Predictors of early trial termination using individual-level participant data and aggregate-level data from multiple trials

项目背景最大限度地保留是试验设计和实施的主要焦点，确定提高试验保留的策略已被确定为重要的研究重点。然而，我们不知道有任何研究已经确定了哪些个体参与者的特征可以预测人员流失。这些知识将有助于为试验设计和实施提供信息。在初步工作中，我们使用行业资助的一系列指标条件和药物类别试验的个人水平参与者数据（IPD），发现合并症与磨损有关（见https://bmcmedicine.biomedcentral.com/articles/10.1186/s12916-019-1427-1分析中使用的试验）。每增加一个合并症，减员的优势比增加约1.1倍（优势比1.11,95% CI:1.07至1.14）。这在不同的条件和药物类别中都是高度一致的。我们还发现，在试验中，年龄和性别并不能预测减员。ryan McChrystal现在正在进行一个博士项目——使用IPD进行一系列试验——以进一步探索参与减少的预测因素。例如，治疗方法与指标相似的合并症（如高血压和心绞痛）可能不如治疗方法截然不同的合并症（如前列腺疾病和心力衰竭）重要。此外，其他因素（如多药、疾病严重程度指数、基线生活质量测量）也可能与消耗有关。除IPD外，分析还将包括试验水平的数据。对于某些试验，FDA要求在clinicaltrials.gov上报告试验完成统计数据（以及未完成的原因，如不良事件、死亡、缺乏疗效、失去随访、医生决定、怀孕、违反协议、受试者退出或其他）。这些数据将与IPD一起在多级元回归模型（doi/10.1111/rssa.12579）中进行分析，该模型在R/Stan包multima （https://github.com/dmphillippo/multinma）中实现。在这些模型中包含聚合级数据将增加样本量，提高精度和普遍性，同时避免聚合偏差和非可折叠性偏差。这些模型和实现它们的软件包都是新颖的，因此博士生将接受DP的培训以及NW和SD的额外建议。在监督小组成员KG （https://www.cochranelibrary.com/cdsr/doi/10.1002/14651858.MR000032.pub3/full）进行的Cochrane综述的基础上，学生将把他们的发现与更广泛的关于随机试验中保留率的文献联系起来。这将使研究结果的效用最大化：为试验设计提供信息，通过允许试验人员预测不同资格标准对消耗风险的影响，为试验行为提供信息，通过确定哪些招募的参与者最有可能需要额外的支持来完成试验，为未来的干预研究提供信息，特别是关于“保留差”人群的研究议程

Background to the projectMaximising retainment is a major focus of trial design and conduct and identifying strategies to improve trial retention has been identified as an important research priority. However, we are not aware of any study which has identified which individual participant characteristics predict attrition. Such knowledge would be useful for informing both trial design and conduct.In preliminary work, using individual-level participant data (IPD) across industry-funded trials for a range of index conditions and drug classes we found that comorbidity was associated with attrition (see https://bmcmedicine.biomedcentral.com/articles/10.1186/s12916-019-1427-1 for trials used in the analysis). Per additional comorbid condition, the odds ratio for attrition increased by around 1.1-fold (odds ratio 1.11, 95% CI:1.07 to 1.14). This was highly consistent across conditions and drug classes. We also found that within trials age and sex did not predict attrition.What the studentship will encompassRyan McChrystal is now undertaking a PhD project - using IPD for a range of trials - to further explore predictors of attrition of participation. For example, comorbid conditions which have similar treatments to the index condition (eg hypertension and angina) may be less important than comorbid conditions with very different treatments (eg prostate disease and heart failure). Moreover, other factors (eg polypharmacy, index disease severity, baseline quality of life measures) may also be associated with attrition.In addition to IPD, the analysis will also include trial-level data. For certain trials the FDA mandates reporting of trial completion statistics (alongside the reasons for non-completion such as adverse event, death, lack of efficacy, lost to follow-up, physician decision, pregnancy, protocol violation, withdrawal by subject or other) on clinicaltrials.gov. These data will be analysed alongside the IPD in multi-level meta regression models (doi/10.1111/rssa.12579) as implemented in the R/Stan package multinma (https://github.com/dmphillippo/multinma). The inclusion of aggregate-level data in these models will increase the sample size, improving precision and generalisability whilst avoiding aggregation bias and non-collapsibility bias. The models and the package in which they are implemented are both novel, so the PhD student will receive training in both from DP and additional advice from NW and SD. Building on a Cochrane review undertaken by KG, a member of the supervisory team (https://www.cochranelibrary.com/cdsr/doi/10.1002/14651858.MR000032.pub3/full), the student will relate their findings to the wider literature on retention in randomised trials. This will maximise the utility of the findings for: -informing trial design, by allowing trialists to predict the effect of different eligibility criteria on the risk of attrition informing trial conduct, by identifying which recruited participants are at most likely to require additional support to complete a trial informing the research agenda for future intervention studies, particularly as regards "retention poor" populations