The following is taken from an evaluation of the data from the report dated 20 November 2015, data cut-off 5 December 2014, which included sensitivity analyses including BICR of approximately 40% of the data.
The preplanned sensitivity analyses 1-7 support the findings from the investigator-assessed analysis for the ITT population as at the data cut-off 5 December 2014; none are presented in support of the analyses of data from subsequent cut-off dates.
Sensitivity analysis 8, designed to test investigator bias (an issue in the earlier Phase I/II Study 1003 due to the open label nature and multiple data driven amendments) used data for the ITT population comprised of BICR-derived data if either the BICR declared an earlier PFS or if there was concordance with the investigator, and investigator data for all other patients (59.5%) in the study. There was an imbalance in the distribution of BICR reviews as this was selected randomly for 40.5% of the population (42.4% treatment arm and 36.8% control arm). This audit included 70 of the total 195 declarations of PD by the investigators: 42 in the palbociclib and fulvestrant arm and 28 in the control arm that is, the random sampling of 40.5% of the study population incorporated assessment of 35.9% of events assessed as PD by the investigators.
The discordance of investigator and BICR assessments were presented for the 40.5% of ITT randomly sampled. The following were calculated by the Clinical Evaluator:
Total agreement on outcomes of timing of PD or no PD occurred in 170/211 (80.6%) results examined from the ITT population (123/147 or 83.7% for the palbociclib and fulvestrant arm compared with 47/64 (73.4%) for the placebo and fulvestrant arm;
The percentage where the investigator recorded an ‘early’ PD, and the BICR did later or not at all was higher in the palbociclib and fulvestrant arm (18/42; 42.9%) than the placebo and fulvestrant arm (6/28; 25%);
The percentage where the BICR recorded an ‘early’ PD, and the investigator did later or not at all was lower in the palbociclib and fulvestrant arm (6/24; 25%) than the placebo and fulvestrant arm (11/17 cases; 64.71%);
A weakness of the audit was that the BICR was not conducted across the entire ITT population. Generating an estimated HR to support the sensitivity analysis from amalgamating predominantly (59.5%) investigator assessed data with the data amended following BICR assessments generates a population where the validity of the outcomes for more than half has not been confirmed; the results and value of such an analysis are limited (Sensitivity analysis 8). As an assessment of the value of the audit, of greater relevance is the lower HR observed when the analysis is confined to the 40.5% whose data were sampled: this was presented as 0.268 (95% CI: 0.158, 0.455; stratified 1-sided p-value <0.000001) in favour of palbociclib plus fulvestrant. However, this figure does not belong in the PI and should be removed – see Comment 7 below.
The total concordance rates are high based on the data.
Where discordance did occur, the BICR was more likely to record an earlier PD for the placebo and fulvestrant arm, and the investigator was more likely to identify an earlier PD for the palbociclib and fulvestrant arm; of note, the HR for the combined BICR/investigator PFS analysis was lower than for the investigator analysis indicating this had the greater effect on the results.
No information is provided on the censoring rates for the two differing groups of assessors. Therefore, the sponsor is requested to present for the 40.5% whose data was assessed by both groups:
Two tables with the same information as in Table 126.96.36.199.1 [not in this document]: one for the BICR and then one for the investigator assessment data;
under the same headings as in Table 188.8.131.52 [not in this document] the results restricted to the 40.5% whose data was assessed by both groups to allow an assessment of the impact of the BICR on that subgroup.
However, as discordance has been established as being relatively low, the subsequent BICR secondary analyses are not as critical.
The PI statement in the Clinical Trials section needs to use the sponsor’ description of the BICR and state that the currently presented PFS ‘was supported by a random sample blinded independent committee review audit analysis conducted on 40.5%...’ .
Given the clinical evaluator’s conclusions, which are in agreement with the sponsor’s own statement in the CSR: ‘The objective of the random sample BICR audit approach was to corroborate the analysis results of the primary endpoint (that is, investigator assessed PFS) and to assist in the evaluation of potential bias. The BICR audit approach was not intended to provide an alternative means of definitive analysis’ the HR for the BICR PFS analysis should be removed from the PI as it has no clinical relevance for prescribers.
Overall, within the limitations of the methodology, the data presented provided some reassurance that investigator bias is unlikely with respect to reporting of the PFS, which is important given the side effect profile of palbociclib (particularly neutropenia) would likely lead to an awareness by the investigator of the treatment allocation.
The sample size for the audit limits the ability of the BICR to provide support for analyses of endpoints that occurred less commonly, given there will be fewer outcomes available for direct comparisons of the declarations by the investigators. There is a lack of power to detect a difference between the arms or between the analyses for such events.
It is noted that the BICR for Study 1008 incorporated all ITT patients thereby avoiding these limitations.
The sponsor included in addition to the CSR (cut-off date 5 December 214), two updated PFS analyses (16 March 2015, 23 October 2015), which also included updated OR, DoR and clinical benefit rate. Neither of these updated analyses was accompanied by BICR-derived analyses.
Comment: The PI does not include any of the updated data from the cut-off date of 23 October 2015, which means all of the results presented in the text, tables and figures in the Clinical Trials section have been superseded. There is a mixture of data from the original CSR (cut-off date 5 December 2014 and 16 March 2015. It is also noted that all the safety data and text in the PI from the cut-off date 5 December 2014 and not from the latest safety update (31 July 2015). The sponsor is requested to provide an updated PI for evaluation as part of the s31 response.
The clinical evaluator will evaluate the data but confine comments on the PI to the parts pertaining to this study that will not be changed by updated data.
The forest plot indicates that for most of the subgroups within the study, defined by stratification factors and baseline characteristics, there appeared to be a consistent treatment effect over time. Those where it was not statistically significant (that is, 95% confidence intervals cross 1) included those with:
Those of ‘Black and other’ or ‘Asian’ race
Those from the Asia/Pacific region;
a shorter disease-free interval;
more >3 lines of prior therapy;
most recently received: ‘anti-estrogen’ or ‘other’
At the 3rd data cut-off (23 October 2015) compared with earlier subgroup analyses, the HR for the PFS analysis for those whose most recent therapy was defined as ‘other’ (defined as anything other than an aromatase inhibitor, tamoxifen or toremifene) now crosses 1 indicating uncertainty about a benefit in this population.
Each of the first 3 of these groups are associated with a poorer prognosis, and the lack of a statistically significant treatment effect (hazard ratios all cross 1) is not unexpected. The non-significant subgroup analysis for those with >3 prior treatments may reflect the uncertain benefit of fulvestrant in more heavily pre-treated this population. However, the number of patients is small especially in the control arm so this subgroup analysis should be interpreted with caution.
The category of ‘other’ includes those no longer receiving an endocrine therapy as their most recent treatment and indicates from that treatment choice, it was likely that their disease was considered at that time to no longer be responsive to aromatase inhibitors or SERMs. Breast cancer is a heterogeneous disease and that there was some degree of response may indicate (albeit not statistically significant) following the last non-endocrine treatment, a degree of progression of the endocrine-sensitive residual disease as well as the known effect of fulvestrant in disease resistant to AIs and SERMs.
It is important that this latest updated forest plot is included in the PI, without truncation, to ensure prescribers are aware of the outcomes. Similarly, the accompanying PI text requires modification as outlined in the PI changes.
Figure 11: Study A5481023 Forest plot of subgroup analyses as of 23 October 2015
Comments: The PI currently shows a truncated forest plot which introduces another race not identified in the study labeled in the PI as ‘unspecified’ and does not included 4 of those subgroups mentioned above. Specifically, the ‘more >3 lines of prior therapy’ has been left off the PI while the other prior treatment line breakdowns are included. The sponsor should include the comprehensive forest plot from the latest update (see PI changes) to ensure provision of accurate and balanced information about the subgroups.
Similarly, under Table 3 in the PI, the statement that prolongation of PFS ‘was demonstrated in patient individual subgroups’ should be changed to ‘most individual patient subgroups’ as the existing statement implies all, and together with just a truncated forest plot, does not demonstrate those subgroups where the HR crossed 1, which introduces uncertainty about the treatment effect.
The PI includes a paragraph about the treatment effect in pre/perimenopausal women which brackets together the PFS for each group receiving palbociclib and fulvestrant and then the placebo and fulvestrant arms. It is difficult to follow and should just present the median PFS and HR for each of these groups
In the update report using an earlier cut-off date of 16 March 2015, discontinuations due to withdrawal of consent occurred in 1.2 % (4 patients) in the palbociclib and fulvestrant arm but now based on the 23 October 2015 cut-off are reported as 0.9% (3 patients) with a later cut-off date - the sponsor is requested to explain why there are now fewer presented (Clinical Questions).
Multivariate analysis of PFS treatment effect indicated that those with ECOG 1 or 2 performance status has a statistically significantly lower chance of a reduced PFS or death (HR 1.402; 95% CI: 1.053, 1.866, p=0.02) compared with those with ECOG 0. Similarly those without visceral metastases had a better outcome than those with visceral metastases.
Comment: These results confirm the already understood poor prognosis associated with these factors.