Background In a clinical trial, survival may be censored by the end of the study, especially for subjects who enter later in the enrollment period. If there is a trend toward better survival over time then longer survivors experience shorter censoring times (heavier censoring). In such a case, the censoring and survival times are correlated, and thus the censoring is nonignorable in the sense that standard survival models that assume independent censoring could yield incorrect inferences. We will demonstrate a graphical method for analyzing sensitivity of estimates of survival model parameters to small departures from nonignorable censoring.
Methods We assume a parametric model of survival together with a scaled beta model for the censoring process that incorporates the dependence of censoring time on survival time. We assess sensitivity using an index of local sensitivity to nonignorability (Troxel et al. [1]). High sensitivity indicates a large impact of nonignorable censoring on the parameter of interest and a need for additional modeling.
Results A simulation study shows that the approach is valid for practical use. We apply our method to a clinical trial evaluating the survival benefit of a surgically implanted left ventricular assist device in subjects with end-stage heart failure. Sensitivity is somewhat larger in estimates of the mean survival in the device arm, where survival is better and the fraction censored is therefore larger. The degree of nonignorability required to substantially affect estimates is larger than seems plausible, however.
Conclusions Our results illustrate how one can apply sensitivity analysis to evaluate the reliability of survival parameter estimates in a clinical trial.
Get full access to this article
View all access options for this article.
References
1.
1. Troxel AB, Ma G, Heitjan DF. An index of local sensitivity to nonignorability. Statistica Sinica2004; 14: 1221–1237.
2.
2. David HA, Moeschberger ML. The theory of competing risks. London: Charles Griffin, 1978.
3.
3. Peterson AV. Bounds for a joint distribution function with fixed sub-distribution functions: application to competing risks. Proceedings of the National Academy of Science, 1976; 73: 11–13.
4.
4. Williams JS, Lagakos SW. Models for censored survival analysis: constant-sum and variable-sum models. Biometrika1977; 64: 215–224.
5.
5. Lagakos SW. General right-censoring and its impact on the analysis of survival data. Biometrics1979; 35: 139–156.
6.
6. Link WA. A model for informative censoring. Journal of the American Statistical Association1989; 84: 749–752.
7.
7. Heitjan DF. Ignorability and coarse data: some biomedical examples. Biometrics1993; 49: 1099–1109.
8.
8. Zheng M, Klein JP. Estimates of marginal survival for dependent competing risks based on an assumed copula. Biometrika1995; 82: 127–138.
9.
9. Scharfstein DO. Estimation of the failure time distribution in the presence of informative censoring. Biometrika2002; 89: 617–634.
10.
10. Heitjan DF, Rubin DB. Ignorability and coarse data. The Annals of Statistics1991; 19: 2244–2253.
11.
11. Rubin DB. Inference and missing data. Biometrika1976; 63: 581–592.
12.
12. Zhang J. Sensitivity analysis of nonignorable coarsening. Doctoral dissertation, Department of Biostatistics & Epidemiology, University of Pennsylvania, 2004.
13.
13. Rose EA, Gelijns AC, Moskowitz AJ et al. Long-term use of a left ventricular assist device for end-stage heart failure. New England Journal of Medicine2001; 345: 1435–1443.
14.
14. Rose EA, Moskowitz AJ, Packer M et al. The REMATCH Trial: rationale, design, and end points. Annals of Thoracic Surgery1999; 67: 723–730.
15.
15. Press W, Flannery B, Teukolsky S, Vetterling W. Numerical recipes: the art of scientific computing. Cambridge: Cambridge University Press, 1986.
16.
16. Xie H. An index of sensitivity to nonignorability: extensions and applications. Doctoral dissertation, Department of Biostatistics, Columbia University, 2003.
