Krummenauer Frank
Institut für Medizinische Biometrie, Epidemiologie und Informatik der Universität Mainz, Germany.
Klin Monbl Augenheilkd. 2002 Sep;219(9):682-5. doi: 10.1055/s-2002-35165.
Sometimes the statistical comparison of diagnostic measurement methods or subsequent measurement series in the same individuals is based on bivariate scattergrams; correlation coefficients and linear regression lines are provided in addition. However, this strategy for data representation must be considered inappropriate, as soon as the measurement series under consideration show the same physical index (e.g. repeated measurement of the intraocular pressure [mm Hg] 2 and 4 weeks after glaucoma surgery). Such data can be handled by computation of the intraindividual difference of an individual's measurement. The resulting series of intraindividual deviations can then be analysed using standard strategies: description can be based on the intraindividual differences' median and quartiles, graphical representation on the differences' distribution's box whisker plot. This procedure is particularly attractive for data arising from ophthalmological studies, when two eyes of one patient are to be compared. The statistical comparison of diagnostic methods can further be based on the Bland/Altman diagram, which not only illustrates the distribution of differences, but also motivates a dependence between the order of deviations and the absolute order of the measurement values. Whereas a box plot of intraindividual differences and a Bland/Altman diagram illustrate the concordance of two measurement series, the correlation scattergram rather visualizes their reproducibility.
有时，对同一受试者的诊断测量方法或后续测量系列进行统计比较是基于双变量散点图；此外还会给出相关系数和线性回归线。然而，一旦所考虑的测量系列显示相同的物理指标（例如青光眼手术后2周和4周重复测量眼内压[毫米汞柱]），这种数据呈现策略就应被视为不合适。此类数据可通过计算个体测量的个体内差异来处理。然后可以使用标准策略对所得的个体内偏差系列进行分析：描述可基于个体内差异的中位数和四分位数，图形表示可基于差异分布的箱线图。当要比较一名患者的两只眼睛时，此程序对于眼科研究产生的数据特别有吸引力。诊断方法的统计比较还可基于布兰德/奥特曼图，该图不仅说明了差异的分布，还揭示了偏差顺序与测量值绝对顺序之间的相关性。个体内差异的箱线图和布兰德/奥特曼图说明了两个测量系列的一致性，而相关散点图则更直观地展示了它们的可重复性。
