NIRG: Evaluation of interventions with rare events: methods for parallel cluster randomised trials and stepped-wedge cluster randomised trials

该项目将在试验结果罕见的情况下改进试验的统计方法。它将专注于将人群而不是个人随机分组的试验。试验是评估干预措施是否对健康研究的所有领域和其他领域产生积极影响的金标准。然而，有时很难对个人进行随机化。这可能是因为正在测试的干预措施影响到整个村庄或医院病房，或者试验正在测试大规模推出个别干预措施的影响。在这种情况下，我们可以将整组个体随机分组，例如我们称之为集群的整个村庄或医院病房。这种分组的随机化对我们分析试验数据的方式有影响，因为同一群组中的个人可能比来自不同群组的个人更相似。如果我们在试验中试图影响的结果是罕见的事件，例如从已经很低的起点降低疾病水平，这种分析就会变得更加复杂。例如，在传染病方面，即使疾病病例数量略有增加，也可能对疾病轨迹产生很大影响。当我们对罕见事件感兴趣时，可能会有一些群体(人群)没有人经历感兴趣的事件(例如感染了疾病)。在这个项目中，我将探索我们在分析这些试验时如何处理这些罕见事件，以便即使我们有罕见的结果，我们也可以相信结果。该项目将专注于两种类型的试验。在第一种类型中，随机分组以在试验持续时间内接受新的干预或控制条件。众所周知，目前分析这些试验的方法不准确地估计了干预的效果，或者低估了我们对这一估计的信心。我将开发一种新的分析方法来克服这个问题。将开发新的软件来帮助研究人员实施我们的新分析方法。第二种类型的试验被称为阶梯式楔形群随机试验。在这里，我们随机分组，以便在试验期间的不同时间切换到干预，以便在试验结束时，所有或大多数分组都接受了干预。当事件罕见时，预计标准的分析方法将难以提供结果，或者将对结果的不确定性给出不准确的估计。该项目将决定如何对阶梯式楔形群随机试验进行现有的分析方法；该方法以前从未被用于这些试验。像第一类试验一样，我还将创建软件来帮助研究人员实施我认为有效的分析方法。至关重要的是，我们拥有经过验证和值得信赖的分析方法，以便与这些试验一起使用，从而能够改善人类健康。到本项目结束时，这两种类型的试验都将有改进的、经过验证的分析方法，当试验结果是罕见事件时，可以使用这些方法。

This project will improve the statistical methods of trials when the outcome of the trial is rare. It will focus on trials that randomise groups of people rather than individual people. Trials are the gold standard for assessing whether an intervention has a positive impact across all areas of health research and beyond. However, it is sometimes difficult to randomise individual people. This could be because the intervention being tested affects a whole village or hospital ward, or the trial is testing the impact of large-scale rollout of an individual intervention. When this is the case, we can randomise groups of individuals such as whole villages or hospital wards that we call clusters. This randomisation of groups has implications for how we analyse data from the trial because individuals in the same cluster are likely to be more like one another than individuals from different clusters. This analysis is further complicated if the outcome we are trying to affect in the trial is a rare event, such as reducing levels of a disease from an already low starting point. For example, in infectious diseases, even a small increase in the number of disease cases can have a large impact on the disease trajectory. When we are interested in a rare event, there might be some clusters (the groups of people) where no one experience the event of interest (e.g. contracted the disease). In this project, I will explore how we deal with these rare events in the analysis of these trials, so that even when we have a rare outcome, we can trust the results. The project will focus on two types of trials. In the first type, the clusters are randomised to receive either the new intervention or a control condition for the duration of the trial. Current methods of analysing these trials are known to inaccurately estimate the effect of the intervention or underestimate our confidence in that estimate. I will develop a new method of analysis to over come this issue. New software will be created to help researchers implement our new analysis method.The second type of trial is called a stepped-wedge cluster randomised trial. Here, we randomise clusters to switch to the intervention at different times during the trial so that by the end of the trial all or most clusters have received the intervention. When events are rare, it is expected that standard methods of analysis will struggle to provide results or will give inaccurate estimates of our uncertainty in results. This project will determine how to conduct an existing analysis method for stepped-wedge cluster randomised trials; the method has not been used for these trials before. Like the first type of trial, I will also create software to help researchers implement the analysis methods I find work well.It is vital that we have validated and trustworthy analysis methods to use with these trials to be able to improve human health. By the end of this project, both types of trials will have improved, validated methods of analysis that can be used when the outcome of the trial is a rare event.