Enhancing the Utility of the Trend-in-Trend Research Design

患者、临床医生和政策制定者通常需要关于药物和其他药物的真实效果的有效信息。 在使用中有明显时间趋势的暴露。例如，罗非昔布的使用迅速增长， 在2000-2013年期间，外源性睾酮的增加伴随着对知识的迫切需求。 这些药物对心血管的影响随机试验很少招募足够数量的受试者，或遵循 足够长的时间，以检查与罕见结果的关联。因此，需要进行非随机研究， 尽管它们容易受到未测量的混杂和其他形式的偏差的影响。为了解决这些限制， 流行病学设计，我们最近介绍了趋势中的趋势设计：一种新颖的，混合流行病学-生态学研究 当研究暴露的使用存在显著趋势时适用的设计。大量额外 需要进行开发，以实现趋势中趋势研究设计对公共卫生的全面影响。最需要的地区 开发包括：评价用于实时上市后安全性监测的序贯分析扩展; 评估机器学习以指定趋势中趋势设计的潜在累积暴露概率 模型;剂量-反应关系评价;以及基于基线的治疗效果异质性评价 患者特征。对于每一种方法的需要，我们提出了一种方法，我们将：1）评估其 统计特性，特别是通过渐近性和有限样本统计功效的大样本有效性; 2） 通过全面的模拟研究评估其性能; 3）使用已知的关联对该方法进行验证测试 罗非昔布与急性心肌梗死的关系。最后，我们将把这些发展应用于 尚未解决的和非常重要的问题，心血管，脑血管和血栓栓塞的影响， 男性的外源性睾丸激素 这项研究将通过极大地扩展一种新型药物的效用，产生广泛的科学、临床和公共卫生影响。 流行病学研究设计具有已证明的有效性，适用于广泛的暴露， 使用趋势突出。它还将评估外源性睾酮的安全性，这是一个具有巨大临床意义的问题。 公共卫生的重要性。检查处方药的安全性是这一方法的一个非常重要的应用 有可能使数百万人受益的设计。由于趋势中趋势设计也适用于非药物 具有明显时间趋势的暴露（例如，医疗程序、生活方式因素、环境暴露等）， 它对人口的潜在影响甚至更广泛。

Patients, clinicians, and policymakers often need valid information about the real-world effects of drugs and other exposures for which there have been pronounced time-trends in use. For example, the rapid growth in the use of rofecoxib in the early 2000s and of exogenous testosterone from 2000–2013 was accompanied by an urgent need for knowledge about the cardiovascular effects of those drugs. Randomized trials seldom enroll a sufficient number of subjects, or follow them long enough, to examine associations with rare outcomes. Therefore, non-randomized studies are needed, even though they are susceptible to unmeasured confounding and other forms of bias. To address these limitations of available epidemiologic designs, we recently introduced the trend-in-trend design: a novel, hybrid epidemiologic-ecologic research design that is applicable when there is a prominent trend in the use of the study exposure. Substantial additional development is needed to realize the full public health impact of the trend-in-trend research design. Areas in most need of development include: evaluation of a sequential analysis extension for real-time post-marketing safety surveillance; evaluation of machine learning to specify the trend-in-trend design’s underlying cumulative probability of exposure model; evaluation of dose-response relationships; and evaluation of treatment effect heterogeneity based on baseline patient characteristics. For each of these methodologic needs, we propose an approach for which we will: 1) assess its statistical properties, in particular large-sample validity through asymptotics and finite sample statistical power; 2) evaluate its performance in thorough simulation studies; and 3) proof-test the approach using the known association between rofecoxib-acute myocardial infarction in healthcare data. Finally, we will apply these developments to the unresolved and highly important question of the cardiovascular, cerebrovascular, and thromboembolic effects of exogenous testosterone in men. This study will have broad scientific, clinical, and public health impact by greatly extending the utility of a novel epidemiologic research design with demonstrated validity and is applicable to a broad range of exposures with a prominent trend in use. It will also evaluate the safety of exogenous testosterone, a question of tremendous clinical and public health importance. Examining the safety of prescription drugs is one enormously important application of this design that has the potential to benefit many millions of people. As the trend-in-trend design is also applicable to non-drug exposures that have a pronounced time-trend (e.g., medical procedures, lifestyle factors, environmental exposures, etc.), its potential impact on the population is even broader.