Beacon H J, Thompson S G
Medical Statistics Unit, London School of Hygiene and Tropical Medicine, U.K.
Stat Med. 1996 Dec 30;15(24):2717-32. doi: 10.1002/(SICI)1097-0258(19961230)15:24<2717::AID-SIM518>3.0.CO;2-E.
Quality of life data present considerable statistical challenges because of their longitudinal and multidimensional nature, and also because the available data are often very unbalanced through missing values. Here we exemplify the potential of multi-level models, that is, hierarchical random coefficient models, for such data. The discussion is developed in the context of analysing the quality of life data from a trial of palliative treatment in non-small-cell lung cancer. Not only do multi-level models provide a flexible modelling framework for the investigation of the underlying behaviour of response, for example, giving simple estimates of treatment effects, but they also permit a description of the differences between subjects and allow the analysis of multi-dimensional outcomes. The assumptions of Normality, homogeneity, and independence of the within- and between-subject variance components can be investigated and the models can be extended to provide explicit modelling of variance heterogeneity. It is concluded that multi-level models, for which software is now available, provide a natural and powerful approach to the analysis of longitudinal data in general, and multi-dimensional quality of life data in particular.
生活质量数据因其纵向和多维度的性质，以及可用数据常常因缺失值而极不均衡，故而呈现出相当大的统计挑战。在此，我们举例说明多层次模型，即分层随机系数模型，对于此类数据的潜力。讨论是在分析非小细胞肺癌姑息治疗试验中的生活质量数据的背景下展开的。多层次模型不仅为研究反应的潜在行为提供了一个灵活的建模框架，例如给出治疗效果的简单估计，而且还允许描述个体之间的差异，并能分析多维度结果。可以研究个体内和个体间方差成分的正态性、同质性和独立性假设，并且可以扩展模型以对方差异质性进行明确建模。得出的结论是，现在已有软件可用的多层次模型，总体上为纵向数据的分析，尤其是多维度生活质量数据的分析，提供了一种自然且强大的方法。
