Tafamidis delays disease progression in patients with early stage transthyretin familial amyloid polyneuropathy: additional supportive analyses from the pivotal trial

Abstract

Background: Tafamidis, a non-NSAID highly specific transthyretin stabilizer, delayed neurologic disease progression as measured by Neuropathy Impairment Score–Lower Limbs (NIS-LL) in an 18-month, double-blind, placebo-controlled randomized trial in 128 patients with early-stage transthyretin V30M familial amyloid polyneuropathy (ATTRV30M-FAP). The current post hoc analyses aimed to further evaluate the effects of tafamidis in delaying ATTRV30M-FAP progression in this trial.

Methods: Pre-specified, repeated-measures analysis of change from baseline in NIS-LL in this trial (ClinicalTrials.gov NCT00409175) was repeated with addition of baseline as covariate and multiple imputation analysis for missing data by treatment group. Change in NIS-LL plus three small-fiber nerve tests (NIS-LL + Σ3) and NIS-LL plus seven nerve tests (NIS-LL + Σ7) were assessed without baseline as covariate. Treatment outcomes over the NIS-LL, Σ3, Σ7, modified body mass index and Norfolk Quality of Life–Diabetic Neuropathy Total Quality of Life Score were also examined using multivariate analysis techniques.

Results: Neuropathy progression based on NIS-LL change from baseline to Month 18 remained significantly reduced for tafamidis versus placebo in the baseline-adjusted and multiple imputation analyses. NIS-LL + Σ3 and NIS-LL + Σ7 captured significant treatment group differences. Multivariate analyses provided strong statistical evidence for a superior tafamidis treatment effect.

Conclusions: These supportive analyses confirm that tafamidis delays neurologic progression in early-stage ATTRV30M-FAP.

Trial registration number: NCT00409175.

Keywords:

• Amyloidosis
• disease-modifying drug
• familial amyloid neuropathy
• polyneuropathy
• tafamidis
• transthyretin

Introduction

Transthyretin familial amyloid polyneuropathy (ATTR-FAP) is a rare, relentlessly progressive, life-threatening condition caused by mutations in the TTR gene that destabilize the TTR protein, facilitating its misfolding and aggregation into insoluble amyloid fibrils [1–4]. Over 100 different pathogenic TTR mutations have been identified to date, the most common of which is the Val30Met (V30M) mutation that is endemic in regions of Portugal, Sweden, Japan and Brazil [3,5].

In ATTR-FAP, amyloid deposition occurs primarily in peripheral and autonomic nerve tissues, causing a progressively debilitating distal-to-proximal sensorimotor neuropathy with autonomic symptoms. Neuropathy may be accompanied by various combinations of cardiac, gastrointestinal, renal or ocular symptoms [3,4,6]. The age at onset of ATTR-FAP varies between the second and ninth decades of life, and without treatment, death occurs on average 10 years after symptom onset [1,3].

Treatment options for ATTR-FAP are evolving. Originally, liver transplantation, which removes the main production site of variant TTR, was the only treatment option for ATTR-FAP. Liver transplantation can halt neuropathy progression in well-selected patients with mild disease and is associated with a 20-year survival rate over 50% in patients with hereditary TTR amyloidosis [7]. However, liver transplantation is an invasive procedure requiring lifelong immunosuppressive therapy and survival rates are generally poorer for late-onset ATTRV30M-FAP and non-V30M TTR mutations than for early-onset ATTRV30M-FAP [7,8]. In recent years, significant progress has been made in non-invasive treatment strategies [4,9,10]. Tafamidis, an orally administered, non-NSAID, highly specific TTR stabilizer, is the first approved disease-modifying medicine for treating ATTR-FAP. Currently indicated for use to delay neurological impairment in adult patients with early-stage ATTR-FAP in Europe, Latin America and Asia, tafamidis is emerging as a first-line treatment option in advance or in lieu of liver transplantation [9–12].

The efficacy of tafamidis in slowing neuropathy progression was evaluated in an 18-month, double-blind, placebo-controlled trial in 128 patients with early-stage ATTRV30M-FAP from Europe (82%) and Latin America (18%), using treatment response in Neuropathy Impairment Score–Lower Limbs (NIS-LL) and change from baseline in Norfolk Quality of Life Questionnaire–Diabetic Neuropathy Total Quality of Life Score (TQOL) as co-primary endpoints (ClinicalTrials.gov NCT00409175) [13]. The co-primary analysis for NIS-LL and TQOL favoured tafamidis over placebo (45.3 versus 29.5% responders, p = .068 and change from baseline of 2.4 versus 6.9 points, p = .116, respectively), and a key secondary analysis of change from baseline in NIS-LL continuous score showed a significant reduction in clinical deterioration with tafamidis relative to placebo (p = .027) [13]. Reduced rates of neuropathic progression were sustained for a total of 30 months in a 12-month open-label extension of this trial [14].

Here, we describe additional post hoc analyses of the data from the pivotal trial that employed alternative methods of analysing change from baseline in NIS-LL by itself or in combination with other pre-specified efficacy measures of nerve function, quality of life and nutritional status to further examine the beneficial effects of tafamidis in delaying neurologic impairment in ATTRV30M-FAP.

Methods

Study design

Study design, eligibility criteria and participant flow have been reported previously [13]. Patients (18–75 years with a documented TTR V30M mutation, symptomatic biopsy-confirmed ATTR-FAP and a Karnofsky performance status ≥50, who provided informed consent) were randomized 1:1 to 20 mg tafamidis meglumine (oral) once daily or placebo for 18 months [13].

Outcome measures

The five pre-specified efficacy measures included in the current analyses have been described in detail previously [13]. The NIS-LL score ranges from 0 (normal) to 88 (total impairment) based on clinical impairment in muscle strength, sensation and reflexes. The summated three nerve tests small-fiber normal deviate score (Σ3 NTSF nds, Σ3), a composite of cooling detection threshold, heat/pain detection threshold and heart rate response to deep breathing, ranges from −11.2 (normal) to 11.2 (total impairment). The summated seven nerve tests normal deviate score (Σ7 NTs nds, Σ7), a composite of five nerve conduction studies, vibration detection threshold at the hallux and heart rate response to deep breathing, ranges from −26 (normal) to 26 (total impairment). For Σ3 and Σ7, individual test data were expressed as normal deviates based on healthy subject cohort data from the Mayo Clinic, Rochester, MN. The modified body mass index [mBMI, BMI (kg/m²)×serum albumin concentration (g/L)] compensates for potential oedema caused by malnutrition due to gastrointestinal dysfunction. TQOL ranges from −2 (best possible TQOL) to 138 (worst possible TQOL).

Two further outcome measures, the neurophysiological composite endpoints NIS-LL + Σ3 and NIS-LL + Σ7, have emerged as endpoints in subsequent clinical trials of ATTR-FAP and were summated as described by Dyck et al. [15] and added post hoc.

Statistical analyses

Analyses were conducted in the intent-to-treat (ITT) population. If not otherwise indicated, only observed cases were included.

Repeated-measures analyses of variance

In the previous, pre-specified, repeated-measures analysis of variance (ANOVA) [13], change from baseline in NIS-LL, Σ3, Σ7, TQOL and mBMI at Months 6, 12 and 18 was analysed using a repeated-measures mixed-effect model with an unstructured covariance matrix, fixed effects for treatment, month, treatment-by-month interaction and patient as a random effect (pre-specified model). For TQOL, missing values at month 18 were imputed using a last observation carried forward method and the analysis model included baseline TQOL as a covariate [13].

To account for a numeric difference in NIS-LL baseline values between treatment groups, a post hoc analysis repeated the pre-specified repeated-measures ANOVA of the LS mean change from baseline in NIS-LL [13] with the addition of the baseline NIS-LL value as covariate. In a second post hoc analysis, multiple imputation sensitivity analysis [16] was performed to impute missing NIS-LL data, primarily for patients who discontinued prematurely. Two approaches were used. The primary approach imputed missing data for patients based on non-missing data within their specific treatment group (i.e. placebo or tafamidis). A second, more conservative approach imputed missing data for all patients based on non-missing data in the placebo group regardless of their assigned treatment. For each approach, the pre-specified repeated-measures ANOVA of change from baseline in NIS-LL (without baseline as covariate) was repeated on three batches of 1000 multiply imputed datasets to add a higher level of reproducibility. Additionally, LS mean changes from baseline in NIS-LL + Σ7 and NIS-LL + Σ3 were analysed post hoc using the repeated-measures ANOVA model without baseline as covariate.

Multivariate analyses

Two multivariate analyses were performed post hoc to quantify the overall evidence that tafamidis improved outcomes compared with placebo based on five key efficacy outcomes: NIS-LL, Σ3, Σ7, mBMI and TQOL. Both analyses included all patients in the ITT population with available data for all five outcomes (tafamidis, n = 48; placebo, n = 44).

The first multivariate analysis used a permutation approach to determine the probability of obtaining the observed outcomes due to chance alone. Change from baseline to Month 18 was derived for each patient for each of the five outcomes. Data for each outcome were ordinally ranked (with Rank 1 assigned to the best clinical response and worsening response given higher numeric ranks). Each patient’s five rank scores were summed up into a total rank score for that patient. The mean total rank score was calculated for each treatment group. The difference in mean total rank scores was calculated across the groups (i.e. tafamidis minus placebo).

If there were no treatment effect, i.e. the null hypothesis was true, treatment labels could be randomly reassigned without effect on outcome (i.e. treatment group difference in mean total rank score). In order to test this hypothesis, the distribution of all possible outcomes under the null hypothesis of no treatment effect was approximated by permuting the treatment labels 200,000 times. In each permutation, 48 of the observed total rank scores were randomly assigned to tafamidis and 44 to placebo, and the treatment group difference in mean total rank score was recalculated. The actual observed treatment effect from the clinical study was then compared with the permutation-based outcomes (null distribution) to estimate the likelihood of the observed treatment effect if there were no tafamidis effect.

The second multivariate analysis used methods by O’Brien [17] and Pocock [18]. The univariate t-statistics for each of the five efficacy endpoints were obtained from the treatment group differences in mean changes from baseline to Month 18. A global test statistic was generated based on a weighted linear combination of univariate standardized normal deviates from the five correlated key outcomes. Weights for each univariate statistic were proportional to correlations between endpoints.

Results

Patient flow through the tafamidis registration trial and baseline characteristics of the ITT population, which comprised 125 patients randomized to receive tafamidis (n = 64) or placebo (n = 61), have been reported previously [13]. Coelho et al. [13] reported that tafamidis treatment was associated with less deterioration in five efficacy outcome measures compared with placebo as assessed by respective, pre-specified, ITT, repeated-measures ANOVA analyses of change from baseline to Month 18. Statistically significant differences between tafamidis and placebo were observed in NIS-LL, Σ3 and mBMI, and numerical differences favouring tafamidis relative to placebo were observed for Σ7 and TQOL (Figure 1) [13].

Figure 1. Placebo-corrected point estimates of the tafamidis treatment effect on NIS-LL, Σ3, Σ7, mBMI and TQOL at Month 18. ^aLeast-square means treatment group difference in least-square means change from baseline to Month 18 between tafamidis and placebo as estimated by the pre-specified, ITT, observed-case (NIS-LL, Σ3, Σ7 and mBMI) or last observation carried forward (TQOL), repeated-measures analysis [13]. ^bmBMI values were divided by 10 to fit all data onto the same scale. ^cThe analysis model included baseline TQOL as a covariate. For patients with post-baseline assessments, missing values at Month 18 were imputed using a last observation carried forward method. For patients without a post-baseline assessment, the mean change from baseline at 18 months for patients with post-baseline assessments was used to impute the change from baseline within each treatment group.

The current baseline-adjusted analysis of change from baseline in NIS-LL—which was performed to account for the numeric difference in NIS-LL baseline values between treatment groups [mean (SD): tafamidis, 8.4 (11.4); placebo, 11.4 (13.5); p = .089)—showed a significant positive treatment effect of tafamidis relative to placebo at Months 12 and 18 (Figure 2(a)). The baseline-adjusted LS mean increase in NIS-LL from baseline to Month 12 was 1.5 points [95% confidence interval (CI) − 0.01–3.0] in the tafamidis group versus 4.5 points (95% CI 3.0–6.1) in the placebo group, a group difference of 3.0 points (95% CI 0.9–5.2; p = .007). By Month 18, the LS mean change from baseline was 2.9 points (95% CI 1.1–4.8) in the tafamidis group versus 5.6 points (95% CI 3.8–7.4) in the placebo group, a significant treatment group difference of 2.7 points (95% CI 0.1–5.2; p = .043).

Figure 2. Observed-case, repeated-measures analyses of change from baseline in (a) NIS-LL (ANCOVA with baseline as covariate); (b) NIS-LL + Σ3 (pre-specified ANOVA model without baseline as covariate); (c) NIS-LL + Σ7 (pre-specified ANOVA model without baseline as covariate). p values relate to the difference between treatment groups at each time point. ANCOVA, analysis of covariance.

When applying sensitivity analysis based on multiple imputation for missing data by treatment group, LS mean estimates of the increase in NIS-LL total score from baseline to Month 18 were significantly lower for tafamidis compared with placebo (Table 1). The more conservative multiple imputation model, wherein missing tafamidis data were imputed as if patients had been on placebo, showed benefit favouring tafamidis but did not reach statistical significance (Table 2).

Table 1. Sensitivity multiple imputation, repeated-measures ANOVA of change from baseline to Month 18 in NIS-LL total score.

Batch run^a	Mean (SE) treatment group difference (placebo − tafamidis)	95% CI	p value
Primary approach incorporating treatment group
1	2.787 (1.345)	0.151–5.423	.041
2	2.815 (1.351)	0.166–5.464	.040
3	2.798 (1.347)	0.157–5.438	.040
Secondary approach imputing all missing data based on placebo model
1	2.310 (1.429)	0.491–5.112	.109
2	2.333 (1.427)	0.464–5.130	.105
3	2.319 (1.429)	0.481–5.119	.108

^a Each batch run represents the combined results from 1000 multiply imputed data sets.

SE: standard error.

Table 2. Multivariate analysis of the overall clinical efficacy of tafamidis compared to placebo using methods by O’Brien [17] and Pocock [18]: univariate t-tests for each of the five efficacy outcome measures.

	Mean (SD) change from baseline to Month 18		Treatment group difference (placebo − tafamidis)
	Tafamidis (n = 48)	Placebo (n = 44)	Mean	t-statistic	p value
NIS-LL	2.2 (4.4)	5.2 (8.7)	3.0	2.12	.037
Σ3	0.3 (2.1)	1.6 (2.5)	1.3	2.67	.009
Σ7	1.2 (3.9)	3.6 (4.7)	2.4	2.69	.009
mBMI^a	39.3 (75.1)	−27.9 (89.0)	67.3	3.93	.0002
TQOL	2.0 (14.8)	7.1 (25.7)	5.1	1.18	.240

^a The placebo minus tafamidis difference for mBMI was converted to a positive number in order to allow for common interpretation of direction with the other four efficacy variables.

SD: standard deviation.

Statistically significant treatment group differences in the LS mean change from baseline in NIS-LL + Σ3 were observed from Month 12 (Figure 2(b)). The mean (SD) baseline NIS-LL + Σ3 score was 13.9 (14.2) points in the tafamidis group and 17.1 (15.7) points in the placebo group. The LS mean increase in NIS-LL + Σ3 from baseline to Month 12 was 1.6 points (95% CI −0.1–3.2) in the tafamidis group versus 6.1 points (95% CI 4.4–7.8) in the placebo group, a treatment group difference of 4.5 points (95% CI 2.2–6.8; p = .0001). The LS mean increase in NIS-LL + Σ3 from baseline to Month 18 was 3.0 points (95% CI 1.1–4.9) in the tafamidis group versus 7.5 points (95% CI 5.6–9.4) in the placebo group, a treatment group difference of 4.5 points (95% CI 1.8–7.2, p = .001).

Similarly, significant treatment group differences in the LS mean change from baseline in NIS-LL + Σ7 were also observed at Months 12 and 18 (Figure 2(c)). The mean (SD) baseline NIS-LL + Σ7 score was 16.1 (18.3) points in the tafamidis group and 20.2 (19.8) points in the placebo group. At Month 12, the LS mean change from baseline in NIS-LL + Σ7 was 2.4 points (95% CI 0.4–4.3) in the tafamidis group versus 7.8 points (95% CI 5.9–9.8) in the placebo group, a treatment group difference of 5.5 points (95% CI 2.8–8.2; p = .0001). At Month 18, the LS mean change from baseline in NIS-LL + Σ7 was 4.3 points (95% CI 2.0–6.6 points) in the tafamidis group versus 9.1 points (95% CI 6.7–11.4) in the placebo group, a treatment group difference of 4.8 points (95% CI 1.5–8.0 points; p = .004).

Multivariate analysis of five key efficacy outcomes (NIS-LL, Σ3, Σ7, mBMI and TQOL) using a permutation approach found that the actual (observed) array of outcomes would have occurred by chance in approximately 1 in 50,000 permutations if there were no beneficial treatment effect (Figure 3).

Figure 3. Multivariate permutation analysis comparing the observed treatment group difference (tafamidis − placebo) in mean total rank score over NIS-LL, Σ3, Σ7, mBMI and TQOL at Month 18 with the null distribution of no difference between tafamidis and placebo. Each bar represents a summed rank score range of 5 (e.g. 0 reflects permutations falling between −2.5 and +2.5).

O’Brien [17] and Pocock [18] global statistical test procedures showed a significant overall improvement across the five key efficacy endpoints in the tafamidis treatment group relative to placebo (p < .0001, Table 2).

Discussion

These multifaceted analyses of data from the pivotal trial of tafamidis in adult patients with early-stage ATTRV30M-FAP provide further support for the benefits of tafamidis in delaying disease progression in ATTR-FAP. There were four main findings: (i) statistically significant treatment group differences in mean change from baseline in NIS-LL over 18 months between tafamidis and placebo were retained when adjusting for baseline NIS-LL and when applying multiple imputation for missing data. Notably, there was a minimal mean increase in NIS-LL from baseline to Month 18 in the tafamidis group, which was reduced by nearly half compared with the placebo group; (ii) two composite measures combining NIS-LL with electrophysiological measures of small-fiber (NIS-LL + Σ3) or predominantly large-fiber (NIS-LL + Σ7) nerve function showed significantly less deterioration in neurologic function in the tafamidis group than in the placebo group; (iii) a permutation analysis of the array of outcomes observed during the 18-month treatment period showed that the chance of obtaining the observed results on the five outcomes studied would be 1 in 50,000 if there was no treatment effect; (iv) use of a global test statistic to assess the efficacy of tafamidis across the same set of five endpoints provided additional strong evidence of a superior overall treatment effect for tafamidis over placebo.

These additional analyses, though post hoc, further confirm the favourable effects of tafamidis treatment and are consistent with those reported previously for the pre-specified primary and key secondary efficacy analyses [13]. All of these analyses found slowing of neurologic disease progression and better preservation of nutritional status and quality of life with tafamidis relative to placebo. The demonstrated beneficial effect of tafamidis on slowing deterioration in neurological function was also maintained in a 12-month open-label extension study [14] and for up to 6 years in a second long-term extension study in this ATTRV30M-FAP study population (ClinicalTrials.gov NCT00925002, Barroso et al., submitted). Further, in an analysis in patients from the tafamidis pivotal trial [13] with mild neurological impairment at baseline, tafamidis treatment was associated with minimal neurologic disease progression throughout 5.5 years of treatment [19].

Being a rare disease with substantial heterogeneity, the optimization of sensitive and comprehensive outcome measures to monitor treatment effects in ATTR-FAP is critical and ongoing. The robust results for the clinical/neurophysiologic composite endpoints (NIS-LL + Σ3 and NIS-LL + Σ7) and the two multivariate analyses support a multifaceted approach based on multiple measurements when assessing treatment effects in patients with ATTR-FAP. Individual outcome measures may not capture all aspects of the multidimensional and highly heterogeneous neurologic deficits caused by ATTR-FAP such as sensation loss, decrease of muscle stretch reflex, muscle weakness, nerve conduction abnormalities and autonomic dysfunction [20]. Multivariate analytic approaches may improve sensitivity and allow a more comprehensive multi-dimensional assessment. It is anticipated that better characterization of the clinical spectrum and natural progression of ATTR-FAP may help to further develop standardized, reliable and responsive, multifaceted assessment tools for measuring overall treatment effects in future ATTR-FAP clinical studies [21,22].

Conducting clinical research in rare disease is challenging. Knowledge regarding appropriate clinical trial design, endpoints, target population, analysis approaches and even the natural history of the disease is often lacking. The pivotal study of tafamidis was the first clinical trial designed to look at the efficacy of a disease-modifying therapy in ATTR-FAP and the fundamental knowledge of this complex disease has continued to evolve since it was conducted. As a result, it is important to look at the pre-specified key secondary endpoints as well as post hoc analyses to further characterize the full treatment effect of tafamidis. Indeed, several of the analyses included in this article were in part conducted at the request of regulatory authorities to further characterize the full treatment effect of tafamidis.

These analyses address a potential baseline imbalance, provide a sensitivity assessment of the possible impact of missing data and missing data assumptions on the outcome, and offer a more global assessment of treatment effect of tafamidis on the multifaceted neurologic deficits associated with ATTR-FAP (i.e. an analysis across outcomes). As with any post hoc analysis, there is potential for bias and interpretation of post hoc results should be made with this in mind. Nevertheless, these findings confirm the reliability and robustness of the original results.

In conclusion, the present analyses, which evaluated the impact of tafamidis on neuropathy progression by further examining NIS-LL continuous scores in conjunction with other endpoints, clearly demonstrate that tafamidis halted or slowed disease progression for up to 18 months in patients with early-stage ATTRV30M-FAP. The global array of outcomes, coupled with the robust NIS-LL results under different methods of analysis, all point to the overall benefit of tafamidis treatment. The totality of evidence supports tafamidis as a safe and effective disease-modifying treatment for ATTR-FAP.

Acknowledgements

The authors thank the study investigators (Supplementary Appendix 1), personnel and patients for their important contributions. Medical writing support was provided by Sharmila Blows, PhD, and Susanne Vidot, PhD, of Engage Scientific Solutions and was funded by Pfizer.

Disclosure statement

D.K., J.S., M.S. and L.A. are employees of Pfizer. B.G. is an employee of inVentiv Health (Burlington, MA, USA), who provided statistical support, which was funded by Pfizer.

Additional information

Funding

This study was sponsored by Pfizer, New York, NY, USA. Medical writing support and statistical support was funded by Pfizer.