学术文献库

Survival analysis of right censored data arises often in many areas of research including medical research. Effect of covariates (and their interactions) on survival distribution can be studied through existing methods which requires to pre-specify the functional form of the covariates including their interactions. Survival trees offer relatively flexible approach when the form of covariates' effects is unknown. Most of the currently available survival tree construction techniques are not based on a formal test of significance; however, recently proposed ctree algorithm (Hothorn et al., 2006) uses permutation test for splitting decision that may be conservative at times. We consider parameter instability test of statistical significance of heterogeneity to guard against spurious findings of variation in covariates' effect without being overly conservative. We have proposed SurvCART algorithm to construct survival tree under conditional inference framework (Hothorn et al., 2006) that selects splitting variable via parameter instability test and subsequently finds the optimal split based on some maximally chosen statistic. Notably, unlike the existing algorithms which focuses only on heterogeneity in event time distribution, the proposed SurvCART algorithm can take splitting decision based in censoring distribution as well along with heterogeneity in event time distribution. The operating characteristics of parameter instability test and comparative assessment of SurvCART algorithm were carried out via simulation. Finally, SurvCART algorithm was applied to a real data setting. The proposed method is fully implemented in R package LongCART available on CRAN.