Methods for observational risk-benefit studies of medical devices: an analysis of big data and simulation studies

医疗器械用于外科手术的许多领域。最近公布的欧盟法规可能意味着需要进行大量的上市后器械监督研究。然而，在比较不同医疗器械的风险和益处的观察性研究中，缺乏关于现有统计模型性能的方法学文献，以尽量减少混淆。在观察性药物安全性和比较有效性研究中，存在多种方法，并得到广泛应用。然而，医疗器械流行病学特有的挑战（例如，外科医生的特点、学习曲线、修订、设备修改）。我们的目标是评估不同统计方法在医疗器械的风险/s和效益的观察性研究中的表现，如在实际操作条件下和潜在的所有患者中使用。我们将使用常规收集的大健康数据，以及模拟数据集。这些将使用不同的方法进行分析，如倾向得分分析、知识产权加权、边际结构建模、中断时间序列、竞争风险和方法学研究产生的新风险。我们将通过临床用例研究解决设备流行病学中的实际问题，并通过模拟研究评估方法在这些用例引起的挑战中的性能。通过这项研究，我们将创建最佳方法指南，以应对观察性上市后器械监测研究中的不同挑战，并给出实际临床问题的答案。

Medical devices are used in many areas of surgery. Recently published EU regulations will likely imply the need for numerous post-marketing device surveillance studies. There is however a scarcity of methodological literature on the performance of existing statistical models to minimise confounding in observational studies comparing the risk and benefit of different medical devices. Many methods exist and are widely applied in observational drug safety and comparative effectiveness research. However, there are challenges (e.g., surgeon characteristics, learning curves, revisions, device modifications) specific to medical device epidemiology. We aim to assess the performance of different statistical methods for the observational study of the risk/s and benefit/s of medical devices, as used in actual practice conditions and in potentially all patients. We will use routinely collected big health data, as well as simulated datasets. These will be analysed using different approaches, such as propensity score analyses, IP weighting, marginal structural modelling, interrupted time series, competing risks and novel ones resulting from methodological research. We will address practical questions in device epidemiology with clinical use case studies and assess methods performance in challenges arising from these use cases with simulation studies. With this research we will create guidance on best methods to answer different challenges in observational post-marketing device surveillance research and give answer to actual clinical questions.