Assessment methods for interlaboratory comparisons of the dicentric assay

To test the performance of different algorithms that can be used in interlaboratory comparisons based on dicentric chromosome analysis, and to evaluate the impact of considering <i>a priori</i> values different to calculate individual laboratory performance based on the ionizing radiation dose estimation. Mean and standard deviation estimations in inter-laboratory comparisons are tested on simulated data and data from previously published inter-laboratory comparisons using three robust algorithms, algorithm A, Algorithm B and Q/Hampel, all programmed in R-project language and implemented in a Shiny application. The simulated data were generated assuming three different probabilities to contaminate inter-laboratory comparisons samples with atypical dose values. Comparison between different algorithms was also done using published exercises where blood samples were irradiated at 0 and 0.7 Gy that represent a challenge for the assessment of an inter-laboratory comparison. The best performance was obtained with the Q/Hampel algorithm for the estimation of the dose mean and with the algorithm B for the estimation of the dose standard deviation under the conditions tested in the simulations. The Q/Hampel algorithm showed the best performance when non-irradiated samples were evaluated and there was a high proportion of identical values. The presence identical values causes the Algorithm B to fail. Real examples illustrating the need to consider standard deviation priors, and the need to use algorithms resistant to a high proportion of identical values are presented. Q/Hampel algorithm is a serious candidate to estimate the dose mean in the inter-laboratory comparisons, and to estimate both parameters when the proportion of identical values equals or higher than the half of the results. When the proportion of identical values is less than the half of the results, the Algorithm B should be considered as a candidate to estimate the standard deviation in the inter-laboratory comparisons with small number of laboratories. We remark that special attention is needed to establish prior definitions of standard deviation in the assessment of inter-laboratory dicentric assay comparisons.

本研究旨在测试可用于基于双着丝粒染色体分析（dicentric chromosome analysis）的实验室间比对的不同算法性能，并评估采用不同先验值（a priori）、基于电离辐射剂量估算来计算各实验室个体表现时所产生的影响。本研究针对实验室间比对中的均值与标准差估算任务，基于模拟数据以及已发表的实验室间比对数据集，对三种稳健算法——算法A、算法B与Q/Hampel——进行测试；上述算法均以R语言（R-project）编写，并已在Shiny应用中实现部署。模拟数据的生成设定了三种不同的概率，用于向实验室间比对样本中掺入非典型剂量值。本研究还采用已发表的比对实验数据开展不同算法间的对比：该实验中血液样本分别接受0Gy与0.7Gy的辐照，该设置对实验室间比对的评估构成了挑战。在模拟实验的测试条件下，Q/Hampel算法在剂量均值估算任务中表现最优，而算法B则在剂量标准差估算任务中表现最佳。当对未辐照样本进行评估且样本中存在高比例相同值时，Q/Hampel算法展现出最优性能。相同值的存在会导致算法B失效。本研究提供了实际案例，用以说明在实验室间比对中需考虑标准差先验值，以及需使用对高比例相同值具有鲁棒性的算法。Q/Hampel算法是实验室间比对中剂量均值估算的有力备选方案，且当相同值占比等于或超过总结果数的一半时，该算法也可用于两项参数的估算。当相同值占比低于总结果数的一半时，对于实验室数量较少的实验室间比对，可考虑采用算法B来估算标准差。本研究强调，在双着丝粒染色体分析的实验室间比对评估中，需特别关注标准差先验定义的设定工作。