Fernández-Fontelo Amanda, Puig Pedro, Ainsbury Elizabeth A, Higueras Manuel
Departament de Matemàtiques, Universitat Autònoma de Barcelona, 08193 Bellaterra (Barcelona), Spain.
Public Health England, Centre for Radiation, Chemical and Environmental Hazards, Chilton, Didcot, Oxon OX11 0RQ, UK.
Radiat Prot Dosimetry. 2018 Jun 1;179(4):317-326. doi: 10.1093/rpd/ncx285.
The goal in biological dosimetry is to estimate the dose of radiation that a suspected irradiated individual has received. For that, the analysis of aberrations (most commonly dicentric chromosome aberrations) in scored cells is performed and dose response calibration curves are built. In whole body irradiation (WBI) with X- and gamma-rays, the number of aberrations in samples is properly described by the Poisson distribution, although in partial body irradiation (PBI) the excess of zeros provided by the non-irradiated cells leads, for instance, to the Zero-Inflated Poisson distribution. Different methods are used to analyse the dosimetry data taking into account the distribution of the sample. In order to test the Poisson distribution against the Zero-Inflated Poisson distribution, several asymptotic and exact methods have been proposed which are focused on the dispersion of the data. In this work, we suggest an exact test for the Poisson distribution focused on the zero-inflation of the data developed by Rao and Chakravarti (Some small sample tests of significance for a Poisson distribution. Biometrics 1956; 12 : 264-82.), derived from the problems of occupancy. An approximation based on the standard Normal distribution is proposed in those cases where the computation of the exact test can be tedious. A Monte Carlo Simulation study was performed in order to estimate empirical confidence levels and powers of the exact test and other tests proposed in the literature. Different examples of applications based on in vitro data and also data recorded in several radiation accidents are presented and discussed. A Shiny application which computes the exact test and other interesting goodness-of-fit tests for the Poisson distribution is presented in order to provide them to all interested researchers.
生物剂量测定的目标是估计疑似受辐照个体所接受的辐射剂量。为此，要对已计数细胞中的畸变（最常见的是双着丝粒染色体畸变）进行分析，并建立剂量响应校准曲线。在用X射线和γ射线进行全身照射（WBI）时，样本中的畸变数量可以用泊松分布恰当地描述，不过在局部身体照射（PBI）中，未受辐照细胞提供的过多零值，例如会导致零膨胀泊松分布。考虑到样本的分布，会使用不同的方法来分析剂量测定数据。为了检验泊松分布与零膨胀泊松分布，已经提出了几种渐近和精确方法，这些方法关注数据的离散度。在这项工作中，我们提出一种针对泊松分布的精确检验，该检验关注由Rao和Chakravarti（《泊松分布的一些小样本显著性检验。生物统计学》1956年；12：264 - 82）提出的、源于占有问题的数据零膨胀。在精确检验的计算可能很繁琐的情况下，提出了基于标准正态分布的近似方法。进行了蒙特卡罗模拟研究，以估计精确检验及文献中提出的其他检验的经验置信水平和功效。给出并讨论了基于体外数据以及几次辐射事故中记录的数据的不同应用示例。展示了一个用于计算泊松分布的精确检验及其他有趣的拟合优度检验的Shiny应用程序，以便将其提供给所有感兴趣的研究人员。
