The Two-Sample Problem with Induced Dependent Censorship
Article first published online: 25 MAY 2004
Volume 55, Issue 4, pages 1108–1113, December 1999
How to Cite
Huang, Y. (1999), The Two-Sample Problem with Induced Dependent Censorship. Biometrics, 55: 1108–1113. doi: 10.1111/j.0006-341X.1999.01108.x
- Issue published online: 25 MAY 2004
- Article first published online: 25 MAY 2004
- Received April 1998. Revised March 1999. Accepted March 1999.
- Medical cost;
- Multivariate failure time;
- Quality-adjusted survival time;
- Survival analysis
Summary. Induced dependent censorship is a general phenomenon in health service evaluation studies in which a measure such as quality-adjusted survival time or lifetime medical cost is of interest. We investigate the two-sample problem and propose two classes of nonparametric tests. Based on consistent estimation of the survival function for each sample, the two classes of test statistics examine the cumulative weighted difference in hazard functions and in survival functions. We derive a unified asymptotic null distribution theory and inference procedure. The tests are applied to trial V of the International Breast Cancer Study Group and show that long duration chemotherapy significantly improves time without symptoms of disease and toxicity of treatment as compared with the short duration treatment. Simulation studies demonstrate that the proposed tests, with a wide range of weight choices, perform well under moderate sample sizes.