|
Abstract:
|
A common problem in medical experiments occurs when an outcome, such as a patient's quality of life (QOL) at a fixed time point, cannot be observed because the patient has died. Such an outcome is said to be "censored by death" or "truncated by death." This differs from conventional instances of missing data, as the data here are nonexistent, not merely unobserved. It has been suggested that the effect on QOL be measured by comparing the QOL of treated and control patients, but only considering those patients who would live under both treatment and control. We consider how the treatment effect on this group can be estimated, using certain normality assumptions and an EM algorithm that considers both the treatment arm and the control arm of the experiment together. We test the performance of this algorithm on simulated data and examine its application to employment data.
|
