Abstract:
|
When measuring quality-of-life of cancer patients in clinical trials, an increasing amount of information from patients may become unavailable as the study continues. Usually the reason for the missing observations is related to the patient's health state and, therefore, the data cannot be said to be missing at random. When this occurs, most standard methods of estimating missing data or analyzing data, including missing observations, become problematic and may no longer apply. Including only patients with complete data is also problematic. Recently, the biostatistical literature has included a number of statistical methods for handling the bias caused by informative censoring or informative missingness. This study evaluates some of these methods as applied to data collected in palliative care research, where observations could be recorded near the end of a patient's life. Information is available about pain, fatigue, and other symptoms over time, with a high "drop-out" rate. The adequacy of these methods for this type of data is discussed, including the potential for improved modeling of the data.
|