Australian public assessment for Ibuprofen - Therapeutic Goods ...

AusPAR Caldolor Ibuprofen Phebra Pty Ltd PM-2010-02393-3-1
Final 2 January 2013
Therapeutic Goods Administration
Table 14. CPI-CL-008A: Primary efficacy outcome. Total morphine usage in the 24 h after the
first dose of study medication (ITT population).
Morphine usage (mg)
q6h
Placebo
Caldolor
400 mg q6h
Caldolor
N=134 N=134 N=138
800 mg q6h
mean ± SD 48.9 ± 27.7 46.3 ± 29.4 43.8 ± 33.7
median 45.3 44.0 35.5
range 0.0 - 144.0 3.0 - 198.25 0.0 - 221.3
Transformed morphine usage (mg) *
LS mean (SE) 223.0 (13.8) 208.5 (13.6) 190.6 (13.1)
LS mean difference versus placebo
(95% CI)
- -14.4 (-44.4, 15.5) -32.4 (-62.1, -
2.6)
P-value versus placebo † - 0.458 (NS) 0.030
* Data were transformed using the rank transformation (a non-parametric test), because the raw, log-transformed and Box-
Cox-transformed data were found to be non-normally distributed. † Linear 4-way ANOVA with fixed effects for age group,
weight group, randomisation centre and treatment group.
The statistical analysis was adjusted for multiplicity (testing two separate Caldolor dose
levels versus placebo) using Dunnett’s method. Dunnett’s method reports adjusted pvalues
that should be compared with the usual critical p-value of 0.05. The results
therefore demonstrate a significant effect in the Caldolor 800 mg group and no significant
effect in the Caldolor 400 mg group, in the rank-transformed data.
However, the FDA statistical reviewers did not accept this analysis and the FDA-approved
PI states that statistically significant efficacy was not demonstrated for either dose in this
study. A central point of the FDA reviewers’ argument is that the statistical analysis plan
provided for the analysis of raw data in the first instance, followed by log-transformed
data and then Box-Cox transformed data if the data at each preceding step were not
normally distributed. The non-parametric analysis based on rank-transformed data was
not prespecified (even though one can argue that its use is logical given the non-normality
of the raw, log-transformed and Box-Cox transformed data). Furthermore, the preplanned
analyses failed to show a significant effect on morphine usage for either Caldolor dose
level, and the practice of resorting to an unplanned analysis to provide a different (rather
than a confirmatory) result is not acceptable. Finally, the FDA statistical reviewers noted
that the violation of normality that was used as the excuse for discarding the planned
analyses should not be important when the sample size is sufficiently large (as it was in
this study). Thus, there is no strong reason for accepting the non-parametric analysis in
preference to the planned analyses and the study should therefore be regarded as
inconclusive.
The FDA reviewer’s arguments are cogent and have not been refuted by the sponsor.
While the study results are sufficient to justify exploration of the 800 mg dose in
subsequent studies, the proposed PI statement that the 800 mg dose significantly reduced
morphine consumption in this study should not be allowed.
Study CPI-CL-008B therefore examined only Caldolor 800 mg q6h and found that regimen
to be significantly superior to placebo (Table 15). Caldolor 800 mg reduced mean
morphine usage by about 10% in CPI-CL-008A and 15% in CPI-CL-008B.
Page 50 of 118