法来识别潜在的混淆变量，然后创建比较组，其中这些变量是平衡的。这使得更容易分离暴露（药物结合种族）对结果（糖尿病血管并发症）的影响。本项目将为研究糖尿病二线治疗疗效的种族差异提供有价值的证据基础。这一发现可能有助于开发有针对性的治疗方案，更有效地预防不同种族的2型糖尿病患者的血管并发症。MRC策略和核心技能本课程符合MRC技能的重点，即提高定量技能。在此范围内，我将获得技能和理解：-评估治疗有效性的不同方法和药物流行病学中常用的统计技术。-不同定量数据源之间的三角测量，包括基因组和健康利用数据。-具备管理和分析大数据的计算能力。-应用复杂的定量技术来控制观测数据集的混淆。关键词：糖尿病，种族，电子病历，药物流行病学，高维倾向评分。

IntroductionThere are enormous inequalities in rates of type 2 diabetes across ethnic groups in the UK. Population of South Asian and African descent develop diabetes earlier in life and progress more quickly to complications, such as stroke and heart disease.A potential cause of ethnic disparities in diabetes outcomes arises because therapeutic guidelines are predominantly derived from white European populations. UK guidelines currently recommend a range of second-line diabetes treatments due to clinical equipoise. It is unknown whether treatment response differs by ethnicity. Differences in the effectiveness of medications may arise from biological factors to cultural acceptability of different medications. Project descriptionIn this project I will compare the effectiveness of antidiabetic medications at preventing complications between ethnic groups with the same class of antidiabetic medication. I also plan to use patterns in prescription data to study adherence to medications and determinants of non-adherence among patients from different ethnic groups. I will then use data from the UK biobank to study genetic susceptibility and drug response to type 2 diabetes across different ethnic groups..One disadvantage of using electronic health records to study treatment affects is the inability to control which patients are assigned to which treatment groups. In clinical trials patients are often randomly assigned to treatment groups to ensure that the groups being compared are as similar. This makes it easier to isolate the effects of the treatment. In reality, there are non-random differences between treatment groups because patients and healthcare providers choose which treatment is most suitable. I will use high-dimensional propensity scoring to control for differences in the treatment groups. This methods uses an algorithm to identify potentially confounding variables and then creates comparison groups among which these variables are balanced. This make it easier to isolate the effects of the exposure (medication combined with ethnicity) on the outcome (vascular complications of diabetes).RelevanceThis project will make a valuable contribution to the evidence base for ethnic differences in the effectiveness of second-line diabetes treatment. The findings may feed into the development targeted treatment regimens that will be more effective at preventing vascular complications for type 2 diabetes patients from different ethnic groups.MRC strategy and core skillsThis studentship meets the MRC skills priority of enhancing quantitative skills. Within this remit, I will gain skills and understanding of:- Different ways of assessing the effectiveness of treatments and commonly used statistical techniques in pharmacoepidemiology.- Triangulation between different quantitative data sources, including genomic and health utilization data.- Computational skills required to managing and analyse big data.- The application of complex quantitative techniques to control for confounding in observational datasets.Keywords: diabetes, ethnicity, electronic health records, pharmacoepidemiology, high dimensional propensity scoring.