Abstract:
|
When health-related quality of life (QOL) is measured in clinical studies where patients experience morbidity or mortality, it is advisable to consider the possibility that missing data are non-ignorable. In a randomized trial of patients with advanced non-small-cell lung cancer (NSCLC), a sensitivity analysis was performed using three models that assumed different missing data mechanisms: a conditional linear model (Wu and Bailey, Biometrics 45:939-955, 1989), a joint mixed-effects and time-to-dropout model (Schluchter, Stat Med. 11: 1861-1870, 1992) and an outcome-dependent selection model (Diggle and Kenward, Appl Stat 43:49-93, 1994). In the first two of the three models there was evidence of non-ignorable missing data, but not in the third model. This study provides an example of a case where if the test of non-ignorable process in not rejected, one can only conclude that the missingness is ignorable only if the particular model is correct. In this example, it appears that the missing data mechanism depends on the random-effects rather than the outcome.
|