Estimating the effect of treatment on quality of life in the presence of missing data due to drop-out and death

Abstract

Quality of life was measured in a study of two multidrug regimens for mycobacterium avium complex MAC bactaeremia using the MOS-HIV questionnaire. The effect of treatment on quality of life was estimated at each follow-up time in three ways: (1) using only the observed data, (2) after assigning the worst possible quality of life scores for individuals who died, and (3) after imputing missing scores for patients who either died or dropped out of the study. The overall quality of life scores were also compared between treatment groups with categorical generalized estimating equation models and three-dimensional graphs. Of the 179 patients included in these analyses, 84 (47%) either died or dropped out during the 16 week study period. When the quality of life scores were compared between the treatment groups with the Wilcoxon rank sum test using only the observed data, there was no significant difference between the groups at 16 weeks of follow-up. When the worst possible quality of life scores were assumed for patients who had died, both the magnitude and the statistical significance of the difference in the quality of life scores between the groups increased. Imputing missing data for patients who either dropped out or died resulted in even larger differences in quality of life between the treatment groups. We conclude that ignoring missing data due to drop-outs and death can result in an underestimation of the treatment effect and overly optimistic statements about the outcome of the participants on both treatment arms due to the selective drop-out of participants with poorer quality of life. To obtain a valid assessment of the effect of treatment on quality of life, the experience of the patients who died or dropped out of the study must be considered.