van den Bor Rutger M, Vaessen Petrus W J, Oosterman Bas J, Zuithoff Nicolaas P A, Grobbee Diederick E, Roes Kit C B
Julius Clinical Ltd., Broederplein 41-43, 3703 CD Zeist, The Netherlands; Julius Center for Health Sciences and Primary Care, UMC Utrecht, P.O. Box 85500, 3508 GA Utrecht, The Netherlands.
Julius Clinical Ltd., Broederplein 41-43, 3703 CD Zeist, The Netherlands.
J Clin Epidemiol. 2017 Jul;87:59-69. doi: 10.1016/j.jclinepi.2017.03.018. Epub 2017 Apr 12.
Central monitoring of multicenter clinical trials becomes an ever more feasible quality assurance tool, in particular for the detection of data fabrication. More widespread application, across both industry sponsored as well as academic clinical trials, requires central monitoring methodologies that are both effective and relatively simple in implementation.
We describe a computationally simple fraud detection procedure intended to be applied repeatedly and (semi-)automatically to accumulating baseline data and to detect data fabrication in multicenter trials as early as possible. The procedure is based on anticipated characteristics of fabricated data. It consists of seven analyses, each of which flags approximately 10% of the centers. Centers that are flagged three or more times are considered "potentially fraudulent" and require additional investigation. The procedure is illustrated using empirical trial data with known fraud.
In the illustration data, the fraudulent center is detected in most repeated applications to the accumulating trial data, while keeping the proportion of false-positive results at sufficiently low levels.
The proposed procedure is computationally simple and appears to be effective in detecting center-level data fabrication. However, assessment of the procedure on independent trial data sets with known data fabrication is required.
多中心临床试验的中央监测成为一种越来越可行的质量保证工具，特别是用于检测数据造假。在行业资助的临床试验和学术临床试验中更广泛地应用，需要有效且实施相对简单的中央监测方法。
我们描述了一种计算简单的欺诈检测程序，旨在反复（半）自动应用于累积的基线数据，并尽早检测多中心试验中的数据造假。该程序基于伪造数据的预期特征。它由七项分析组成，每项分析标记约10%的中心。被标记三次或更多次的中心被视为“潜在欺诈”，需要进一步调查。使用已知欺诈的经验性试验数据对该程序进行了说明。
在示例数据中，在对累积试验数据的大多数重复应用中都检测到了欺诈中心，同时将假阳性结果的比例保持在足够低的水平。
所提出的程序计算简单，似乎在检测中心层面的数据造假方面有效。然而，需要在具有已知数据造假的独立试验数据集上对该程序进行评估。
