Data synthesis over a network of multiple treatment comparisons for joint longitudinal and event-time outcomes.

当检查关于特定疾病的数据时，检查来自多个数据源的相关数据是有益的，例如多个研究的结果、来自几个不同医院或中心的数据集等。这些不同数据源之间的人群可能不同，从而导致来自不同研究、医院或中心的数据之间的响应的可变性。这一点需要在分析中加以考虑。荟萃分析是一种分析来自多个不同来源的数据的方法，同时考虑数据来源之间的差异。通常从个人收集有关其健康的各种不同信息。这些可能包括随着时间的推移重复测量的项目，例如每周血液检查的结果。它们还可以包括直到感兴趣的特定事件发生的时间，例如直到疾病进展到下一阶段的时间。联合建模作为一种同时对随时间重复测量的结果进行建模的方法越来越受欢迎，沿着时间直到事件结果。最近，已开展工作，将联合建模扩展到多个数据源的情况。对于给定的病症或疾病，通常有两种以上可能的治疗方案。实际上，很可能在不同的数据源中检查不同的可能治疗方案。例如，一项研究可能检查了治疗方案A、B和C，而另一项研究检查了B、C和D，第三项研究仅检查了A和C。目前的多数据源联合建模方法要求在所有数据源中直接比较治疗方案，这意味着有助于分析的数据源必须比较同一组治疗方案。然而，网络荟萃分析方法允许在数据源之间研究的治疗选项集不相同的情况下进行数据分析，因为它允许直接信息（例如A与C相比）和间接信息（A与B相比，B与C相比）来告知A与C的治疗比较。因此，假设可能的治疗方案之间存在路径，网络荟萃分析不需要数据源检查相同的治疗方案集。网络荟萃分析尚未扩展到联合建模方法。该奖学金将开发这种方法，并在免费，易于使用的统计软件中实施。然而，联合建模分析中的一个已知问题是，联合模型可能是时间密集型的，以适应，一个问题，将越来越明显的大型多数据源分析。在此研究期间，计算机科学和机器学习方法，如顺序蒙特卡罗或SMC采样器将被用于尝试显着减少联合模型的模型拟合时间。开发的方法和软件将用于三项分析;在包含复发时间和标志物（如CD 4细胞计数）的多医院数据分析中比较HIV患者的治疗选择，在包含收缩压和死亡时间测量值的多研究数据集中比较高血压患者的治疗选择，并在分析停药时间沿着各种生物标志物时，比较不同重症监护室患者的心血管药物支持治疗方案。

When examining data concerning a particular disease, it is beneficial to examine relevant data from multiple data sources, for example the results of multiple studies, datasets from several different hospitals or centres etc. The populations may differ between these different data sources, resulting in variability in response between data from different studies, hospitals or centres. This needs to be accounted for in the analysis. Meta-analysis is an approach to analyse data from multiple different sources, whilst accounting for differences between the data sources.Commonly a range of different information is collected from individuals concerning their health. These could include items repeatedly measured over time, such as the results from weekly blood tests. They could also include the time until a specific event of interest occurs, such as the time until a disease progresses to the next stage. Joint modelling has been growing in popularity as a method to simultaneously model outcomes repeatedly measured over time, along with time until event outcomes. Recently, work has been conducted to expand joint modelling to the case of multiple data sources. For a given condition or disease, commonly there are more than two possible treatment options. In reality, it is likely that different sets of possible treatment options are examined across the different data sources. For example one study may have examined treatment options A, B and C, whilst another examined B, C, and D, and a third examined only A and C. Current multi-data source joint modelling methods require treatment options to be directly compared in all data sources, meaning that the data sources contributing to the analysis would have to compare the same set of treatment options. However, network meta-analysis methods allow analysis of data where the set of investigated treatment options is not identical between data sources, as it allows both direct information (e.g. A compared to C) and indirect information (A compared to B, B compared to C) to inform about the treatment comparison A versus C. As such, provided paths exist between the possible treatment options, network meta-analyses do not require the data sources to examine the same sets of treatment options.Network meta-analyses have not been extended into joint modelling methodology. This fellowship will develop this methodology, and implement it in free, easy to use statistical software. However, a known problem in joint modelling analyses is that joint models can be time intensive to fit, an issue that would be increasingly noticeable with large multi-data source analyses. During this fellowship, computer science and machine learning methods such as Sequential Monte Carlo or SMC samplers will be employed in an attempt to significantly reduce model fitting times for joint models. The developed methodology and software will be applied to three analyses; to compare treatment options for patients with HIV in an analysis of multi-hospital data containing time until relapse and markers such as CD4 cell count, to compare treatment options for hypertensive patients in a multi-study dataset containing systolic blood pressure and time to death measurements, and to compare treatment options for cardiovascular drug support for patients in different intensive care units in an analysis of time to treatment withdrawal along with various biomarkers.