Taylor J M, Kim D K
Department of Radiation Oncology, Jonsson Comprehensive Cancer Center, School of Public Health, UCLA 90024.
Int J Radiat Biol. 1989 Aug;56(2):161-7. doi: 10.1080/09553008914551311.
The poor statistical properties of the Fe-plot as a method of estimating the alpha/beta ratio in the linear quadratic model for multifraction data are summarized. A Monte Carlo simulation of quantal response data shows that the Fe-plot method of analysis is about half as efficient as logistic regression, indicating that twice the amount of data would be needed to obtain the same accuracy in the estimate of alpha/beta if the Fe-plot was used rather than logistic regression. In addition, the confidence intervals from the Fe-plot are inappropriately too narrow and do not have the correct coverage rate. Model-based statistical methods which directly link the treatment variables, dose and number of fractions to the response are advocated.
总结了Fe-图作为估计多分次数据线性二次模型中α/β比值方法时较差的统计特性。对定量反应数据的蒙特卡罗模拟表明，Fe-图分析方法的效率约为逻辑回归的一半，这表明如果使用Fe-图而非逻辑回归，要在α/β估计中获得相同的准确性，所需的数据量将是原来的两倍。此外，Fe-图得出的置信区间过窄且不合适，覆盖率不正确。提倡使用直接将治疗变量、剂量和分次次数与反应联系起来的基于模型的统计方法。
