Re: Ganz ML, Chavan A, Dhanda R, et al. Cost-effectiveness of valbenazine compared with deutetrabenazine for the treatment of tardive dyskinesia. J Med Econ. 2021;24(1):103–113

This article refers to:

Cost-effectiveness of valbenazine compared with deutetrabenazine for the treatment of tardive dyskinesia

This article is related to:

In Reply to Xue W, Ribalov R, Zhou Z-Y, et al. Re: Ganz ML, Chavan A, Dhanda R, et al. Cost-effectiveness of valbenazine compared with deutetrabenazine for the treatment of tardive dyskinesia. J Med Econ. 2021;24(1):103–113

Dear Editor,

We read with great interest the study by Ganz et al.¹, a cost-effectiveness analysis (CEA) of valbenazine versus deutetrabenazine for the treatment of tardive dyskinesia (TD) from a third-party payer perspective in the United States (US). The study reported that valbenazine was associated with a higher likelihood of treatment response at one year and more quality-adjusted life years (QALYs) compared with deutetrabenazine. Ganz et al. used two outcome measures to assess response to treatment: the Abnormal Involuntary Movement Scale (AIMS) total score and the Clinical Global Impression of Change (CGIC) score. Using AIMS total score, the estimated incremental cost-effectiveness ratio (ICER) was $9,951/QALY for valbenazine compared with deutetrabenazine. When using CGIC score, valbenazine was dominant with 3.4 QALYs and a lifetime cost of $252,311, compared with 3.3 QALYs and a lifetime cost of $283,208 for deutetrabenazine.

As valbenazine and deutetrabenazine are the standard of care for TD in the US, the results of a rigorously conducted CEA of these therapies are of high interest to payers, patients, and healthcare providers affected by the substantial economic burden of TD. However, some important limitations of the model by Ganz et al. resulted in the overestimation of the cost-effectiveness of valbenazine versus deutetrabenazine, impacting its validity.

First, the model inappropriately weighted the deutetrabenazine doses in terms of both efficacy and medication costs. As there are no clinical trials directly comparing valbenazine and deutetrabenazine, Ganz et al.’s CEA based the probability of treatment response on an indirect treatment comparison (ITC) by Aggarwal et al.² that used pooled data from the therapies’ randomized, placebo-controlled trials: the KINECT 2/KINECT 3^3,4 trials of valbenazine and the ARM-TD/AIM-TD^5,6 trials of deutetrabenazine. However, while deutetrabenazine 12 mg/d (a less efficacious dose versus 24 or 36 mg/d in AIM-TD) was included in the model’s efficacy estimates, it was not included in the medication cost estimates. Instead, the medication cost of deutetrabenazine was weighted by the doses of 24 mg/d (27%), 36 mg/d (27%), and 48 mg/d (46%). The more effective, and more expensive, deutetrabenazine dose (48 mg/d) accounted for less than 23% of patients when estimating effectiveness, but 46% of patients when calculating medication costs. Therefore, the inconsistent application of weighting for deutetrabenazine doses led to underestimation of its effectiveness, while failing to adjust the medication costs.

Second, the ITC used as the basis for probability of treatment response may itself be subject to bias due to important differences in study design and patient characteristics between the included trials. The proportions of female and white patients were lower in KINECT 2/KINECT 3 (females: 43.0% and 45.8%, respectively; white: 63.0% and 56.4%)^3,4 compared to ARM-TD/AIM-TD (females: 52.1% and 55.0%, respectively; white: 69.2% and 79.0%)^5,6. Additionally, the mean AIMS score at baseline was lower in KINECT 2 compared with other trials^3–6. Methods such as matching adjusted indirect comparison or simulated treatment comparisons could have addressed some of the population differences, but were not utilized in the ITC. Furthermore, substantial differences in the reduction of the AIMS score from baseline (least squares mean change) were observed between the placebo groups in KINECT 2 and KINECT 3 (−0.2 and −0.1, respectively, at 6 weeks)^3,4 versus ARM-TD and AIM-TD (−1.6 and −1.4, respectively, at 12 weeks)^5,6, suggesting underlying differences in the recruited populations. The ITC reported odds ratios (ORs) for the AIMS response (≥50% total score improvement from baseline) associated with valbenazine and deutetrabenazine (in Table 3 of that publication)², but none of the individual or pooled comparisons resulted in any significant differences. The large confidence interval of the OR was not fully tested in the sensitivity analysis of the CEA.

Third, the assumption of sustained treatment effect based on the ITC using short-term trial data is not supported by clinical evidence and it underestimates the treatment response to deutetrabenazine. The model estimated response to deutetrabenazine by applying the OR from the ITC to the efficacy of valbenazine at 8, 16, 24, and 48 weeks. However, the ITC was based on only 6 weeks of treatment with valbenazine and 12 weeks of treatment with deutetrabenazine². The AIMS-based response for deutetrabenazine estimated using the above method (i.e. 23% at Week 16 and 31% at Week 48) is lower than that observed in the deutetrabenazine open-label extension study (38% at Week 15 and 48% at Week 54)⁷. The model further assumed responders at Week 48 remained on treatment until death. These assumptions are not supported by clinical evidence and require further validation.

Finally, the results of Ganz et al.’s model are counterintuitive and the assessments regarding parameter and structural uncertainties are limited. Varying the time horizon by ±20% in the deterministic sensitivity analysis appears counterintuitive because the base case time horizon was over a lifetime. In addition, the cost-effectiveness acceptability curve seems implausible because the probabilities of valbenazine and deutetrabenazine being cost-effective at a given willingness-to-pay threshold do not always sum to 100%. On the other hand, the validity of key structural assumptions, such as the use of constant ORs across all response assessment time points, was not assessed in a sensitivity analysis. The assumption that responders remain on treatment after 48 weeks is not justified in the literature yet the uncertainty was not evaluated. Similarly, the uncertainty of varying the values of utility inputs in the model was not evaluated. For the base case analyses, the incremental life years (LYs) were 0.027 between valbenazine and deutetrabenazine while the incremental QALYs were 0.118. Considering the low magnitude of life extension compared with the difference in QALY, the utility inputs are expected to be important drivers, and a small change in the input value could substantially alter the base-case ICER.

Moreover, the authors may have overlooked some potential errors. There are substantial differences between the LY gains estimated in this publication compared to a previous conference poster presented by the same authors (approximately 4 years versus 11 years for valbenazine, respectively)⁸, despite using similar demographic inputs and time horizon.

In summary, the CEA published by Ganz et al. may be subject to bias due to substantial and unaddressed limitations in the inputs and assumptions of the model, as well as inadequate sensitivity analyses and limited assessment of alternative parameters. Because these factors could have greatly impacted the results, we urge caution in their interpretation to avoid erroneous conclusions.

Transparency

Declaration of funding

This work was supported by Teva Pharmaceuticals.

Declaration of financial/other relationships

SL and RR are employees of Teva Pharmaceuticals and hold stock/options.

WX, ZZ, and RA are employees of Analysis Group, Inc., which has received consulting fees from Teva Pharmaceuticals.

Acknowledgements

Medical writing assistance was provided by Shelley Batts, an employee of Analysis Group, Inc. Support for this assistance was provided by Teva Pharmaceuticals.