The analysis of repeated multivariate binary quality of life data: a hierarchical model approach.

Many quality of life measuring instruments consist of a number of questions that are answered on ordinal scales. Often these responses are then totalled to give a summary score for each quality of life domain within the instrument. This, however, may lose valuable information about individual aspects of patient quality of life and also can have little intuitive meaning. Here we present an alternative analysis, in which dichotomized individual items of the questionnaire are analyzed. We first show how a hierarchical logistic regression model for repeated binary data can be extended to the multivariate case. We then use such a model for analyzing the prevalence of six symptoms in a palliative treatment trial in non-small-cell lung cancer. The analysis provides information about the correlations between symptoms, both between and within person. If appropriate, it also permits the estimation of a treatment effect common to all symptoms. Methods for model checking are discussed. We conclude that this methodology can provide a more intuitive and informative analysis of quality of life data than that obtained by considering summary scores.