Understanding the Disease Course and Therapeutic Benefit of Tafamidis Across Real-World Studies of Hereditary Transthyretin Amyloidosis with Polyneuropathy: A Proof of Concept for Integrative Data Analytic Approaches

Understanding the Disease Course and Therapeutic Benefit of Tafamidis Across Real-World Studies... Neurol Ther (2018) 7:141–154 https://doi.org/10.1007/s40120-018-0096-x BRIEF REPORT Understanding the Disease Course and Therapeutic Benefit of Tafamidis Across Real-World Studies of Hereditary Transthyretin Amyloidosis with Polyneuropathy: A Proof of Concept for Integrative Data Analytic Approaches . . Daniel Serrano Christopher B. Atzinger Marc F. Botteman Received: February 21, 2018 / Published online: April 2, 2018 The Author(s) 2018 data. IDA permits increased understanding of ABSTRACT outcomes in tafamidis-treated and untreated persons with hATTR-PN by optimally pooling all Introduction: Hereditary transthyretin (TTR) available information. amyloidosis with polyneuropathy (hATTR-PN) is Methods: Summary statistics corresponding to a rare, autosomal dominant amyloidosis charac- the Neuropathy Impairment Score-Lower Limb terized primarily by progressive ascending sen- (NIS-LL) from five published studies were sorimotor neuropathy often associated with pooled, converted to change from baseline autonomic involvement. hATTR-PN is caused by means and variances, and analyzed using IDA. a mutation in the TTR gene leading to protein IDA-based synthetic cohorts were generated by misfolding and amyloid accumulation in averaging across studies stratified on treatment peripheral nerves and vital organs. The latest versus control cohort. Trends in change from global prevalence estimates point to 10,000 cases baseline in each study and the corresponding worldwide, with an upper end of about 40,000. synthetic cohorts were plotted. Patient-level Tafamidis has been approved in over 40 countries data were simulated from the synthetic cohort for delaying neurologic disease progression in trends in a Monte Carlo simulation to highlight early-stage hATTR-PN. Multiple observational the ability to contrast synthetic cohort trends studies have examined clinical outcomes in using the mixed model for repeated measures hATTR-PN patients treated with tafamidis in the (MMRM). routine clinical setting. Integrative data analysis Results: The average sample size among the (IDA) is a technique for optimally constructing five studies was 71 (37–128) patients. The aver- synthetic treatment and control cohorts from age NIS-LL trends indicated that tafamidis- multiple independent studies, which allows treated patients experienced slower progression meta-analysis of patient-level data. Herein, we in neuropathy compared to untreated patients. provide a proof of concept for the application of Synthetic cohort trends reflected the trends IDA to real-world and natural history hATTR-PN observed in the contributing studies, while simultaneously shrinking the width of corre- Enhanced content To view enhanced content for this article go to https://doi.org/10.6084/m9.figshare.59997 sponding confidence bands. Monte Carlo sim- ulation results demonstrated precise recovery of the synthetic cohort and time-dependent sim- D. Serrano  C. B. Atzinger  M. F. Botteman (&) ulated NIS-LL means by the MMRM. Pharmerit International, 4350 East-West Highway, Discussion: This proof of concept demonstrates Suite 1110, Bethesda, MD 20814, USA the utility of IDA-based synthetic cohorts for e-mail: mbotteman@pharmerit.com 142 Neurol Ther (2018) 7:141–154 increased precision in characterizing and testing INTRODUCTION hypotheses about treatment outcomes and prognosis in hATTR-PN. Hereditary transthyretin (TTR) amyloidosis with Funding: Pfizer. polyneuropathy (hATTR-PN) is a rare, autoso- Plain Language Summary: Plain language mal dominant, systemic amyloidosis that is summary available for this article. characterized primarily by progressive ascend- ing sensorimotor neuropathy, with or without autonomic involvement, although mixed phe- notypes are common [1, 2]. The central estimate PLAIN LANGUAGE SUMMARY of global hATTR-PN prevalence is approxi- mately 10,000 persons, but it may be as high as Hereditary transthyretin (TTR) amyloidosis with 40,000 [3]. hATTR-PN is traditionally catego- polyneuropathy (hATTR-PN) is a rare inherited rized as either Val30Met [4] or non-Val30Met. disease. It is caused by genetic mutations that The former is the most common variant glob- change the structure of TTR proteins, causing an ally [5]. These mutations make TTR tetramers abnormal buildup of amyloid protein deposits prone to dissociating into monomers that (amyloidosis) in your body’s nerves and organs. undergo misfolding due to their physical struc- This damages your nerves and organs and cau- ture; the misfolded proteins aggregate into ses weakness, numbness, and pain. The drug insoluble amyloid fibrils that are deposited on tafamidis stabilizes TTR proteins and slows the peripheral nerves and in vital organs, leading to disease’s progression. Tafamidis has been the symptoms of hATTR-PN. If untreated, the approved in over 40 countries based on clinical average survival is 10–15 years after symptom trial results. However, researchers and doctors onset [5–8]. are still studying how it works for patients in Tafamidis is a selective TTR stabilizer that the real world. Over the past few years, holds TTR tetramers together to prevent for- researchers have published multiple real-world mation of misfolded TTR, and is approved in studies about tafamidis, but it is difficult to get a over 40 countries to delay neurologic disease full, uniform picture of how well tafamidis progression in early-stage hATTR-PN [9]. The works, because the studies are too different. tafamidis clinical development program hATTR-PN is a rare disease, so the number of demonstrated the drug’s long-term safety and patients per study is small; small sample sizes effectiveness in delaying hATTR-PN disease can make it more difficult to tell true effects progression for up to 5.5 years [10–15], with from statistical noise, whereas large samples can comparable outcomes observed in Val30Met be more precise. Integrative data analysis (IDA) and non-Val30Met patients compared to pla- is a technique that allows researchers to com- cebo [16]. bine the results of multiple studies into a large Disease progression in hATTR-PN is typically pool of data and analyze the larger data set (or measured according to standardized staging ‘‘synthetic cohort’’) instead. This makes optimal criteria that reflect the severity of systematic use of each study’s information and creates a neurological involvement. One of the most fuller picture of the treatment’s real-world frequently used staging systems is the results. This manuscript is a ‘‘proof of concept’’ polyneuropathy disability (PND) score [17], to demonstrate how the IDA statistical method which ranges from stage 0 (no impairment) to can be used to build synthetic cohorts from real- stage IV (confined to a wheelchair or world hATTR-PN data, improving our under- bedridden). standing of hATTR-PN patients’ outcomes. Since the initial approval of tafamidis in 2011 by the European Medicines Agency [18], various observational open-label studies have Keywords: Integrated data analysis; Hereditary assessed its effectiveness among samples com- transthyretin amyloidosis; Meta-analysis; posed predominantly of stage I patients in the Tafamidis Neurol Ther (2018) 7:141–154 143 routine clinical (i.e., ‘‘real world’’) setting and control/natural history studies to create [10, 19–21]. Several key characteristics were synthetic treatment and control arms. These consistent across these studies, including the synthetic treatment and control cohorts can assessments used to measure neuropathy pro- then be contrasted to determine the time-de- gression and the duration of assessment inter- pendent value of therapeutic intervention. vals. However, the mutant variant distributions, These synthetic cohorts yield greater precision ages of onset, and timing of treatment initiation by increasing sample size while shrinking error relative to disease stage differed among the variance. studies. This inter-study heterogeneity—in Four extant studies were selected for analysis addition to small sample sizes, different ana- in addition to the tafamidis registration study, lytical approaches, and variable follow-up because each was among the largest recent times—has made it difficult to interpret the studies and contributed comprehensive exami- uniformity of the effect of tafamidis on hATTR- nations of the relationship between disease PN progression. Key unresolved questions progression and tafamidis treatment in a man- include whether progression and treatment ner commensurate with the approach taken in response differ between mutation type, age of the registration pivotal study [23]. Note that onset, and/or disease staging schemes. only three of the four were completely inde- We present herein a proof-of-concept study pendent samples, and that the Coelho et al. [10] and applied example of a statistical method that cohort considered here was the tafamidis can be used to pool real-world and randomized crossover extension of the placebo arm in the trial tafamidis study data. Methodological registration trial. Table 1 provides a summary of details are provided in an overview, and an these studies. applied example is described. The example The five included studies were used to char- constructs synthetic cohorts from summary acterize study-specific trends in average change statistics reported in the literature, and then from baseline in Neuropathy Impairment Score- contrasts the generated synthetic cohorts in a Lower Limb (NIS-LL) scores. In addition, the mixed model for repeated measures (MMRM) to study-specific trends were averaged within characterize therapeutic benefit. treatment arms to construct synthetic cohorts for treatment and controls (i.e., natural history cohorts). Averages were used to construct the METHODS synthetic cohort trends because more sophisti- cated pooling procedures are not available with Integrative data analysis (IDA) [22] is a statisti- summary data. The averaging procedure was cal pooling method to combine studies and used to serve as a proof of concept for a forth- then construct synthetic treatment and control coming work in which patient-level data from cohorts. Aggregation of heterogeneous studies some of the studies described herein (as well as into synthetic cohorts can be thought of as a others) will be pooled to create synthetic meta-analytic technique for raw data. Optimal cohorts for direct analysis of treatment versus weighting and scaling techniques are used to control/natural history cohorts. produce a synthetic cohort from each individ- Summary statistics reported in each of the ual study that up-weights each study’s unique included studies were used to obtain or con- and usable information while simultaneously struct study-stratified change from baseline down-weighting its idiosyncratic noise. Avail- means and corresponding confidence limits. In able optimal pooling techniques range from the some studies—notably the 2012 registration use of fixed and random study effects to inverse study by Coelho et al. [23]—change from base- probability weighting (IPW) through propensity line means and 95% confidence limits were not methods. This produces a synthetic cohort that tabulated but rather were presented in fig- is maximally representative of each study’s ures only. In such cases, tracing software was useful information. These techniques can be used to recover as precisely as possible the used to aggregate data from treatment studies numerical values presented in the figure. For the 144 Neurol Ther (2018) 7:141–154 Table 1 Summary of trial-based and real-world prospective studies in patients with hATTR-PN treated with tafamidis Cortese et al. [19] Coelho et al. [23] Plante- Lozeron et al. [20] Coelho et al. [10] Bordeneuve et al. [21] Study Multicenter, Multicenter, Single center, Single center, Multicenter, design observational interventional observational observational interventional Country Italy Global France France Global No. of 61 125 43 37 33 patients Male 69% 50% tafamidis, 43% 56% 67% 45% placebo b b c Mean age 59 36 59 58 36 at onset, Duration 3.4 3.4 3.3 4.0 3.1 of disease, y Val30Met 28% 100% 47% 100% 100% Stage I 72% NR NR 67% NR Follow-up, 36 18 36 12 12 mo Key 33% of pts remained In the efficacy At 6-12 mo, At 6 mo, 38% of 29 In pts switched from outcomes stable and did not evaluable population 58% (25/43) evaluable pts showed placebo, the show significant (n = 87), of pts no meaningful monthly rate of progression, significantly more showed a progression in NIS- change in NIS-LL regardless of tafamidis pts than response to LL (progression declined mutation type and placebo pts had\ 2 tafamidis defined as change Pts treated with baseline disease point NIS-LL from baseline C 2 At 30-36 mo, tafamidis for 30 mo stage worsening from points) 9% (2/22) had 55.9% greater baseline (60% vs. Neuropathy and were still Of 13 pts evaluated at preservation of 38%; p = 0.041) cardiomyopathy stable 12 mo, 69% had no neurologic function progressed in a Tafamidis pts had meaningful than pts in whom proportion of pts better-preserved progression in NIS- tafamidis was despite treatment TQOL LL initiated later hATTR-PN hereditary transthyretin amyloidosis with polyneuropathy, mo months, NIS-LL Neuropathy Impairment Score-Lower Limb, NR not reported, pts patients, TQOL total quality of life, Val30Met methionine replacement of valine at position 30 in the TTR gene, y years Notes Only the placebo-tafamidis arm was used from this study Age of onset computed from reported age minus disease duration Computed from reported median age minus median symptom duration Median symptom duration Neurol Ther (2018) 7:141–154 145 control group, the only published data were For these data, the change from baseline statis- from the placebo arm of the 18-month tafami- tics were computed as a function of the distri- dis registration study [23]. A separate cohort of bution for the difference in Gaussian variables controls was simulated that behaved in a man- scaled by a constant. Once the SD was com- ner consistent with our expectation of natural puted, 95% Wald confidence limits were com- history disease progression in neuropathy. puted for the corresponding standard error. This Specifically, this simulated cohort bore the procedure yielded change from baseline means characteristics of the 2012 registration study’s (l ) and corresponding 95% confidence limits. placebo arm, except that it had worse progres- These estimates were averaged to construct the sion (to reflect an assumed attenuation of any tabulated and plotted synthetic cohort average placebo effect) and it included a projection up trend and corresponding 95% Wald confidence to 30 months. limits, stratified by treatment arm. The outcome measure for this exercise was the Multivariate normal data were simulated average change from baseline in NIS-LL scores and from the synthetic cohort time-dependent corresponding 95% confidence limits. However, treatment-arm-stratified means and variances not all studies reported NIS-LL scores in the using the ‘‘mvrnorm’’ function in R version change from baseline metric, nor did they neces- 3.4.3 [24]. The simulated data were constructed sarily report 95% confidence limits when change under a balanced design with n = 100 patients from baseline means were available. Where aver- per cohort and complete data in repeated mea- age change from baseline and corresponding 95% sures from month 6 through month 30 at confidence limits were reported (i.e., Coelho et al. 6-month intervals. The variances were used to [10, 23] and Cortese et al. [19]), the statistics were construct treatment-arm-stratified unstructured used directly. Where time-specific NIS-LL means covariance matrices. In addition, correlated and standard deviations were reported (i.e., baseline covariates were simulated for both Lozeron et al. [20] and Plante´-Bordeneuve treatment arms generated to have a mean of 5 et al. [21]), change from baseline means and and a standard deviation of 2. Simulated change standard deviations were computed using prop- from baseline data was modeled via the MMRM. erties of the distribution for the difference in cor- This model was estimated using the MIXED related Gaussian variables. Specifically, if Y1 and procedure in SAS 9.4 software [25]. The model Y2 are correlated Gaussian vectors, with distribu- was parameterized using reference cell coding, 2 2 treating the synthetic placebo as reference for tion Y * N (l , r )and Y * N (l , r ), then 1 1 2 2 1 2 2 2 the treatment effect and month 6 assessment as (Y - Y ) * N (l  l ; r þ r þ 2qr r ), with 2 1 1 2 2 1 1 2 reference for time effect with continuous base- corresponding standard deviation of qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi line covariate. Least squares means (LSMs) were 2 2 SD ¼ r þ r þ 2qr r . In every case where D 1 2 1 2 estimated for each treatment by assessment change from baseline had to be constructed from level. The estimated LSMs were then plotted time-dependent means and standard deviations, over the observed estimated synthetic cohort the correlation coefficient, q, was not reported. In means to assess the model’s ability to recover these cases, q was conservatively estimated as 0.4 the observed assessment- and treatment-de- for the purpose of approximating the difference pendent means. This last component was con- standard deviation. ducted as part of the proof of concept to In the case of the Plante-Bordeneuve demonstrate that the model proposed for anal- et al. [21] data, time-dependent means and ysis of the final synthetic cohort data would standard deviations were reported for the NIS successfully recover the functional form and but not the NIS-LL. As the NIS-LL is a subset of observed means with acceptable precision. the NIS, NIS-LL estimates were approximated Fixed-effect point and interval estimates and from these summary statistics by scaling them variance components are not reported. to a range consistent with that observed for the Tables were generated using the REPORT other NIS-LL data. Specifically, the means were procedure in SAS 9.4 software, while fig- divided by 5.4 and the variances divided by 2.0. ures were generated using the ‘‘ggplot2’’ package 146 Neurol Ther (2018) 7:141–154 in R version 3.4.3. This article is based on pre- in several figures. Figure 1 presents the NIS-LL viously conducted studies and does not contain change from baseline trends stratified by study. any new data collected from human partici- Two clusters of trends are observed: the Cortese pants or animals. et al. and Lozeron et al. trends were comparable, This research is based on previously con- and the Coelho et al. [23] and Plante´-Bor- ducted studies and does not contain any studies deneuve et al. [21] trends were comparable. In with human participants or animals performed all cases, the confidence bands were wide, by any of the authors. reflecting in part the studies’ small sample sizes, with the exception of Coelho et al. [23], which had a notably larger sample (n = 125) than the RESULTS other studies (mean n = 57). The only slightly outlying trend was associated with Coelho et al. The reported or, in some cases, computed (e.g., [10]. However, the Coelho et al. [10] trend is the data computed for Plante´-Bordeneuve [21]) distinct, since the original placebo cohort from values are tabulated for review in Table 2 (trea- the registration trial was switched to tafamidis ted cohorts) and Table 3 (untreated or placebo treatment for the open-label continuation cohorts). In addition, the estimates are plotted Table 2 Study-stratified NIS-LL change from baseline means for tafamidis treatment cohorts Assessment Mean (95% confidence limits) NIS-LL change from baseline period Coelho Coelho Lozeron Cortese Plante-Bordeneuve Synthetic et al. [23] et al. [10] et al. [20] et al. [19] et al. [21] treatment cohort Baseline 0 (0, 0) 0 (0, 0) 0 (0, 0) 0 (0, 0) 0 (0, 0) 0 (0, 0) Month 6 1.3 (0.7, 1.8) NA 4.8 (3.7, 5.9) 4.5 (2.9, 6.1) 1.2 (0.5, 1.9) 2.9 (2.6, 3.3) Month 12 1.4 (0.5, 2.1) NA 6.6 (3.8, 9.4) 5.9 (3.6, 8.2) NA 4.6 (4.1, 5.2) Month 18 2.8 (1.9, 3.8) NA NA 8.0 (4.6, 11.4) 2.3 (0.9, 3.7) 4.4 (3.7, 5.0) Month 24 2.5 (1.6, 3.4) 7.8 (5.7, 9.7) NA NA NA 5.1 (4.4, 5.9) Month 30 3.0 (1.5, 4.4) 6.8 (5.1, 8.5) NA NA 4.3 (2.4, 6.2) 4.7 (4.3, 5.1) NA not assessed (per study design), NIS-LL Neuropathy Impairment Score-Lower Limb Table 3 Study-stratified NIS-LL change from baseline means for cohorts not receiving tafamidis Assessment period Mean (95% confidence limits) change from baseline Coelho et al. [23] Natural history simulation Synthetic control cohort Baseline 0 (0, 0) 0 (0, 0) 0 (0, 0) Month 6 2.0 (1.5, 2.7) 3.2 (2.4, 4.0) 2.6 (2.4, 2.8) Month 12 4.7 (3.9, 5.6) 6.2 (5.9, 6.6) 5.5 (5.2, 5.7) Month 18 5.8 (4.9, 6.8) 7.9 (7.4, 8.5) 6.9 (6.6, 7.1) Month 24 NA 10.5 (9.2, 11.7) 10.5 (10.1, 10.9) Month 30 NA 12.8 (11.4, 14.1) 12.8 (12.3, 13.3) NA not assessed (per study design), NIS-LL Neuropathy Impairment Score-Lower Limb Neurol Ther (2018) 7:141–154 147 Fig. 1 Study-stratified mean (95% confidence limits) NIS-LL change from baseline trend for tafamidis treatment cohorts. BL baseline, M month, NIS-LL Neuropathy Impairment Score-Lower Limb, TX treatment study; the plotted trend is the change from uncontrolled in the untreated synthetic cohort baseline in NIS-LL scores post-crossover. and progression slows within the tafamidis- Broadly, the trends demonstrate a slowing of treated synthetic cohort. disease progression in NIS-LL associated with Within Fig. 6, the observed values and cor- tafamidis. The average of these study-specific responding colors reported in Fig. 5 are retained trends is presented in Fig. 2. The average trend, (control = gray; tafamidis = black). These trends plotted in black, fits through the center of all are overlaid with the model-estimated trends, study-specific trends, with a shape consistent which are also color-coded (control = orange; with the Gompertz function suggested as tafamidis = blue). As seen in Fig. 6, the observed appropriate for the NIS-LL data in hATTR-PN synthetic cohort means (OBS) were precisely [26]. The same process was used to generate recovered by the MMRM-based values (LSMs). Figs. 3 and 4 for the Coelho et al. placebo arm Notably, the discrepancy in estimates was zero [23] and the simulated natural history data. The between baseline and month 6 in both treat- treatment and placebo synthetic cohort trends ment and placebo synthetic cohorts, and zero were plotted together in Fig. 5. The trends between month 18 and month 24 for the pla- overlap early, but as expected, diverge around cebo synthetic cohort. All other discrepancies month 12, as disease progression is were minor, and none evinced a departure from 148 Neurol Ther (2018) 7:141–154 Fig. 2 Study-stratified mean (95% confidence limits) NIS- baseline, M month, NIS-LL Neuropathy Impairment LL change from baseline trend for tafamidis treatment Score-Lower Limb, TX treatment cohorts, overlaying synthetic treatment cohort trend. BL the observed functional form. Thus, the dis- variance, as measured by the width of the 95% crete-time MMRM is expected to precisely confidence bands, shrank relative to any indi- recover the observed means in the forthcoming vidual study, but not excessively so. In addition, analyses. within the treatment synthetic cohort, the average trend and confidence bands mimicked a Gompertz function, which is a well-known DISCUSSION function for modeling decelerating exponential effects that asymptote asymmetrically. This is of In this work, a synthetic cohort approach was interest as the Gompertz function has been applied to the analysis of real-world outcomes proposed elsewhere for the analysis of neu- for tafamidis for the treatment of hATTR-PN, rodegenerative outcome measures within including comparison to natural history data. hATTR-PN [26]. Our findings demonstrate the merits of While the Gompertz function may be a good employing synthetic cohorts. The average trend approximation to the average trend, one might lines for the synthetic cohorts did not distort encounter difficulty in properly specifying the any of the study-specific trends. The error model in the context of repeated measures and Neurol Ther (2018) 7:141–154 149 Fig. 3 Study-stratified mean (95% confidence limits) NIS-LL change from baseline trend for natural history cohorts. BL baseline, CTRL control, M month, NIS-LL Neuropathy Impairment Score-Lower Limb random effects. In contrast, a discrete-time have artificially limited the variance and caused MMRM is easily parameterized and can flexibly the confidence limits to be underestimated. accommodate non-linear trends. Therefore, a Summary statistics available in the literature second part of this proof of concept was to were used, limiting the methods available for demonstrate that if individual-level data were optimally weighting the pooling procedure simulated from synthetic cohort means and averaging across studies. variances, a discrete-time MMRM could pre- In addition, three of the five studies consid- cisely recover the synthetic cohort-stratified ered were composed of samples that were 100% mean trends. In fact, the discrete-time MMRM Val30Met. The remaining two studies com- did succeed in recovering the observed means, prised mixed samples (\ 50% Val30Met). Pub- pointing to the ability of this model to detect lished evidence has suggested that progression and accurately reflect synthetic cohort treat- and treatment response outcomes differ sub- ment arm differences in NIS-LL disease stantially between Val30Met and non-Val30Met progression. populations [27–29]. However, the recent anal- This proof-of-concept report has some limi- ysis by Gundapaneni et al. [16] suggested that tations. The included studies may have had progression and treatment responses were no some overlap in the patient samples, which may different between Val30Met and non-Val30Met 150 Neurol Ther (2018) 7:141–154 Fig. 4 Study-stratified mean (95% confidence limits) NIS- CTRL control, M month, NIS-LL Neuropathy Impair- LL change from baseline trend for non-tafamidis cohorts, ment Score-Lower Limb overlaying synthetic control cohort trend. BL baseline, populations treated with tafamidis, after techniques (i.e., responder analyses). It is our adjusting for baseline neuropathy status. A contention that a precise answer is likely limitation of this proof-of-concept study is its achievable only by optimally pooling available inability to address the difference in progression data via IDA, and this issue speaks to the need to and treatment response between these impor- conduct this pooling research. tant sub-populations. However, given the The next step in this line of research is to modest sample sizes in the real-world data apply similar methods to the raw data corre- available to date, no single study has been able sponding to a larger set of real-world data to do this either. As a consequence, given the studies. Doing so will allow for an IDA approach findings of Gundapaneni et al., a new question [22, 30] in which the patient-level data from a that IDA may be uniquely positioned to answer group of independent studies is pooled, rather is whether mutation-dependent progression than the aggregate. With patient-level data, and treatment differences are important, or more sophisticated and sensitive methods of whether they are artifacts arising from modest pooling studies under optimal weighting para- sample sizes and potentially insensitive analysis digms can be employed. These include, but are Neurol Ther (2018) 7:141–154 151 Fig. 5 Synthetic cohort-stratified mean (95% confidence limits) NIS-LL change from baseline trend. BL baseline, CTRL control, M month, NIS-LL Neuropathy Impairment Score-Lower Limb, Tx treatment not limited to, incorporation of fixed and ran- CONCLUSION dom study effects, propensity-matching proce- dures, and the preferable hybrid of these Beyond the registration trial, evidence pub- approaches (i.e., doubly robust propensity lished to date on the natural history, disease weighting). With access to individual patient progression, and tafamidis treatment outcomes characteristics and outcomes, more sophisti- associated with hATTR-PN has demonstrated cated statistical techniques and models can also some heterogeneity and has been derived from be applied to achieve a greater understanding of studies with modest sample sizes (due to the low clinical outcomes by using a unified process prevalence of this disease). IDA and synthetic that adjusts for baseline and changes over time. cohorts are a technique that can be used to By better characterizing the natural history of analyze myriad studies with shared features to untreated hATTR-PN cases and the relative increase the precision of the characterization of benefit of tafamidis treatment, IDA would sig- hATTR-PN treatment outcomes. In so doing, nificantly facilitate clinician–patient commu- modest samples can be aggregated to form large nication regarding available treatment regimens cohorts from which increased precision of and their respective risks and benefits. inference may be obtained. As this is a proof of 152 Neurol Ther (2018) 7:141–154 Fig. 6 Treatment arm-stratified observed synthetic cohort MMRM mixed model for repeated measures, NIS-LL means versus MMRM estimated LSMs. BL baseline, Neuropathy Impairment Score-Lower Limb, OBS observed CTRL control, LSM least squares mean, M month, synthetic cohort, TX treatment concept for the application of IDA to patient- Inc. (New York, USA). All authors had full access level progression data in hATTR-PN, no definite to all of the data in this study and take complete conclusions about the effectiveness of tafamidis responsibility for the integrity of the data and can be made from these results. Rather, one can accuracy of the data analysis. only conclude from the evidence presented Medical Writing and/or Editorial Assis- herein whether IDA is a method that may be tance. Editorial assistance in the preparation of useful in the future for characterizing disease this article was provided by Catherine A. progression and drug effectiveness in a larger O’Connor and Caroline Seo of Pharmerit Inter- cohort using patient-level data. national. Support for this assistance was funded by Pfizer, Inc. ACKNOWLEDGEMENTS Authorship. All named authors meet the International Committee of Medical Journal Editors (ICMJE) criteria for authorship for this Funding. Sponsorship for this research and article, take responsibility for the integrity of article processing fees were funded by Pfizer, Neurol Ther (2018) 7:141–154 153 5. Coelho T, Merlini G, Bulawa CE, et al. Mechanism the work as a whole, and have given their of action and clinical application of Tafamidis in approval for this version to be published. hereditary transthyretin amyloidosis. Neurol Ther. 2016;5(1):1–25. https://doi.org/10.1007/s40120- Disclosures. This research was supported by 016-0040-x. Pfizer, Inc. Marc F. Botteman is a shareholder of 6. Bekircan-Kurt CE, Gunes N, Yilmaz A, Erdem-Oz- Pharmerit International. Daniel Serrano is an damar S, Tan E. Three Turkish families with differ- employee of Pharmerit International. Christo- ent transthyretin mutations. Neuromuscular pher B. Atzinger is an employee of Pharmerit Disord. 2015;25(9):686–92. https://doi.org/10. International. Marc F. Botteman is an employee 1016/j.nmd.2015.05.010. of Pharmerit International. Pharmerit received 7. Bulawa CE, Connelly S, Devit M, et al. Tafamidis, a funding from Pfizer, Inc., for study design, exe- potent and selective transthyretin kinetic stabilizer cution, analysis, and manuscript development. that inhibits the amyloid cascade. Proc Natl Acad Sci USA. 2012;109(24):9629–34. https://doi.org/10. 1073/pnas.1121005109. Compliance with Ethics Guidelines. This research is based on previously conducted 8. Reines JB, Vera TR, Martin MU, et al. Epidemiology studies and does not contain any studies with of transthyretin-associated familial amyloid human participants or animals performed by polyneuropathy in the Majorcan area: Son Llatzer Hospital descriptive study. Orphanet J Rare Dis. any of the authors. 2014;9:29. https://doi.org/10.1186/1750-1172-9- Data availability. All the data used in the analyses are presented in Tables 2 and 3. 9. Benson MD, Kincaid JC. The molecular biology and clinical features of amyloid neuropathy. Muscle Nerve. 2007;36(4):411–23. https://doi.org/10.1002/ Open Access. This article is distributed mus.20821. under the terms of the Creative Commons Attribution-NonCommercial 4.0 International 10. Coelho T, Maia LF, da Silva AM, et al. Long-term License (http://creativecommons.org/licenses/ effects of tafamidis for the treatment of transthyr- by-nc/4.0/), which permits any noncommer- etin familial amyloid polyneuropathy. J Neurol. 2013;260(11):2802–14. https://doi.org/10.1007/ cial use, distribution, and reproduction in any s00415-013-7051-7. medium, provided you give appropriate credit to the original author(s) and the source, provide 11. Keohane D, Schwartz J, Gundapaneni B, Stewart M, a link to the Creative Commons license, and Amass L. Tafamidis reduces disease progression in patients with transthyretin familial amyloid indicate if changes were made. polyneuropathy: supportive post hoc analyses of a pivotal trial. Orphanet J Rare Dis. 2015;10(Suppl 1):P11. https://doi.org/10.1186/1750-1172-10-s1- p11. REFERENCES 12. Barroso FA, Judge DP, Ebede B, et al. Long-term safety and efficacy of tafamidis for the treatment of 1. Benson MD. The hereditary amyloidoses. Best Pract hereditary transthyretin amyloid polyneuropathy: Res Clin Rheumatol. 2003;17(6):909–27. https:// results up to 6 years. Amyloid. 2017;24(3):194–204. doi.org/10.1016/j.berh.2003.09.001. https://doi.org/10.1080/13506129.2017.1357545. 2. Hund E, Linke RP, Willig F, Grau A. Transthyretin- 13. Keohane D, Schwartz J, Gundapaneni B, Stewart M, associated neuropathic amyloidosis.Pathogenesis Amass L. Tafamidis delays disease progression in and treatment. Neurology. 2001;56(4):431–5. patients with early stage transthyretin familial amyloid polyneuropathy: additional supportive 3. Schmidt HH, Waddington-Cruz M, Botteman MF, analyses from the pivotal trial. Amyloid. et al. Estimating the global prevalence of trans- 2017;24(1):30–6. https://doi.org/10.1080/ thyretin familial amyloid polyneuropathy. Muscle 13506129.2017.1301419. Nerve. 2017. https://doi.org/10.1002/mus.26034. 14. Merlini G, Plante-Bordeneuve V, Judge DP, et al. 4. Connors LH, Richardson AM, Theberge R, Costello Effects of tafamidis on transthyretin stabilization CE. Tabulation of transthyretin (TTR) variants as of and clinical outcomes in patients with non-Val30- 1/1/2000. Amyloid. 2000;7(1):54–69. Met transthyretin amyloidosis. J Cardiovasc Transl 154 Neurol Ther (2018) 7:141–154 Res. 2013;6(6):1011–20. https://doi.org/10.1007/ 22. Curran PJ, Hussong AM. Integrative data analysis: s12265-013-9512-x. the simultaneous analysis of multiple data sets. Psychol Methods. 2009;14(2):81. 15. Waddington Cruz M, Amass L, Keohane D, et al. Early intervention with tafamidis provides long- 23. Coelho T, Maia LF, Martins da Silva A, et al. Tafa- term (5.5-year) delay of neurologic progression in midis for transthyretin familial amyloid polyneu- transthyretin hereditary amyloid polyneuropathy. ropathy: a randomized, controlled trial. Neurology. Amyloid. 2016;23(3):178–83. https://doi.org/10. 2012;79(8):785–92. https://doi.org/10.1212/WNL. 1080/13506129.2016.1207163. 0b013e3182661eb1. 16. Gundapaneni BK, Sultan MB, Keohane DJ, Schwartz 24. Wickham H. ggplot2: Elegant graphics for data JH. Tafamidis delays neurological progression analysis. New York: Springer; 2009. comparably across Val30Met and non-Val30Met genotypes in transthyretin familial amyloid 25. SAS version 9.4. SAS Institute: Cary, NC, USA; 2013. polyneuropathy. Eur J Neurol. 2017. https://doi. org/10.1111/ene.13510. 26. Adams D, Coelho T, Obici L, et al. Rapid progres- sion of familial amyloidotic polyneuropathy: a 17. Ando Y, Coelho T, Berk JL, et al. Guideline of multinational natural history study. Neurology. transthyretin-related hereditary amyloidosis for 2015;85(8):675–82. https://doi.org/10.1212/WNL. clinicians. Orphanet J Rare Dis. 2013;8:31. https:// 0000000000001870. doi.org/10.1186/1750-1172-8-31. 27. Ericzon BG, Wilczek HE, Larsson M, et al. Liver transplantation for hereditary transthyretin amy- 18. VYNDAQEL [summary of opinion—initial authori- loidosis: after 20 years still the best therapeutic sation] [online]. London: European Medicines alternative? Transplantation. 2015;99(9):1847–54. Agency 2011-21-July Contract No.: EMA/CHMP/ https://doi.org/10.1097/TP.0000000000000574. 546201/2011. 28. Kerschen P, Plante-Bordeneuve V. Current and 19. Cortese A, Vita G, Luigetti M, et al. Monitoring effectiveness and safety of Tafamidis in transthyr- future treatment approaches in transthyretin etin amyloidosis in Italy: a longitudinal multicenter familial amyloid polyneuropathy. Curr Treat study in a non-endemic area. J Neurol. Options Neurol. 2016;18(12):53. https://doi.org/10. 2016;263(5):916–24. https://doi.org/10.1007/ 1007/s11940-016-0436-z. s00415-016-8064-9. 29. Suhr OB, Larsson M, Ericzon BG, Wilczek HE, 20. Lozeron P, Theaudin M, Mincheva Z, et al. Effect on FAPWTR’s investigators. Survival after transplanta- disability and safety of Tafamidis in late onset of tion in patients with mutations other than Val30- Met30 transthyretin familial amyloid polyneu- Met: extracts from the FAP world transplant ropathy. Eur J Neurol. 2013;20(12):1539–45. registry. Transplantation. 2016;100(2):373–81. https://doi.org/10.1111/ene.12225. https://doi.org/10.1097/TP.0000000000001021. 21. Plante-Bordeneuve V, Gorram F, Salhi H, et al. 30. Hofer SM, Piccinin AM. Integrative data analysis Long-term treatment of transthyretin familial through coordination of measurement and analysis amyloid polyneuropathy with tafamidis: a clinical protocol across independent longitudinal studies. and neurophysiological study. J Neurol. Psychol Methods. 2009;14(2):150. 2017;264(2):268–76. https://doi.org/10.1007/ s00415-016-8337-3. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Neurology and Therapy Springer Journals

Understanding the Disease Course and Therapeutic Benefit of Tafamidis Across Real-World Studies of Hereditary Transthyretin Amyloidosis with Polyneuropathy: A Proof of Concept for Integrative Data Analytic Approaches

Free
14 pages

Loading next page...
 
/lp/springer_journal/understanding-the-disease-course-and-therapeutic-benefit-of-tafamidis-0JdFv7cIA6
Publisher
Springer Journals
Copyright
Copyright © 2018 by The Author(s)
Subject
Medicine & Public Health; Internal Medicine; Neurology
ISSN
2193-8253
eISSN
2193-6536
D.O.I.
10.1007/s40120-018-0096-x
Publisher site
See Article on Publisher Site

Abstract

Neurol Ther (2018) 7:141–154 https://doi.org/10.1007/s40120-018-0096-x BRIEF REPORT Understanding the Disease Course and Therapeutic Benefit of Tafamidis Across Real-World Studies of Hereditary Transthyretin Amyloidosis with Polyneuropathy: A Proof of Concept for Integrative Data Analytic Approaches . . Daniel Serrano Christopher B. Atzinger Marc F. Botteman Received: February 21, 2018 / Published online: April 2, 2018 The Author(s) 2018 data. IDA permits increased understanding of ABSTRACT outcomes in tafamidis-treated and untreated persons with hATTR-PN by optimally pooling all Introduction: Hereditary transthyretin (TTR) available information. amyloidosis with polyneuropathy (hATTR-PN) is Methods: Summary statistics corresponding to a rare, autosomal dominant amyloidosis charac- the Neuropathy Impairment Score-Lower Limb terized primarily by progressive ascending sen- (NIS-LL) from five published studies were sorimotor neuropathy often associated with pooled, converted to change from baseline autonomic involvement. hATTR-PN is caused by means and variances, and analyzed using IDA. a mutation in the TTR gene leading to protein IDA-based synthetic cohorts were generated by misfolding and amyloid accumulation in averaging across studies stratified on treatment peripheral nerves and vital organs. The latest versus control cohort. Trends in change from global prevalence estimates point to 10,000 cases baseline in each study and the corresponding worldwide, with an upper end of about 40,000. synthetic cohorts were plotted. Patient-level Tafamidis has been approved in over 40 countries data were simulated from the synthetic cohort for delaying neurologic disease progression in trends in a Monte Carlo simulation to highlight early-stage hATTR-PN. Multiple observational the ability to contrast synthetic cohort trends studies have examined clinical outcomes in using the mixed model for repeated measures hATTR-PN patients treated with tafamidis in the (MMRM). routine clinical setting. Integrative data analysis Results: The average sample size among the (IDA) is a technique for optimally constructing five studies was 71 (37–128) patients. The aver- synthetic treatment and control cohorts from age NIS-LL trends indicated that tafamidis- multiple independent studies, which allows treated patients experienced slower progression meta-analysis of patient-level data. Herein, we in neuropathy compared to untreated patients. provide a proof of concept for the application of Synthetic cohort trends reflected the trends IDA to real-world and natural history hATTR-PN observed in the contributing studies, while simultaneously shrinking the width of corre- Enhanced content To view enhanced content for this article go to https://doi.org/10.6084/m9.figshare.59997 sponding confidence bands. Monte Carlo sim- ulation results demonstrated precise recovery of the synthetic cohort and time-dependent sim- D. Serrano  C. B. Atzinger  M. F. Botteman (&) ulated NIS-LL means by the MMRM. Pharmerit International, 4350 East-West Highway, Discussion: This proof of concept demonstrates Suite 1110, Bethesda, MD 20814, USA the utility of IDA-based synthetic cohorts for e-mail: mbotteman@pharmerit.com 142 Neurol Ther (2018) 7:141–154 increased precision in characterizing and testing INTRODUCTION hypotheses about treatment outcomes and prognosis in hATTR-PN. Hereditary transthyretin (TTR) amyloidosis with Funding: Pfizer. polyneuropathy (hATTR-PN) is a rare, autoso- Plain Language Summary: Plain language mal dominant, systemic amyloidosis that is summary available for this article. characterized primarily by progressive ascend- ing sensorimotor neuropathy, with or without autonomic involvement, although mixed phe- notypes are common [1, 2]. The central estimate PLAIN LANGUAGE SUMMARY of global hATTR-PN prevalence is approxi- mately 10,000 persons, but it may be as high as Hereditary transthyretin (TTR) amyloidosis with 40,000 [3]. hATTR-PN is traditionally catego- polyneuropathy (hATTR-PN) is a rare inherited rized as either Val30Met [4] or non-Val30Met. disease. It is caused by genetic mutations that The former is the most common variant glob- change the structure of TTR proteins, causing an ally [5]. These mutations make TTR tetramers abnormal buildup of amyloid protein deposits prone to dissociating into monomers that (amyloidosis) in your body’s nerves and organs. undergo misfolding due to their physical struc- This damages your nerves and organs and cau- ture; the misfolded proteins aggregate into ses weakness, numbness, and pain. The drug insoluble amyloid fibrils that are deposited on tafamidis stabilizes TTR proteins and slows the peripheral nerves and in vital organs, leading to disease’s progression. Tafamidis has been the symptoms of hATTR-PN. If untreated, the approved in over 40 countries based on clinical average survival is 10–15 years after symptom trial results. However, researchers and doctors onset [5–8]. are still studying how it works for patients in Tafamidis is a selective TTR stabilizer that the real world. Over the past few years, holds TTR tetramers together to prevent for- researchers have published multiple real-world mation of misfolded TTR, and is approved in studies about tafamidis, but it is difficult to get a over 40 countries to delay neurologic disease full, uniform picture of how well tafamidis progression in early-stage hATTR-PN [9]. The works, because the studies are too different. tafamidis clinical development program hATTR-PN is a rare disease, so the number of demonstrated the drug’s long-term safety and patients per study is small; small sample sizes effectiveness in delaying hATTR-PN disease can make it more difficult to tell true effects progression for up to 5.5 years [10–15], with from statistical noise, whereas large samples can comparable outcomes observed in Val30Met be more precise. Integrative data analysis (IDA) and non-Val30Met patients compared to pla- is a technique that allows researchers to com- cebo [16]. bine the results of multiple studies into a large Disease progression in hATTR-PN is typically pool of data and analyze the larger data set (or measured according to standardized staging ‘‘synthetic cohort’’) instead. This makes optimal criteria that reflect the severity of systematic use of each study’s information and creates a neurological involvement. One of the most fuller picture of the treatment’s real-world frequently used staging systems is the results. This manuscript is a ‘‘proof of concept’’ polyneuropathy disability (PND) score [17], to demonstrate how the IDA statistical method which ranges from stage 0 (no impairment) to can be used to build synthetic cohorts from real- stage IV (confined to a wheelchair or world hATTR-PN data, improving our under- bedridden). standing of hATTR-PN patients’ outcomes. Since the initial approval of tafamidis in 2011 by the European Medicines Agency [18], various observational open-label studies have Keywords: Integrated data analysis; Hereditary assessed its effectiveness among samples com- transthyretin amyloidosis; Meta-analysis; posed predominantly of stage I patients in the Tafamidis Neurol Ther (2018) 7:141–154 143 routine clinical (i.e., ‘‘real world’’) setting and control/natural history studies to create [10, 19–21]. Several key characteristics were synthetic treatment and control arms. These consistent across these studies, including the synthetic treatment and control cohorts can assessments used to measure neuropathy pro- then be contrasted to determine the time-de- gression and the duration of assessment inter- pendent value of therapeutic intervention. vals. However, the mutant variant distributions, These synthetic cohorts yield greater precision ages of onset, and timing of treatment initiation by increasing sample size while shrinking error relative to disease stage differed among the variance. studies. This inter-study heterogeneity—in Four extant studies were selected for analysis addition to small sample sizes, different ana- in addition to the tafamidis registration study, lytical approaches, and variable follow-up because each was among the largest recent times—has made it difficult to interpret the studies and contributed comprehensive exami- uniformity of the effect of tafamidis on hATTR- nations of the relationship between disease PN progression. Key unresolved questions progression and tafamidis treatment in a man- include whether progression and treatment ner commensurate with the approach taken in response differ between mutation type, age of the registration pivotal study [23]. Note that onset, and/or disease staging schemes. only three of the four were completely inde- We present herein a proof-of-concept study pendent samples, and that the Coelho et al. [10] and applied example of a statistical method that cohort considered here was the tafamidis can be used to pool real-world and randomized crossover extension of the placebo arm in the trial tafamidis study data. Methodological registration trial. Table 1 provides a summary of details are provided in an overview, and an these studies. applied example is described. The example The five included studies were used to char- constructs synthetic cohorts from summary acterize study-specific trends in average change statistics reported in the literature, and then from baseline in Neuropathy Impairment Score- contrasts the generated synthetic cohorts in a Lower Limb (NIS-LL) scores. In addition, the mixed model for repeated measures (MMRM) to study-specific trends were averaged within characterize therapeutic benefit. treatment arms to construct synthetic cohorts for treatment and controls (i.e., natural history cohorts). Averages were used to construct the METHODS synthetic cohort trends because more sophisti- cated pooling procedures are not available with Integrative data analysis (IDA) [22] is a statisti- summary data. The averaging procedure was cal pooling method to combine studies and used to serve as a proof of concept for a forth- then construct synthetic treatment and control coming work in which patient-level data from cohorts. Aggregation of heterogeneous studies some of the studies described herein (as well as into synthetic cohorts can be thought of as a others) will be pooled to create synthetic meta-analytic technique for raw data. Optimal cohorts for direct analysis of treatment versus weighting and scaling techniques are used to control/natural history cohorts. produce a synthetic cohort from each individ- Summary statistics reported in each of the ual study that up-weights each study’s unique included studies were used to obtain or con- and usable information while simultaneously struct study-stratified change from baseline down-weighting its idiosyncratic noise. Avail- means and corresponding confidence limits. In able optimal pooling techniques range from the some studies—notably the 2012 registration use of fixed and random study effects to inverse study by Coelho et al. [23]—change from base- probability weighting (IPW) through propensity line means and 95% confidence limits were not methods. This produces a synthetic cohort that tabulated but rather were presented in fig- is maximally representative of each study’s ures only. In such cases, tracing software was useful information. These techniques can be used to recover as precisely as possible the used to aggregate data from treatment studies numerical values presented in the figure. For the 144 Neurol Ther (2018) 7:141–154 Table 1 Summary of trial-based and real-world prospective studies in patients with hATTR-PN treated with tafamidis Cortese et al. [19] Coelho et al. [23] Plante- Lozeron et al. [20] Coelho et al. [10] Bordeneuve et al. [21] Study Multicenter, Multicenter, Single center, Single center, Multicenter, design observational interventional observational observational interventional Country Italy Global France France Global No. of 61 125 43 37 33 patients Male 69% 50% tafamidis, 43% 56% 67% 45% placebo b b c Mean age 59 36 59 58 36 at onset, Duration 3.4 3.4 3.3 4.0 3.1 of disease, y Val30Met 28% 100% 47% 100% 100% Stage I 72% NR NR 67% NR Follow-up, 36 18 36 12 12 mo Key 33% of pts remained In the efficacy At 6-12 mo, At 6 mo, 38% of 29 In pts switched from outcomes stable and did not evaluable population 58% (25/43) evaluable pts showed placebo, the show significant (n = 87), of pts no meaningful monthly rate of progression, significantly more showed a progression in NIS- change in NIS-LL regardless of tafamidis pts than response to LL (progression declined mutation type and placebo pts had\ 2 tafamidis defined as change Pts treated with baseline disease point NIS-LL from baseline C 2 At 30-36 mo, tafamidis for 30 mo stage worsening from points) 9% (2/22) had 55.9% greater baseline (60% vs. Neuropathy and were still Of 13 pts evaluated at preservation of 38%; p = 0.041) cardiomyopathy stable 12 mo, 69% had no neurologic function progressed in a Tafamidis pts had meaningful than pts in whom proportion of pts better-preserved progression in NIS- tafamidis was despite treatment TQOL LL initiated later hATTR-PN hereditary transthyretin amyloidosis with polyneuropathy, mo months, NIS-LL Neuropathy Impairment Score-Lower Limb, NR not reported, pts patients, TQOL total quality of life, Val30Met methionine replacement of valine at position 30 in the TTR gene, y years Notes Only the placebo-tafamidis arm was used from this study Age of onset computed from reported age minus disease duration Computed from reported median age minus median symptom duration Median symptom duration Neurol Ther (2018) 7:141–154 145 control group, the only published data were For these data, the change from baseline statis- from the placebo arm of the 18-month tafami- tics were computed as a function of the distri- dis registration study [23]. A separate cohort of bution for the difference in Gaussian variables controls was simulated that behaved in a man- scaled by a constant. Once the SD was com- ner consistent with our expectation of natural puted, 95% Wald confidence limits were com- history disease progression in neuropathy. puted for the corresponding standard error. This Specifically, this simulated cohort bore the procedure yielded change from baseline means characteristics of the 2012 registration study’s (l ) and corresponding 95% confidence limits. placebo arm, except that it had worse progres- These estimates were averaged to construct the sion (to reflect an assumed attenuation of any tabulated and plotted synthetic cohort average placebo effect) and it included a projection up trend and corresponding 95% Wald confidence to 30 months. limits, stratified by treatment arm. The outcome measure for this exercise was the Multivariate normal data were simulated average change from baseline in NIS-LL scores and from the synthetic cohort time-dependent corresponding 95% confidence limits. However, treatment-arm-stratified means and variances not all studies reported NIS-LL scores in the using the ‘‘mvrnorm’’ function in R version change from baseline metric, nor did they neces- 3.4.3 [24]. The simulated data were constructed sarily report 95% confidence limits when change under a balanced design with n = 100 patients from baseline means were available. Where aver- per cohort and complete data in repeated mea- age change from baseline and corresponding 95% sures from month 6 through month 30 at confidence limits were reported (i.e., Coelho et al. 6-month intervals. The variances were used to [10, 23] and Cortese et al. [19]), the statistics were construct treatment-arm-stratified unstructured used directly. Where time-specific NIS-LL means covariance matrices. In addition, correlated and standard deviations were reported (i.e., baseline covariates were simulated for both Lozeron et al. [20] and Plante´-Bordeneuve treatment arms generated to have a mean of 5 et al. [21]), change from baseline means and and a standard deviation of 2. Simulated change standard deviations were computed using prop- from baseline data was modeled via the MMRM. erties of the distribution for the difference in cor- This model was estimated using the MIXED related Gaussian variables. Specifically, if Y1 and procedure in SAS 9.4 software [25]. The model Y2 are correlated Gaussian vectors, with distribu- was parameterized using reference cell coding, 2 2 treating the synthetic placebo as reference for tion Y * N (l , r )and Y * N (l , r ), then 1 1 2 2 1 2 2 2 the treatment effect and month 6 assessment as (Y - Y ) * N (l  l ; r þ r þ 2qr r ), with 2 1 1 2 2 1 1 2 reference for time effect with continuous base- corresponding standard deviation of qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi line covariate. Least squares means (LSMs) were 2 2 SD ¼ r þ r þ 2qr r . In every case where D 1 2 1 2 estimated for each treatment by assessment change from baseline had to be constructed from level. The estimated LSMs were then plotted time-dependent means and standard deviations, over the observed estimated synthetic cohort the correlation coefficient, q, was not reported. In means to assess the model’s ability to recover these cases, q was conservatively estimated as 0.4 the observed assessment- and treatment-de- for the purpose of approximating the difference pendent means. This last component was con- standard deviation. ducted as part of the proof of concept to In the case of the Plante-Bordeneuve demonstrate that the model proposed for anal- et al. [21] data, time-dependent means and ysis of the final synthetic cohort data would standard deviations were reported for the NIS successfully recover the functional form and but not the NIS-LL. As the NIS-LL is a subset of observed means with acceptable precision. the NIS, NIS-LL estimates were approximated Fixed-effect point and interval estimates and from these summary statistics by scaling them variance components are not reported. to a range consistent with that observed for the Tables were generated using the REPORT other NIS-LL data. Specifically, the means were procedure in SAS 9.4 software, while fig- divided by 5.4 and the variances divided by 2.0. ures were generated using the ‘‘ggplot2’’ package 146 Neurol Ther (2018) 7:141–154 in R version 3.4.3. This article is based on pre- in several figures. Figure 1 presents the NIS-LL viously conducted studies and does not contain change from baseline trends stratified by study. any new data collected from human partici- Two clusters of trends are observed: the Cortese pants or animals. et al. and Lozeron et al. trends were comparable, This research is based on previously con- and the Coelho et al. [23] and Plante´-Bor- ducted studies and does not contain any studies deneuve et al. [21] trends were comparable. In with human participants or animals performed all cases, the confidence bands were wide, by any of the authors. reflecting in part the studies’ small sample sizes, with the exception of Coelho et al. [23], which had a notably larger sample (n = 125) than the RESULTS other studies (mean n = 57). The only slightly outlying trend was associated with Coelho et al. The reported or, in some cases, computed (e.g., [10]. However, the Coelho et al. [10] trend is the data computed for Plante´-Bordeneuve [21]) distinct, since the original placebo cohort from values are tabulated for review in Table 2 (trea- the registration trial was switched to tafamidis ted cohorts) and Table 3 (untreated or placebo treatment for the open-label continuation cohorts). In addition, the estimates are plotted Table 2 Study-stratified NIS-LL change from baseline means for tafamidis treatment cohorts Assessment Mean (95% confidence limits) NIS-LL change from baseline period Coelho Coelho Lozeron Cortese Plante-Bordeneuve Synthetic et al. [23] et al. [10] et al. [20] et al. [19] et al. [21] treatment cohort Baseline 0 (0, 0) 0 (0, 0) 0 (0, 0) 0 (0, 0) 0 (0, 0) 0 (0, 0) Month 6 1.3 (0.7, 1.8) NA 4.8 (3.7, 5.9) 4.5 (2.9, 6.1) 1.2 (0.5, 1.9) 2.9 (2.6, 3.3) Month 12 1.4 (0.5, 2.1) NA 6.6 (3.8, 9.4) 5.9 (3.6, 8.2) NA 4.6 (4.1, 5.2) Month 18 2.8 (1.9, 3.8) NA NA 8.0 (4.6, 11.4) 2.3 (0.9, 3.7) 4.4 (3.7, 5.0) Month 24 2.5 (1.6, 3.4) 7.8 (5.7, 9.7) NA NA NA 5.1 (4.4, 5.9) Month 30 3.0 (1.5, 4.4) 6.8 (5.1, 8.5) NA NA 4.3 (2.4, 6.2) 4.7 (4.3, 5.1) NA not assessed (per study design), NIS-LL Neuropathy Impairment Score-Lower Limb Table 3 Study-stratified NIS-LL change from baseline means for cohorts not receiving tafamidis Assessment period Mean (95% confidence limits) change from baseline Coelho et al. [23] Natural history simulation Synthetic control cohort Baseline 0 (0, 0) 0 (0, 0) 0 (0, 0) Month 6 2.0 (1.5, 2.7) 3.2 (2.4, 4.0) 2.6 (2.4, 2.8) Month 12 4.7 (3.9, 5.6) 6.2 (5.9, 6.6) 5.5 (5.2, 5.7) Month 18 5.8 (4.9, 6.8) 7.9 (7.4, 8.5) 6.9 (6.6, 7.1) Month 24 NA 10.5 (9.2, 11.7) 10.5 (10.1, 10.9) Month 30 NA 12.8 (11.4, 14.1) 12.8 (12.3, 13.3) NA not assessed (per study design), NIS-LL Neuropathy Impairment Score-Lower Limb Neurol Ther (2018) 7:141–154 147 Fig. 1 Study-stratified mean (95% confidence limits) NIS-LL change from baseline trend for tafamidis treatment cohorts. BL baseline, M month, NIS-LL Neuropathy Impairment Score-Lower Limb, TX treatment study; the plotted trend is the change from uncontrolled in the untreated synthetic cohort baseline in NIS-LL scores post-crossover. and progression slows within the tafamidis- Broadly, the trends demonstrate a slowing of treated synthetic cohort. disease progression in NIS-LL associated with Within Fig. 6, the observed values and cor- tafamidis. The average of these study-specific responding colors reported in Fig. 5 are retained trends is presented in Fig. 2. The average trend, (control = gray; tafamidis = black). These trends plotted in black, fits through the center of all are overlaid with the model-estimated trends, study-specific trends, with a shape consistent which are also color-coded (control = orange; with the Gompertz function suggested as tafamidis = blue). As seen in Fig. 6, the observed appropriate for the NIS-LL data in hATTR-PN synthetic cohort means (OBS) were precisely [26]. The same process was used to generate recovered by the MMRM-based values (LSMs). Figs. 3 and 4 for the Coelho et al. placebo arm Notably, the discrepancy in estimates was zero [23] and the simulated natural history data. The between baseline and month 6 in both treat- treatment and placebo synthetic cohort trends ment and placebo synthetic cohorts, and zero were plotted together in Fig. 5. The trends between month 18 and month 24 for the pla- overlap early, but as expected, diverge around cebo synthetic cohort. All other discrepancies month 12, as disease progression is were minor, and none evinced a departure from 148 Neurol Ther (2018) 7:141–154 Fig. 2 Study-stratified mean (95% confidence limits) NIS- baseline, M month, NIS-LL Neuropathy Impairment LL change from baseline trend for tafamidis treatment Score-Lower Limb, TX treatment cohorts, overlaying synthetic treatment cohort trend. BL the observed functional form. Thus, the dis- variance, as measured by the width of the 95% crete-time MMRM is expected to precisely confidence bands, shrank relative to any indi- recover the observed means in the forthcoming vidual study, but not excessively so. In addition, analyses. within the treatment synthetic cohort, the average trend and confidence bands mimicked a Gompertz function, which is a well-known DISCUSSION function for modeling decelerating exponential effects that asymptote asymmetrically. This is of In this work, a synthetic cohort approach was interest as the Gompertz function has been applied to the analysis of real-world outcomes proposed elsewhere for the analysis of neu- for tafamidis for the treatment of hATTR-PN, rodegenerative outcome measures within including comparison to natural history data. hATTR-PN [26]. Our findings demonstrate the merits of While the Gompertz function may be a good employing synthetic cohorts. The average trend approximation to the average trend, one might lines for the synthetic cohorts did not distort encounter difficulty in properly specifying the any of the study-specific trends. The error model in the context of repeated measures and Neurol Ther (2018) 7:141–154 149 Fig. 3 Study-stratified mean (95% confidence limits) NIS-LL change from baseline trend for natural history cohorts. BL baseline, CTRL control, M month, NIS-LL Neuropathy Impairment Score-Lower Limb random effects. In contrast, a discrete-time have artificially limited the variance and caused MMRM is easily parameterized and can flexibly the confidence limits to be underestimated. accommodate non-linear trends. Therefore, a Summary statistics available in the literature second part of this proof of concept was to were used, limiting the methods available for demonstrate that if individual-level data were optimally weighting the pooling procedure simulated from synthetic cohort means and averaging across studies. variances, a discrete-time MMRM could pre- In addition, three of the five studies consid- cisely recover the synthetic cohort-stratified ered were composed of samples that were 100% mean trends. In fact, the discrete-time MMRM Val30Met. The remaining two studies com- did succeed in recovering the observed means, prised mixed samples (\ 50% Val30Met). Pub- pointing to the ability of this model to detect lished evidence has suggested that progression and accurately reflect synthetic cohort treat- and treatment response outcomes differ sub- ment arm differences in NIS-LL disease stantially between Val30Met and non-Val30Met progression. populations [27–29]. However, the recent anal- This proof-of-concept report has some limi- ysis by Gundapaneni et al. [16] suggested that tations. The included studies may have had progression and treatment responses were no some overlap in the patient samples, which may different between Val30Met and non-Val30Met 150 Neurol Ther (2018) 7:141–154 Fig. 4 Study-stratified mean (95% confidence limits) NIS- CTRL control, M month, NIS-LL Neuropathy Impair- LL change from baseline trend for non-tafamidis cohorts, ment Score-Lower Limb overlaying synthetic control cohort trend. BL baseline, populations treated with tafamidis, after techniques (i.e., responder analyses). It is our adjusting for baseline neuropathy status. A contention that a precise answer is likely limitation of this proof-of-concept study is its achievable only by optimally pooling available inability to address the difference in progression data via IDA, and this issue speaks to the need to and treatment response between these impor- conduct this pooling research. tant sub-populations. However, given the The next step in this line of research is to modest sample sizes in the real-world data apply similar methods to the raw data corre- available to date, no single study has been able sponding to a larger set of real-world data to do this either. As a consequence, given the studies. Doing so will allow for an IDA approach findings of Gundapaneni et al., a new question [22, 30] in which the patient-level data from a that IDA may be uniquely positioned to answer group of independent studies is pooled, rather is whether mutation-dependent progression than the aggregate. With patient-level data, and treatment differences are important, or more sophisticated and sensitive methods of whether they are artifacts arising from modest pooling studies under optimal weighting para- sample sizes and potentially insensitive analysis digms can be employed. These include, but are Neurol Ther (2018) 7:141–154 151 Fig. 5 Synthetic cohort-stratified mean (95% confidence limits) NIS-LL change from baseline trend. BL baseline, CTRL control, M month, NIS-LL Neuropathy Impairment Score-Lower Limb, Tx treatment not limited to, incorporation of fixed and ran- CONCLUSION dom study effects, propensity-matching proce- dures, and the preferable hybrid of these Beyond the registration trial, evidence pub- approaches (i.e., doubly robust propensity lished to date on the natural history, disease weighting). With access to individual patient progression, and tafamidis treatment outcomes characteristics and outcomes, more sophisti- associated with hATTR-PN has demonstrated cated statistical techniques and models can also some heterogeneity and has been derived from be applied to achieve a greater understanding of studies with modest sample sizes (due to the low clinical outcomes by using a unified process prevalence of this disease). IDA and synthetic that adjusts for baseline and changes over time. cohorts are a technique that can be used to By better characterizing the natural history of analyze myriad studies with shared features to untreated hATTR-PN cases and the relative increase the precision of the characterization of benefit of tafamidis treatment, IDA would sig- hATTR-PN treatment outcomes. In so doing, nificantly facilitate clinician–patient commu- modest samples can be aggregated to form large nication regarding available treatment regimens cohorts from which increased precision of and their respective risks and benefits. inference may be obtained. As this is a proof of 152 Neurol Ther (2018) 7:141–154 Fig. 6 Treatment arm-stratified observed synthetic cohort MMRM mixed model for repeated measures, NIS-LL means versus MMRM estimated LSMs. BL baseline, Neuropathy Impairment Score-Lower Limb, OBS observed CTRL control, LSM least squares mean, M month, synthetic cohort, TX treatment concept for the application of IDA to patient- Inc. (New York, USA). All authors had full access level progression data in hATTR-PN, no definite to all of the data in this study and take complete conclusions about the effectiveness of tafamidis responsibility for the integrity of the data and can be made from these results. Rather, one can accuracy of the data analysis. only conclude from the evidence presented Medical Writing and/or Editorial Assis- herein whether IDA is a method that may be tance. Editorial assistance in the preparation of useful in the future for characterizing disease this article was provided by Catherine A. progression and drug effectiveness in a larger O’Connor and Caroline Seo of Pharmerit Inter- cohort using patient-level data. national. Support for this assistance was funded by Pfizer, Inc. ACKNOWLEDGEMENTS Authorship. All named authors meet the International Committee of Medical Journal Editors (ICMJE) criteria for authorship for this Funding. Sponsorship for this research and article, take responsibility for the integrity of article processing fees were funded by Pfizer, Neurol Ther (2018) 7:141–154 153 5. Coelho T, Merlini G, Bulawa CE, et al. Mechanism the work as a whole, and have given their of action and clinical application of Tafamidis in approval for this version to be published. hereditary transthyretin amyloidosis. Neurol Ther. 2016;5(1):1–25. https://doi.org/10.1007/s40120- Disclosures. This research was supported by 016-0040-x. Pfizer, Inc. Marc F. Botteman is a shareholder of 6. Bekircan-Kurt CE, Gunes N, Yilmaz A, Erdem-Oz- Pharmerit International. Daniel Serrano is an damar S, Tan E. Three Turkish families with differ- employee of Pharmerit International. Christo- ent transthyretin mutations. Neuromuscular pher B. Atzinger is an employee of Pharmerit Disord. 2015;25(9):686–92. https://doi.org/10. International. Marc F. Botteman is an employee 1016/j.nmd.2015.05.010. of Pharmerit International. Pharmerit received 7. Bulawa CE, Connelly S, Devit M, et al. Tafamidis, a funding from Pfizer, Inc., for study design, exe- potent and selective transthyretin kinetic stabilizer cution, analysis, and manuscript development. that inhibits the amyloid cascade. Proc Natl Acad Sci USA. 2012;109(24):9629–34. https://doi.org/10. 1073/pnas.1121005109. Compliance with Ethics Guidelines. This research is based on previously conducted 8. Reines JB, Vera TR, Martin MU, et al. Epidemiology studies and does not contain any studies with of transthyretin-associated familial amyloid human participants or animals performed by polyneuropathy in the Majorcan area: Son Llatzer Hospital descriptive study. Orphanet J Rare Dis. any of the authors. 2014;9:29. https://doi.org/10.1186/1750-1172-9- Data availability. All the data used in the analyses are presented in Tables 2 and 3. 9. Benson MD, Kincaid JC. The molecular biology and clinical features of amyloid neuropathy. Muscle Nerve. 2007;36(4):411–23. https://doi.org/10.1002/ Open Access. This article is distributed mus.20821. under the terms of the Creative Commons Attribution-NonCommercial 4.0 International 10. Coelho T, Maia LF, da Silva AM, et al. Long-term License (http://creativecommons.org/licenses/ effects of tafamidis for the treatment of transthyr- by-nc/4.0/), which permits any noncommer- etin familial amyloid polyneuropathy. J Neurol. 2013;260(11):2802–14. https://doi.org/10.1007/ cial use, distribution, and reproduction in any s00415-013-7051-7. medium, provided you give appropriate credit to the original author(s) and the source, provide 11. Keohane D, Schwartz J, Gundapaneni B, Stewart M, a link to the Creative Commons license, and Amass L. Tafamidis reduces disease progression in patients with transthyretin familial amyloid indicate if changes were made. polyneuropathy: supportive post hoc analyses of a pivotal trial. Orphanet J Rare Dis. 2015;10(Suppl 1):P11. https://doi.org/10.1186/1750-1172-10-s1- p11. REFERENCES 12. Barroso FA, Judge DP, Ebede B, et al. Long-term safety and efficacy of tafamidis for the treatment of 1. Benson MD. The hereditary amyloidoses. Best Pract hereditary transthyretin amyloid polyneuropathy: Res Clin Rheumatol. 2003;17(6):909–27. https:// results up to 6 years. Amyloid. 2017;24(3):194–204. doi.org/10.1016/j.berh.2003.09.001. https://doi.org/10.1080/13506129.2017.1357545. 2. Hund E, Linke RP, Willig F, Grau A. Transthyretin- 13. Keohane D, Schwartz J, Gundapaneni B, Stewart M, associated neuropathic amyloidosis.Pathogenesis Amass L. Tafamidis delays disease progression in and treatment. Neurology. 2001;56(4):431–5. patients with early stage transthyretin familial amyloid polyneuropathy: additional supportive 3. Schmidt HH, Waddington-Cruz M, Botteman MF, analyses from the pivotal trial. Amyloid. et al. Estimating the global prevalence of trans- 2017;24(1):30–6. https://doi.org/10.1080/ thyretin familial amyloid polyneuropathy. Muscle 13506129.2017.1301419. Nerve. 2017. https://doi.org/10.1002/mus.26034. 14. Merlini G, Plante-Bordeneuve V, Judge DP, et al. 4. Connors LH, Richardson AM, Theberge R, Costello Effects of tafamidis on transthyretin stabilization CE. Tabulation of transthyretin (TTR) variants as of and clinical outcomes in patients with non-Val30- 1/1/2000. Amyloid. 2000;7(1):54–69. Met transthyretin amyloidosis. J Cardiovasc Transl 154 Neurol Ther (2018) 7:141–154 Res. 2013;6(6):1011–20. https://doi.org/10.1007/ 22. Curran PJ, Hussong AM. Integrative data analysis: s12265-013-9512-x. the simultaneous analysis of multiple data sets. Psychol Methods. 2009;14(2):81. 15. Waddington Cruz M, Amass L, Keohane D, et al. Early intervention with tafamidis provides long- 23. Coelho T, Maia LF, Martins da Silva A, et al. Tafa- term (5.5-year) delay of neurologic progression in midis for transthyretin familial amyloid polyneu- transthyretin hereditary amyloid polyneuropathy. ropathy: a randomized, controlled trial. Neurology. Amyloid. 2016;23(3):178–83. https://doi.org/10. 2012;79(8):785–92. https://doi.org/10.1212/WNL. 1080/13506129.2016.1207163. 0b013e3182661eb1. 16. Gundapaneni BK, Sultan MB, Keohane DJ, Schwartz 24. Wickham H. ggplot2: Elegant graphics for data JH. Tafamidis delays neurological progression analysis. New York: Springer; 2009. comparably across Val30Met and non-Val30Met genotypes in transthyretin familial amyloid 25. SAS version 9.4. SAS Institute: Cary, NC, USA; 2013. polyneuropathy. Eur J Neurol. 2017. https://doi. org/10.1111/ene.13510. 26. Adams D, Coelho T, Obici L, et al. Rapid progres- sion of familial amyloidotic polyneuropathy: a 17. Ando Y, Coelho T, Berk JL, et al. Guideline of multinational natural history study. Neurology. transthyretin-related hereditary amyloidosis for 2015;85(8):675–82. https://doi.org/10.1212/WNL. clinicians. Orphanet J Rare Dis. 2013;8:31. https:// 0000000000001870. doi.org/10.1186/1750-1172-8-31. 27. Ericzon BG, Wilczek HE, Larsson M, et al. Liver transplantation for hereditary transthyretin amy- 18. VYNDAQEL [summary of opinion—initial authori- loidosis: after 20 years still the best therapeutic sation] [online]. London: European Medicines alternative? Transplantation. 2015;99(9):1847–54. Agency 2011-21-July Contract No.: EMA/CHMP/ https://doi.org/10.1097/TP.0000000000000574. 546201/2011. 28. Kerschen P, Plante-Bordeneuve V. Current and 19. Cortese A, Vita G, Luigetti M, et al. Monitoring effectiveness and safety of Tafamidis in transthyr- future treatment approaches in transthyretin etin amyloidosis in Italy: a longitudinal multicenter familial amyloid polyneuropathy. Curr Treat study in a non-endemic area. J Neurol. Options Neurol. 2016;18(12):53. https://doi.org/10. 2016;263(5):916–24. https://doi.org/10.1007/ 1007/s11940-016-0436-z. s00415-016-8064-9. 29. Suhr OB, Larsson M, Ericzon BG, Wilczek HE, 20. Lozeron P, Theaudin M, Mincheva Z, et al. Effect on FAPWTR’s investigators. Survival after transplanta- disability and safety of Tafamidis in late onset of tion in patients with mutations other than Val30- Met30 transthyretin familial amyloid polyneu- Met: extracts from the FAP world transplant ropathy. Eur J Neurol. 2013;20(12):1539–45. registry. Transplantation. 2016;100(2):373–81. https://doi.org/10.1111/ene.12225. https://doi.org/10.1097/TP.0000000000001021. 21. Plante-Bordeneuve V, Gorram F, Salhi H, et al. 30. Hofer SM, Piccinin AM. Integrative data analysis Long-term treatment of transthyretin familial through coordination of measurement and analysis amyloid polyneuropathy with tafamidis: a clinical protocol across independent longitudinal studies. and neurophysiological study. J Neurol. Psychol Methods. 2009;14(2):150. 2017;264(2):268–76. https://doi.org/10.1007/ s00415-016-8337-3.

Journal

Neurology and TherapySpringer Journals

Published: Apr 2, 2018

References

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off