User Manual Part 2

C-16
CLINICAL STUDY - MADIT II
Table C-6. Cumulative mortality and percentage reduction (continued)
Year
Conventional
Arm
ICD Arm Differe nce
Reduction
CI
a
%
p-value
b
2 Years 21.5 15.5 6.0
28%
5, 46 0.02
3 Years 30.4 21.6 8.8
29%
6, 46 0.02
a. Indicates Confidence Interval for the percentage reduction in cumulative mortality. The cumulative mortality (and associated
standard errors) is taken from the Kaplan-Meier analyses; percentage reduction analyses are based on a log transform method.
b. For null hypothesis that the percentage (%) reduction is zero.
The pre-specified primary analysis of the trial was based on computation of a
hazard ratio, based on an assumption th at the two survival curves satisfy a
proportional hazards condition (one is a power —the ‘hazard ratio’— of the
other), and recognizing the sequential stopping rule of the trial. The hazard
ratio is interpreted as the ratio of instantaneous risks of dying, at each point in
time, in the two treatment groups. The hazard ratio for the ICD group relative to
the conventional therapy group was found to be 0.69, indicating a 31percent
reduction in instantan e ous risk (95 percent confidence interval, 0.51 to 0.93;
p=0.016, reduced from p=0.027 when reaching the stopping boundary, by
incorporation of lagged data). The Cox regression analyses used for this
purpose were stratified by enrollment centers, thus allowing for somewhat
different patient pools at differing locations.
The proportional hazards assumption was evaluated by several standard
statistical methods, all providing support. O ne method is derived from
finding p aralle lism in so-called log (-lo g ) plots o f t he cumulative haz a rd s.
Another is from fitting models that allow for differing hazard ratios in differing
intervals of time, and demonstrating that any apparent differences among the
period-specific hazard ratios can be attributed to chance. One such analysis is
summarizedinTableC-7onpageC-16.
Table C-7. Year-specific hazard ratio (HR)
Year Estimate
CI
1 Year 0.87 0.59, 1.29
2 Years 0.56 0.29, 1.07
3+ Years 0.61 0.28, 1.34
Overall
0.69 0.51, 0.93
The p-value = 0.16 for differences among the 3 HRs, and the p-value = 0.016
for the overall HR. The e xponential mortality curves fit the data ve ry well,
with risks of mortality of 0.0100 each month for patients in the conventional
- DRAFT -