Biomedical Engineering Reference
In-Depth Information
when no data abnormality exists, treating the right-point of the time inter-
val (the first documented progression date) as the true event time, and using
standard right-censored methods for PFS analysis, is conventionally accepted.
Adopting the convention by Sun and Chen (2010), we call this approach
\right-point imputation." Sometimes, a naive interval-censored method, the
\mid-point imputation," which treats the average of the left-point and right-
point of the time interval as the true event time, is also used in practice.
When potential biases may be introduced, simulation-based sensitivity analy-
ses have also been suggested to assess the robustness of analysis results based
on conventional methods (FDA (2007); Bhattacharya et al. (2009)).
Many interval-censored time-to-event data analysis methods have been de-
veloped in the past two decades (Sun (2006); Zhang and Sun (2010)). They
can be grouped into two types of methods: multiple-imputation or simulation-
based methods, and analytical methods based on either parametric or non-
parametric assumptions. In this chapter, we focus on nonparametric analyti-
cal methods for interval-censored data that are directly analogous to those in
right-censored data analysis, such as logrank test and the Cox PH model. In
particular, Finkelstein (1986) proposed a nonparametric maximum likelihood
estimation method that can be roughly viewed as a generalization of the Cox
PH model for interval-censored data. Finkelstein's method provides estimates
for covariate coecients that are compatible with those from the Cox PH
model. Furthermore, while the logrank test can be obtained as a score test
under the Cox PH model in right-censored data, the score test based on Finkel-
stein's method can be similarly used for hypothesis testing purposes. Sun et al.
(2005) further developed a class of k-sample test for interval-censored data,
and the Finkelstein's score test can be viewed as one of its special cases.
Relatively limited research has been conducted to compare the perfor-
mance of interval-censored methods with conventional methods in practi-
cal settings. Lindsey and Ryan (1998) compared conventional methods with
several parametric and nonparametric interval-censored methods, including
 
Search WWH ::




Custom Search