Growth curve model analysis for quality of life data.

There is increasing interest in measuring health related quality of life in cancer clinical trials. Most quality of life data are measured repeatedly over a fixed time schedule to capture changes and to reflect relative advantages of study treatments. A multivariate repeated measures model is usually used to analyse this type of data. However, one of the difficulties of this analysis is that quality of life may be affected by the occurrence of some critical events experienced by patients. We may separate a patient's lifetime during study into different 'health states'. The duration of these health states may vary among patients, and may relate to the efficacy of the study treatment. In some cases quality of life data may be missing due to one of the many different types of missing data mechanisms specific for a health state. It is reasonable to assume that the missing data mechanism for a treatment arm is homogeneous within a defined health state, and to control for the potential confounding effect to appropriately assess the impact of treatment on the quality of life. In this paper, we propose a growth curve model conditional on a time-dependent variable of defined health states in order to assess the overall treatment effect while taking into account occurrences of missing data and measurements from irregular visits. A specific contrast can be drawn within the overall model for testing a specific hypothesis without relying on the analysis of subgroups of patients based on a smaller number of repeated measurements. Quality of life data from a recently completed small-cell lung cancer randomized trial are used to illustrate this method.