Using auxiliary time-dependent covariates to recover information in nonparametric testing with censored data.

摘要: Murray and Tsiatis (1996) described a weighted survival estimate that incorporates prognostic time- dependent covariate information to increase the efficiency of estimation. We propose test statistic based on Pepe Fleming (1989, 1991) these estimates. As in Fleming, is an integrated difference two estimated curves. This has been shown be effective at detecting differences crossing hazards settings where logrank performs poorly. method uses stratified longitudinal get more precise estimates underlying curves when there censored this leads powerful tests. Another important feature it remains valid informative censoring captured by incorporated covariate. In case, Pepe-Fleming known biased should not used. These methods could useful clinical trials with heavy include collection over time covariates, such as laboratory measurements, are subsequent or capture related censoring. Information from auxiliary variables relate endpoint can used aug- ment provided treatment comparisons overall, marginal, survival. Some researchers have developed tests model relationships be- tween progression within specific marginal models. Gray (1994) considered three-state which influence sur- vival following via kernel estimators. Finkelstein Schoenfeld similar three state semi-Markov case using parametric models for given progression. Others augmented failure overall fewer as- sumptions relationship between covariates outcome. Kosorok (1993) combined linear rank statistics primary secondary endpoints. Malani (1995) suggested covariate-based modification Efron's redistribution right algo- rithm adjust score Cox (1972) proportional model. Robins Rotnitzky (1992) presented inverse probability weighting strategies estimation testing, weights distribution