Regression analysis based on semicompeting risks data

• Published in: Journal of the Royal Statistical Society. Series B, Statistical methodology Vol. 70; no. 1; pp. 3 - 20
• Main Authors: Hsieh, Jin-Jian, Wang, Weijing, Adam Ding, A
• Format: Journal Article
• Language: English
• Published: Oxford, UK Oxford, UK : Blackwell Publishing Ltd 01.02.2008; Blackwell Publishing Ltd; Blackwell Publishing; Blackwell; Royal Statistical Society; Oxford University Press
• Series: Journal of the Royal Statistical Society Series B
• Subjects: Biology; Bone marrow; bone marrow transplant; Censored data; Censorship; Copula model; data collection; death; Decision theory; Dependent censoring; disease course; Estimation methods; Estimators; Exact sciences and technology; General topics; Linear inference, regression; Mathematics; Medical research; Model selection; Model testing; Modeling; Multiple events data; Myeloid leukemia; Nonparametric inference; Probability and statistics; Proportions; Regression analysis; Risk; Sciences and techniques of general use; Standard deviation; Statistical analysis; Statistical methods; Statistics; Stem cell transplantation; Studies; Transformation model; Transplants & implants; Weighting functions; Rank statistic; Transformation model; Estimator robustness; Censored sample; Covariate; Parametric method; Multiple events data; Model matching; Misspecification model; Copula model; Marginal distribution; Bayes decision; Censored data; Copulas; Bad specification; Statistical analysis; Model selection; Finite sample; Regression analysis; Statistical decision; Risk analysis; Dependent censoring; Statistical method; Statistical regression; Selection problem; Simulation; Regression model; Decision theory
• ISSN: 1369-7412; 1467-9868
• DOI: 10.1111/j.1467-9868.2007.00621.x

Semicompeting risks data are commonly seen in biomedical applications in which a terminal event censors a non-terminal event. Possible dependent censoring complicates statistical analysis. We consider regression analysis based on a non-terminal event, say disease progression, which is subject to censoring by death. The methodology proposed is developed for discrete covariates under two types of assumption. First, separate copula models are assumed for each covariate group and then a flexible regression model is imposed on the progression time which is of major interest. Model checking procedures are also proposed to help to choose a best-fitted model. Under a two-sample setting, Lin and co-workers proposed a competing method which requires an additional marginal assumption on the terminal event and implicitly assumes that the dependence structures in the two groups are the same. Using simulations, we compare the two approaches on the basis of their finite sample performances and robustness properties under model misspecification. The method proposed is applied to a bone marrow transplant data set.