Analysis of Quality of Life Data with Death and Drop-out in Advanced Non-Small-Cell Lung Cancer Patients

Abstract

In measuring quality of life (QOL), outcome-dependent missing values are inevitable because of longitudinal nature of the study. In particular, in clinical trials of advanced-stage disease, it is desirable to distinguish differences between reasons for missing, death and drop-out, because QOL scores for death cases are not really missing data, but are nonexistent and are simply undefined. We focus on estimating the local average treatment effect among survivors. Standard randomized treatment comparisons cannot be performed because the QOL scores are only defined in the non-randomly selected subgroup of survivors. We propose a new estimation method of the survivor average causal effect (SACE) in the presence of both death and dropout. The proposed estimator is a weighted average of the standard estimators for survivors where the weight is the probability that the patient would have survived had he/she received the other treatment. For drop-out cases, the multiple imputation method is applied. Two analysis methods (proposed method and analysis based on only observed survivors) were compared by simulation studies. The proposed estimator had smaller biases with smaller MSEs compared with those of the standard estimator. The proposed method was applied to data from a randomized phase III clinical trial for advanced non-small-cell lung cancer patients.