Biomedical Engineering Reference
In-Depth Information
10.2
Statistical Methods for PFS Analysis
In this section, we briefly introduce the statistical aspects of methods and
their implementation for PFS analysis.
10.2.1
Conventional Approach
The goal of right-point and mid-point imputations is to transform the interval-
censored data into right-censored data, and use standard right-censored data
methods in PFS analysis. It is noted that when the assessment schedules are
exactly the same between treatment arms and no deviation in assessments oc-
curs, the ranks of event times are the same for both right-point and mid-point
imputations. As a result, statistical analysis based on rank-based statistical
methods (e.g., logrank test and Cox PH model), yields identical results for
right-point and mid-point imputations in ideal situations. In addition, be-
cause interval-censored data from clinical trials arise from regular assessment
schedules, and thus have heavy ties in most cases, it is also important to
consider which method of tie-handling should be used. In this study, we use
Efron's method (Efron (1977)) to handle tied event times, which approximates
the exact method (Kalbfleisch and Prentice (2002)) reasonably well but is less
computationally intensive. The logrank test and Cox PH model are imple-
mented in SAS procedure
PHREG
and
LIFETEST
or R function
coxph
in the
survival
package.
10.2.2
Finkelstein's Method for Interval-Censored Data
Finkelstein's method can be briey viewed as a natural extension of the Cox
PH model for interval-censored data. The main idea can be sketched as fol-
lows: Based on the observed interval-censored event times, we create a series
of nonoverlapping bins such that any observation can be viewed as a sum of
Search WWH ::
Custom Search