Matter 1 - Indirect data comparison and subgroup analysis

Background

As no head-to-head clinical trials have been performed to evaluate the comparative efficacy of teriparatide and aledronate for the requested PBS listing (i.e. SQ3 patient subgroup who have sustained a painful fracture while being treated with an antiresorptive agent), a common comparator (i.e. placebo) and modelled evaluation were supplied by the sponsor to predict the relative outcomes and cost-effectiveness of the two therapies in this high risk patient group.

The data supporting teriparatide is derived from the GHAC trial which was a randomised, placebo-controlled trial of two doses of teriparatide (20ug/day and 40ug/day) using morphometric vertebral fracture as the primary efficacy outcome. Only data from patients on the approved dose of teriparatide (i.e. 20ug/day) and placebo was evaluated in the submissions. Subgroup analyses based on risk factors for future fracture were included as apriori analyses. The sponsor identified patients with SQ3 grade fractures using a combination of baseline risk factors, and then performed treatment by subgroup interaction analyses (post-hoc). The ITT results of the complete trial population have been published in a peer-reviewed medical journal (New England Journal of Medicine). In addition, the results from the post-hoc subgroup analysis of patients with prevalent vertebral fractures from the GHAC study have been published by Gallagher et al in The Journal of Clinical Endocrinology and Metabolism.

The data supporting aledronate in patients with established osteoporosis was derived from the FIT-VFA, an arm of the Fracture Intervention Trial which included a subgroup of patients with postmenopausal osteoporosis and a prior vertebral fracture. The trial has also been published in a peer-reviewed journal (The Lancet). Pre-planned subgroup analyses of the FIT-VFA suggest that the effect of aledronate is not attenuated by baseline risk factors. Consequently, the sponsor decided to use the ITT analysis in the common comparator evaluation. However, the submissions to PBAC by Eli Lilly used the full trial data for aledronate on the basis of a published report by Ensrud et al (1997) which was published in the Archives of Internal Medicine journal. The study concluded that aledronate compared with placebo resulted in an overall 47% reduction in the risk of new radiographic vertebral fractures. Treatment effect was uniform across a variety of predefined subgroups with all p-values >0.3 for the interaction between treatment (i.e. aledronate) and subgroup. The subgroups were defined at baseline by femoral neck bone mineral density, number of existing vertebral fractures, and history of postmenopausal fracture.

The choice of comparator had been an issue of conjecture in earlier submissions to the PBAC but appears to have been agreed upon in the final (fourth) submission. It is beyond the scope of this review to directly comment on the choice of an appropriate comparator. However, a brief relevant background to this issue is provided because the validity of the indirect data comparison is dependent on the choice of the comparator and how this choice was derived. Guidelines for submission to the PBAC of new drug listings indicate that if the proposed drug is in a new therapeutic class for which there are other drugs widely used to treat that indication, then the main comparator will usually be the drug most widely prescribed on the PBS to treat that indication. As such, it was recommended by the PBAC and agreed to by the sponsor to have aledronate as the comparator in the fourth submission. However, in terms of available scientific data, the only agent for osteoporosis that has published evidence of efficacy in patients with severe (SQ3) vertebral fractures is raloxifene. This data is derived from a post-hoc analysis of the Multiples Outcomes of Raloxifene Evaluation (MORE) trial. In summary, both methods with respect to the appropriate choice of a comparator drug have merit and may be a potential subject of difference in opinion.
Top of page
A summary of the key efficacy data derived from the GHAC study and the FIT-VFA trial which was presented in the fourth submission is shown in Table 1 (page 15). This data was also presented in the economic modelling evaluation.

The indirect data comparison using a common comparator is presented in Table 2 (page 16). As stated previously, there have been no clinical trials of an antiresorptive agent specifically designed and conducted in a SQ3 subgroup of patients. However, Delmas et al have conducted a post-hoc sub-group analysis of the MORE trial by semiquantitive grades. For the purpose of showing a complete comparative dataset, the result is compared with the GHAC study SQ3 subgroup and is also shown in Table 2. This information was used in a sensitivity analysis for the modelled economic evaluation.

Both tables were presented in Section 2.9 and Section 3 of the November 2006 sponsor submission and have been inserted into this review unchanged from that submission.

Table 1 Incidence of new Vertebral Fractures in Postmenopausal Women with osteoporotic fractures. Treatment with Teriparatide for 19 Months (ITT and SQ3 subgroup) and treatment with alendronate for 2 years 9 months.

	Placebo	Active	RR 95% CI	NNT
	n (%)	n (%)
Teriparatide	N=448	N=444
ITT analysis Number of patients with 1 new VFx (%) Subgroup analyses BMD T-score -3.0 BMD T-score -2.5 BMD T-score -2.0 Baseline VFx = 0 Baseline VFx = 1 Baseline VFx 1 Baseline VFx 2 VFx 2 and T -3.0 VFx 2 and T -2.5 VFx 2 and T -2.0 SQ3 VFx	64 (14.3%) 41(21.24%) 50 (18.59%) 56 (16.14%) 2 (4.0%) 9 (6.8%) 59 (15.01%) 34 (21.38% 22 (27.16%) 29 (27.62%) 33 (24.81%) 27/95 (28.42%)	22 (5.0%) 12 (5.85%) 14 (4.98%) 18 (5.13%) 0 (0.0%) 5 (3.4%) 22 (5.51%) 11 (7.38%) 6 (7.89%) 7 (6.60%) 10 (7.75%) 5/86 (5.81%)	0.347 (0.22, 0.55) p<0.001 0.276 (0.15, 0.51) 0.268 (0.15, 0.47) 0.318 (0.19, 0.53) 0.502 (0.17, 1.46) 0.367 (0.23, 0.59) 0.345 (0.18, 0.66) 0.291 (0.12, 0.68) 0.239 (0.11, 0.52) 0.312 (0.16, 0.61) 0.205 (0.08, 0.51)	11 6.50 7.35 9.08 25.0 29.5 10.5 7.14 5.19 4.76 5.86 4.42
Alendronate	N=1005	N=1022
ITT analysis Number of patients with 1 new VFx (%) Subgroup analyses 2 VFx baseline BMD <0.59 g/cm2	145 (15%) 28.3% (86)	78 (8%) 14.6% (42)	0.53 (0.41 – 0.68) 0.52 (0.37 – 0.72) 0.54 (0.40-0.72)	14.29 7.3

Abbreviations and Notes: VFx= vertebral fracture; CI = confidence interval; SQ = semi-quantitative grading: SQ0 = Normal <20% reduction in vertebral height, SQ1= mild 20 – 25% reduction in vertebral height, SQ2 =moderate 25 - 40% reduction in vertebral height, SQ3 = severe 40% reduction in vertebral height

Top of page
In response to the data displayed in Table 1, the sponsor claims that for the patient population proposed in the final PBS listing restriction, it is warranted to claim that teriparatide treatment results in a treatment effect of 0.20 (95% CI: 0.08, 0.51; p<0.001), or an 80% reduction in vertebral fractures in a severe (SQ3) subgroup of patients.

Additionally, the sponsor states “the improvement in RR is consistent with biological mechanisms of action associated with teriparatide and the pathophysiology of severe vertebral osteoporosis. However, the treatment by subgroup effects for alendronate remains consistent across patient subgroups, therefore, in the absence of data to suggest otherwise, we suggest that the reduction in risk of new vertebral fracture in SQ3 patients treated with alendronate will be 0.53. A common comparator analysis demonstrates that the difference in RR associated with teriparatide in the SQ3 group and alendronate in an osteoporotic population of postmenopausal women is not zero (p=0.0487).”

Table 2: Indirect comparison of teriparatide and alendronate on incident vertebral fractures using a common comparator

Comparison	Teriparatide	Antiresorptive	Difference
Vertebral Fx	RR	RR	Difference	RRR	P
Paired radiographs	0.347 (0.22, 0.55)	0.529 (0.14, 0.69)	-0.422 (-0.96, 0.11)	0.66 (0.38, 1.12)	0.1211
All randomised	0.346 (0.22, 0.55)	0.529 (0.41, 0.69)	-0.426 (-0.96, 0.11)	0.65 (0.38, 1.12)	0.1206
BMD T-score < -2.5 SD	0.268 (0.15, 0.47)	0.529 (0.41, 0.69)	-0.680 (-1.31, 0.06)	0.53 (0.28, 1.01)	0.0330
>2 VFx at baseline	0.345 (0.18, 0.66)	0.522 (0.37, 0.73)	-0.414 (-1.14, 0.31)	0.62 (0.30, 1.26)	0.2624
T-score and > 2 VfX	0.239 (0.11, 0.52)	0.529 (0.41, 0.69)	-0.794 (-1.62, 0.03)	0.46 (0.20, 1.08)	0.0583
SQ3 (paired)	0.187 (0.07, 0.48)	0.529 (0.41, 0.69)	-1.041 (-2.02, 0.06)	0.35 (0.13, 0.94)	0.0374
SQ3 (GHAC subgroup) vs ITT	0.205 (0.08, 0.51)	0.529 (0.41, 0.69)	-0.950 (-1.90, 0.01)	0.39 (0.15, 0.99)	0.0487
SQ3 subgroup (MORE)	0.205 (0.08, 0.51)	0.332 (0.26, 0.42)	0.730 (0.548, 0.971)	0.28 (0.11, 0.73)	0.009

The results of the indirect comparison (Table 2) do not indicate a treatment difference by statistical significance between teriparatide and aledronate for the overall treated population. However, when analysed by severity of disease there is evidence to possibly suggest that teriparatide is superior to aledronate in a subgroup of patients with more severe forms of vertebral osteoporosis at baseline. The result reaches statistical significance in the above presented analysis, though, the finding needs to be assessed for validity before acceptance as being clinically meaningful. In general, the statistical analyses are complex and there are several features of the data and analysis that are open to varying interpretation. This will be discussed in further detail later in the review.

As recorded in the minutes of the March 2006 PBAC meeting, the teriparatide results for the paired radiograph population and the SQ3 subgroup are summarized in Table 3. In addition, the respective heterogeneity tests from that document are presented in Table 4 (shown here unchanged).
Top of page

Table 3: Results for vertebral fractures in paired radiograph population and SQ3 and sub-group of trial GHAC

Patients with a new vertebral fracture	Teriparatide	Placebo	RD (95% CI)	RR (95% CI)
Paired radiograph population (n=892)	22/444 (5.0%)	64/448 (14.3%)	-9.3% (-13.3,– 5.6)	0.35 (0.22, 0.55)
SQ3 sub-group population (n=181)	5/86 (5.8%)	27/95 (28.4%)	-22.6% (-34.5, -10.7)	0.20 (0.08, 0.51)

Table 4: Test for heterogeneity for effect sizes of sub-groups

Fracture sub-groups at baseline	Risk difference (95% CI)	Relative risk (95% CI)
SQ3 sub-group	-0.226 (-0.332, -0.122)	0.205 (0.084, 0.484)
Non-SQ3 sub-group (nil + mild + moderate)	-0.057 (-0.098, -0.019)	0.453 (0.261, 0.782)
Test for heterogeneity*	p = 0.0027	p = 0.1433
Entire trial results (for reference)	-0.093 (-0.133, -0.056)	0.35 (0.22, 0.55)

*Cochran Q test as calculated during the evaluation for the July 2005 PBAC meeting

Top of page

Background to the matter arising

The sponsor described the matter as “PBAC concern that the indirect comparison and subgroup analysis does not support the claim of superiority – paragraph 4 of PBAC minutes.” The relevant excerpt from the minutes read “However, some issues remained outstanding, including (a) the reliance on an indirect comparison across placebo-controlled trials to infer the superiority claim for teriparatide over alendronate rather than a head-to-head randomized trial; (b) use of the results of the post hoc sub-group analysis in place of the overall ITT results for teriparatide in the clinical conclusions and the economic model…”

The discussion in the March 2006 PBAC short minutes is as follows:
“Item 7.8.31 The PBAC noted the submission had appropriately chosen alendronate as the comparator. The Committee further noted that the submission had compared the results of the high risk group for teriparatide with the overall trial population for alendronate, in the absence of the available data for the high risk group for alendronate.

Item 7.8.32 The PBAC also noted that although the tests for interaction between baseline fracture severity and treatment effect indicate that there is a statistically significant difference in absolute risk between the high risk sub-group and the remainder of the trial period, as previously, there was no statistically significant difference for the relative risk (which is used in the modelled economic evaluation) across the two sub-groups. This statistical conclusion is unchanged from that reached previously, namely that the relative risk results of the overall trial population applied to the different baseline risk of the high-risk sub-group should form the basis of any clinical or economic evaluation. The re-submission again relies on a test for trend between the teriparatide and placebo populations in the key trial to support its claim for treatment effect modification. However, this test does not assess treatment effect modification on the relative treatment effect (or multiplicative scale) – which is relied upon in the model, or the absolute treatment effect (or additive scale) – which as concluded above would be expected to vary given that the baseline risk varies across the populations examined in the sub-groups.”

Reviewer’s understanding of the matter

The key issues involved in this matter appear to be two-fold. Firstly, the dependence upon an indirect comparison across placebo-controlled trials (rather than head-to-head studies) to infer the superiority of teriparatide over an appropriate comparator treatment. Secondly, the scientific validity of using the results of a post hoc subgroup analysis in place of the overall ITT results for teriparatide in the clinical conclusion. The two issues are fundamental to the robustness of the sponsor’s scientific claim. Although the two issues are inter-related, at this stage of the review they will be discussed separately to optimize clarity of understanding.

Materials considered

All materials considered by the PBAC relating to the indirect data comparison and subgroup analysis were reviewed. In addition, key reference studies listed on pages 48-49 were used in forming an opinion.