A probabilistic model was developed to evaluate the 5-year HE impact (cost-effectiveness and budget impact) of certolizumab pegol (CZP) with and without an RSS at treatment initiation as an alternative biologic option for biologic-naïve patients with moderate-to-severe RA (moderate RA, 3.2 < DAS28 ≤ 5.1; severe RA, DAS28 > 5.1) compared with the treatment mix of reimbursed biologic disease-modifying antirheumatic drugs (bDMARDs; abatacept, adalimumab, CZP, etanercept, golimumab, infliximab, tocilizumab) currently used in Finland. The use of a treatment mix as a base case comparator for the CZP RSS was based on the following rationale: (1) a CZP RSS would be more likely to complement a mix of treatments over time, rather than a single treatment; (2) inclusion of both incident and prevalent RA patients necessitates a mix; (3) budget impact estimates are more relevant for a mix, and (4) decision makers often consider the comparator to be “current care” and not a single treatment. However, sensitivity analyses report results of CZP RSS versus single treatment scenarios.
RSSs or patient access schemes for publicly reimbursed pharmaceuticals were not part of Finnish reimbursement practice at the time of analysis [13–15]. However, since January 1, 2017, the Finnish Pharmaceuticals Pricing Board has considered RSSs proposed as part of new reimbursement applications on a drug-by-drug basis. The general framework is an agreement-based conditional reimbursement that needs to be separately applied [16, 17] and that can include an RSS. This study models a potential—yet hypothetical—RSS for the Finnish situation. We are not aware of any other formal analyses that explore the option to include an RSS as part of a Finnish reimbursement agreement.
Estimated outcomes included per-patient incremental cost-effectiveness and net budget impact at the target population level. Secondary estimated outcomes included per-patient survival (life years), quality-adjusted survival (quality-adjusted life years, QALYs), and lost productivity (absenteeism in terms of work days lost, valued by the human-capital approach, HCA ) over 5 years. Treatment costs were estimated at the patient and target population levels. Although lifetime modeling is often considered in modeled HE evaluations, 5-year modeling was used in this study on the basis of clinical and economic rationales, including changes in treatment recommendations, improved care practices, market fluctuations (changes in the market shares of drugs, biosimilars, and new treatments), and potential advances in our understanding of RA within 5–10 years. In addition, a 5-year timeframe is associated with significantly less extrapolation uncertainty than a lifetime horizon, and follow-up data related to disease progression in Finnish RA patients treated with bDMARDs for more than 5 years are lacking.
A cost-effectiveness acceptability frontier was drawn to determine optimal treatments, in terms of net monetary benefit, at different willingness-to-pay levels. The commonly referenced willingness-to-pay values include €30,000 and €50,000 per QALY gained in Finland [19–24]. However, Finland does not have an official threshold for cost-effectiveness [15, 25]. Threshold estimates for the UK could potentially be applicable in Finland: most plausible threshold, around €25,000 per QALY gained, plausible threshold, around €37,000 per QALY gained , or end-of-life threshold, around €55,000  per QALY gained, on the basis of population-weighted decisions. The Finnish Medicines Agency considered €68,000 per QALY gained to approach the maximum cost-effectiveness threshold for a life-threatening disease , which concurs with earlier Finnish average opinion . Because RA is not a life-threatening disease or end-of-life condition, incremental cost-effectiveness ratios exceeding €37,000 per QALY gained are probably not acceptable for RA in Finland, or at least necessitate additional evidence.
Clinical response, assessed by American College of Rheumatology (ACR) response criteria, and population-level budget impact were estimated without discounting , whereas a 3% annual rate  was applied to the remaining outcome measures. ACR response was selected as the primary surrogate outcome because it is the most widely reported outcome in RA trials (see Appendix A in the electronic supplementary material). ACR response was also included in the Finnish register of biologic treatments and the FIN-RACo trial.
CZP is one of the most recent tumor necrosis factor inhibitors to be licensed for the treatment of RA. Studies have shown that an early response to CZP treatment at week 12 can predict the likelihood of long-term response [32–34] (see Appendix A in the electronic supplementary material). This suggests that an RSS based on a 12-week assessment of efficacy would be feasible for CZP.
The impact of a hypothetical RSS based on the achievement of ACR20 response at week 12 was estimated. Under the RSS, biologic-naïve patients achieving ACR20 response would continue with CZP treatment, whereas failure to respond would lead to a treatment switch and subsequent refund of the costs associated with CZP acquisition. This scenario is in line with the current European League Against Rheumatism (EULAR)  and Finnish  treatment guidelines, which emphasize a treat-to-target approach in RA and the consideration of alternative treatment in the instance of early nonresponse.
The Summary of Product Characteristics (SPCs) for some first-line RA bDMARDs (e.g., abatacept, golimumab, and infliximab) do not encourage a change in treatment if response is not achieved by week 12, but rather advocate continuation of treatment, with assessment of clinical response at later time points (e.g., 24 weeks). Therefore, the RSS was applied only to CZP-treated patients. In the model base case, the continuation or switching decision for the current treatment mix was based on attainment of ACR20/50/70 response at week 24 (at minimum, ACR20 response was required for continuation).
Health Economic Modeling
The treatment of moderate-to-severe RA after inadequate response to a conventional disease-modifying antirheumatic drug (cDMARD) was modeled. This model was structured as a fully probabilistic (including standard errors/deviations where available), open-cohort model with a 12-week cycle length, implemented in Microsoft Excel with Visual Basic for Applications (Fig. 1). Normal distributions were used to propagate the uncertainty of the following inputs: initial Health Assessment Questionnaire Disability Index (HAQ-DI) score; prediction of HAQ-DI score based on ACR response; multivariate prediction of quality of life (QoL) based on HAQ-DI score, or HAQ-DI bands-based prediction of QoL (sensitivity analysis); background mortality based on age and sex; and elevated mortality, hospitalizations, and work days lost based on HAQ-DI score. For ACR responses, logistic distributions were implemented.
CZP as the first-line bDMARD for all biologic-naïve patients (2015 onwards) was compared with a current mix of bDMARDs, with and without an RSS. The current mix of bDMARDs was assumed to consist of subcutaneously administered, reimbursed bDMARDs, and was based on the market shares of available drugs (subcutaneously administered abatacept 5.1%, adalimumab 46.9%, CZP 3.7%, etanercept 31.2%, golimumab 13.1%; 2013 IMS Health data ), SPCs, and Finnish clinical practice.
Cost-effectiveness analysis was performed for an “average patient” from the first bDMARD for moderate-to-severe RA over a period of 5 years. For the current treatment mix, the model simulated patients starting with each of the different bDMARDs, and then averaged these results on the basis of drug market share.
On the basis of reimbursement statistics from the Finnish Social Insurance Institution and sales data from IMS Health, we estimated that 799 RA patients received new first-line bDMARDs in 2015 (biologic-naïve patients, the initial target population for budget impact analysis), a number that is estimated to increase by 3% per annum until 2019 on the basis of historical trends. Patients were assumed to use the first-line bDMARD in combination with methotrexate (90% of bDMARD-treated Finnish patients with RA had received cDMARD treatment at bDMARD treatment initiation) .
In the compared settings, patients continued with bDMARD and methotrexate treatment beyond week 24 if the minimum response criterion of ACR20 was met and no adverse event precluding treatment continuation occurred (incidence of 1.9% during 24 weeks ). The maximum duration of first-line bDMARD treatment was set to 144 weeks. Patients discontinuing first-line bDMARD treatment were assumed to switch to rituximab, which has been shown to be a cost-effective subsequent treatment in Finnish settings [39, 40]. Patients exited the model through death (constant mortality rate adjusted for disease severity based on HAQ-DI score) or once they reached the model maximum of 260 weeks.
Efficacy data for bDMARDs were estimated by a meta-analysis, applying a random effects model of efficacy responses from clinical trials, accounting for the correlations between outcomes, and separating outcomes reported at week 12 or 24 (see Appendix A in the electronic supplementary material). In the absence of long-term randomized data on treatment effects, and because of the intention-to-treat setting of clinical trials, the first-line treatment response (without adverse events) was assumed to persist until death or the modeled maximum timeframe of 144 weeks with the first-line treatment [12, 41].
ACR20 and ACR50 treatment outcomes for second-line rituximab treatment (infliximab treatment followed by rituximab treatment in a sensitivity analysis scenario) after initial bDMARD treatment were modeled on patient response and persistence data from the South Swedish Arthritis Treatment Group Register and the Spanish BIOBADASER database [41–43].
Within the analysis timeframe, patients had a constant mortality rate (annual rate of 3.05 per 1000 patient-years at age 52 years for 2013 in Finland [44, 45]). This was adjusted upward on the basis of the HAQ-DI status of the patient to compensate for the elevated risk of death associated with RA, given as an odds ratio of 2.93 (95% confidence interval 2.43–3.54) per unit increase in HAQ-DI score (based on this being the estimate with the best Bayesian information criterion and Z score among RA factors ). The impact of excluding elevated mortality due to RA was tested in sensitivity analyses.
At the patient level, the HAQ-DI score takes values in multiples of 0.125. However, the representative cohort’s initial mean HAQ-DI score was assumed to be 1.2 (standard deviation 0.7), in line with the values used in an earlier Finnish RA assessment of first-line bDMARDs (mean HAQ-DI score 1.2 ) among Finnish RA register bDMARD users (mean HAQ-DI score 1.1 ) and from mortality information (mean HAQ-DI score 1.2, standard deviation 0.76) . HAQ-DI score was predicted to change in relation to ACR response level, as reported in an earlier Finnish HE analysis . HAQ-DI score was not assumed to increase as a result of RA. For cost–utility outcomes, QoL was modeled with use of the published linear relationship between HAQ-DI and the Health Utility Index . The impact of QoL values was tested in a sensitivity analysis using HAQ-DI bands.
A societal 5-year perspective was used in the analyses, including direct medical and traveling costs, and HCA-based productivity losses. Payer perspective results were also calculated (i.e., excluding the productivity losses). Input of drug administration was based on SPC guidelines. The cheapest reimbursed retail costs for drugs were sourced from the Finnish Medicines Tariff, June 2015 (including biosimilar infliximab, and subcutaneously administered abatacept and tocilizumab; Table 1).
The incidence of hospitalization was modeled according to HAQ-DI scores , and Finnish productivity losses were included on the basis of the reported association between ACR responses and work days lost . The HCA was used to analyze the potential productivity losses, since ACR responses and HAQ-DI scores are important for the analysis of productivity losses in RA, and HCA-based productivity losses are associated with both HAQ-DI score  and ACR response .
Initiation of bDMARD treatment consisted of a nurse visit (for subcutaneously administered treatments only), an antibody test, an QuantiFERON test, and chest X-ray. Resource use for the 12-week treatment cycles included 0.5 GP visits, 0.5 outpatient visits, 1.5 laboratory visits and phone consultations [1.5 liver value tests (alanine aminotransferase), 1.5 blood counts, and 0.5 creatine tests], and related traveling. Intravenous administration costs were based on a Finnish study . Unit costs are listed in Table 1. The potential extra specialist visit needed to assess the RSS response criterion at week 12 was also included.
The sensitivity of the modeling assumptions was assessed in the following settings: no RSS for CZP, drug-to-drug comparisons for commonly used bDMARDs (100% adalimumab, etanercept, golimumab, or biosimilar infliximab treatment mix assumed); three bDMARDs modeled (infliximab included in the treatment sequence between the current mix or CZP and rituximab); inclusion of ACR70 response for the subsequent treatment; first-line treatment duration of 72 weeks; first-line treatment duration of 240 weeks; additional mortality due to RA ignored; QoL based on HAQ-DI bands ; no discounting; ACR response assessment at week 12, and ACR response assessment at week 24.
Compliance with Ethics Guidelines
This article does not contain any new studies with human or animal subjects performed by any of the authors.