Modeling of the clinical and economic impact of a risk-sharing agreement supporting a treat-to-target strategy in the management of patients with rheumatoid arthritis in France

Abstract

Objectives: To evaluate the cost-effectiveness of a Treat-to-Target strategy with certolizumab pegol in patients with rheumatoid arthritis in the context of a pay-for-performance agreement in which medication costs are refunded in case of discontinuation during the first 3 months of treatment.

Methods: The Treat-to-Target strategy consisted of a systematic switch to second-line tumor necrosis factor (TNF)α inhibitor in case of an unmet ACR50 response at 3 months compared to current routine clinical practice. A reference cohort treated first-line with certolizumab pegol according to current practice without systematic switching was considered as the comparator. A decision-tree model was constructed to estimate clinical outcome (health assessment questionnaire-disability index or HAQ-DI score), time spent in ACR50 response (ACR 50), and direct costs of treatment over a 2-year period. HAQ scores were derived from American College of Rheumatology 50 (ACR50) responses. All TNFα inhibitors were assumed to have equivalent efficacy and tolerability. Costs were estimated at 2013 French retail prices (date of the pay-for-performance agreement).

Results: The mean duration of an ACR50 response was 1.23 years in the Treat-to-Target strategy certolizumab pegol cohort vs 0.98 years in the reference cohort, resulting in a mean gain in HAQ at 24 months of 0.117. The Treat-to-Target strategy with a mix of TNFα inhibitors as second-line therapy was more expensive than the reference strategy in absolute terms, but this difference was entirely offset by the pay-for-performance agreement. The Treat-to-Target strategy was, thus, cost-neutral over a 2-year period after the payback of CZP cost for patients not achieving the target at 3 months.

Conclusions: In the context of a pay-for-performance agreement, the management of patients with rheumatoid arthritis using a Treat-to-Target strategy with certolizumab pegol in first line is dominant compared to standard use of this drug in the French setting in 2013.

Keywords:

• Rheumatoid arthritis
• Certolizumab pegol
• Treat-to-Target strategy
• Cost-effectiveness
• Pay-for-performance
• Risk-sharing
• ACR50

Introduction

International recommendations^1,2 on the management of rheumatoid arthritis (RA) emphasize the importance of obtaining either clinical remission or the lowest level of disease activity as quickly as possible in order to optimize long-term response. This is an important issue since response rates to first-line therapy with methotrexate (MTX) and first- or second-line therapy with biological agents are generally <50% using the American College of Rheumatology 50 (ACR50) or Disease Activity Score 28 (DAS28) criteria^3–6 and even lower when considering the treatment goal of remission^5,7. The 2013 European League Against Rheumatism (EULAR) recommendations² have specified that significant clinical improvement should be measurable at 3 months after treatment initiation. The rationale for choosing an early response assessment criterion is that the likelihood of treatment failures at 3 months subsequently achieving a clinical response is low and, thus, the 3-month non-response rate can be taken to be a surrogate marker of long-term outcome⁸. For this reason, it is recommended to evaluate treatment response at 3 months and change the treatment strategy systematically in the case of inadequate response at this time. The benefits associated with such a “Tight Control” or “Treat-to-Target” approach have been demonstrated clearly in a number of studies over the last decade^9–17. However, in everyday care, the proportion of patients changing treatments in case of inadequate response is low, suggesting that switching is still sub-optimal⁶.

Certolizumab pegol (CZP) is a PEGylated recombinant TNFα inhibitor effective in the treatment of RA¹⁸. In association with MTX, it is indicated “for the treatment of active, moderate to severe, RA in adults, when the response to maintenance treatments (disease-modifying anti-rheumatic drugs; DMARDs), including MTX, is inadequate”¹⁹. CZP was accepted for reimbursement in France by the French Health Authorities (Commission de la Transparence opinion dated March 2010²⁰) and has been marketed since September 2010. A risk-sharing agreement has been negotiated with Health Authorities in 2013, based on a pay-for-performance basis. Such agreements allow risks and costs to be shared between public health insurance and the manufacturer^21–23. In the case of CZP, the manufacturer agreed to refund the medication cost for all patients who discontinued treatment, for whatever reason, during the 3 months following initiation of CZP (as these patients were considered “early failures”). The criterion applied in this risk-sharing agreement thus matches the notion of Treat-to-Target as set out in the EULAR guidelines, in which treatment failures can be identified as early as 3 months and patients switched to another treatment.

In the absence of real-world data on the clinical and economic consequences of implementation of a Treat-to-Target strategy in the framework of the pay-for-performance agreement, we have undertaken a modeling study to address these issues. The objective was to evaluate the clinical and economic consequences of using a Treat-to-Target strategy, consisting of first-line CZP, followed by systematic switch to another biological agent in case of inadequate clinical response to CZP at 3 months, compared to a conventional treatment strategy currently used in standard care (reference strategy) in the French healthcare system, which does not include a systematic switch of therapy at 3 months.

Methods

Modeling approach

A model was developed to estimate the clinical and economic consequences of two different therapeutic strategies in hypothetical cohorts of RA patients treated with first-line CZP. The reference strategy (CZP-Ref) simulated current practice observed in the principal French health insurance claims database for persistence with first-line subcutaneous TNFα inhibitors over the first year after initiation; in this scenario, switching rates are much lower than would be expected if the EULAR guidelines were being followed. In the Treat-to-Target strategy (CZP-TTT), a hypothetical cohort of RA patients treated with first-line CZP was simulated with patients implemented consistently with the EULAR guidelines. These strategies are described in more detail in the next section.

Outcomes evaluated were clinical benefit and costs, which were modeled for cohorts of 1000 patients over a 2-year period. Mortality was not taken into consideration in the analysis because it was considered to be negligible in view of the short duration of follow-up (2 years). It was assumed that the clinical efficacy of all four TNFα inhibitors was identical and, thus, that the effectiveness estimated in the model would be independent of the individual TNFα inhibitor used.

The hypothesis tested in the modeling study was that a first-line CZP-TTT strategy would be cost-effective compared to conventional use of CZP without systematic switching (CZP-Ref).

Treatment strategies compared

The reference strategy (CZP-Ref) consisted of first-line therapy with CZP using treatment discontinuation rates (for all reasons, including inadequate efficacy, intolerance or patient request) based on an analysis of the French national medical insurance claims databases²⁴. This analysis suggested a constant monthly discontinuation rate during the first year of follow-up, yielding a rate of 78.1% (67.4–85.6%) at 1 year of patients still in first-line. These results are consistent with those described in a number of European patient registries^25–29, in which switch rates of ∼20% after 1 year and 30% after 2 years have been reported. In our model, these discontinuation rates were applied, irrespective of the clinical response. This reflects current routine practice, where the decision to stop treatment and change line is not necessarily made as a function of failure to obtain a given level of clinical response such as the ACR50 response. A consequence of this approach is that a proportion of patients on first- or second-line treatment are maintained on treatment, even in the absence of a marked clinical response.

In the investigational strategy, the CZP-TTT cohort was treated with CZP at first-line and re-evaluated after 3 months. The 3-month evaluation point was chosen to match the period for which prescriptions in France are routinely filled and the point at which patients would attend a follow-up consultation with their rheumatologist to renew their prescription. In addition, it represents the time-point at which clinical improvement should be manifest in the EULAR treatment guidelines². Those patients who failed to meet the ACR50 response criteria were systematically switched to another TNFα inhibitor, and then followed an identical therapeutic strategy to the patients in the reference strategy. In contrast, patients who were ACR50 responders at 3 months continued CZP treatment until the end of the period.

Maintenance of a second-line TNFα inhibitor was considered to conform to routine practice in both cohorts. For simplicity, the second-line TNFα inhibitor therapy was limited to one of the three subcutaneous TNFα inhibitors most commonly used as in France, namely etanercept, adalimumab, and golimumab³⁰. Infliximab, an intravenous TNFα inhibitor also widely prescribed in France, was not included in the model since its mode of administration (6–8-weekly infusions under medical supervision) means that decisions related to treatment switching, as well as adherence issues, are likely to be quite different to those encountered for subcutaneous (SC) TNFα inhibitors administered by the patient at home. In addition, infliximab tends to be prescribed to more difficult to treat patients with more severe disease or multiple co-morbidities.

The specific treatment sequences studied in the model, and which conform to practice guidelines^31–33, were the following: (i) prescription of CZP as first-line therapy; and (ii) switch to another SC TNFα inhibitor as second-line. For simplicity, a mix of etanercept, adalimumab, or golimumab distributed according to their respective utilization in France was used³⁴. The percentages of patients treated with these three TNFα inhibitors in 2013 were 55% (etanercept), 36.4% (adalimumab), and 8.6% (golimumab)³⁴. This method is justified by our hypothesis of equivalence in efficacy and tolerability of all three TNFα inhibitors considered; (iii) after the use of a second-line TNFα inhibitor, switch to another biological agent (according to current treatment practices, rituximab as third-line, then tocilizumab as fourth-line). In the model used, a patient stopping a biological treatment is immediately treated with another biological agent according to this sequence.

Efficacy of treatments

Clinical source data used to establish the model are presented in Table 1. The model is based on the hypothesis that all TNFα inhibitors considered possess comparable efficacy and tolerance as first-line therapy^2,35–38 and that all TNFα inhibitors possess comparable adherence rates^39,40.

Table 1. Clinical input variables for the economic model.

Clinical input variable		Source
Proportion of patients achieving ACR50 response thresholds		Launois et al.^41
TNFα inhibitors	45%
Rituximab/tocilizumab*	28%
Change in HAQ-DI, mean (SE) as a function of ACR response		Soini et al.^49
No response	−0.136 (0.017)
ACR20	−0.443 (0.0183)
ACR50	−0.668 (0.0263)
ACR70	−0.923 (0.0323)

* In patients failing TNFα inhibitor therapy.

Derived from a recent meta-analysis⁴¹, the percentages of patients achieving an ACR20, ACR50, and ACR70 response were ∼65%, 45%, and 20%, whatever the TNFα inhibitor considered. This hypothesis is also consistent with the guidance of the Commission de la Transparence in France on the different TNFα inhibitors available⁴². On this basis, we assumed that a proportion of 55% of patients receiving CZP as first-line therapy do not achieve an ACR50 response at 3 months (defined as insufficient response in the current analysis) and these patients are, therefore, switched to a second-line TNFα inhibitor in the TTT strategy, as recommended by the French HTA practice guidelines⁴³. In everyday care, systematic switching does not occur and the proportions of patients stopping a first-line anti-TNF are 20% at 1 year and 30% at 2 years^25,26. In the TTT strategy, it was assumed that 5% of responders who continued their first-line treatment beyond 3 months would discontinue treatment for whatever reason before 2 years.

The model also considers that the efficacy of a second-line TNFα inhibitor is reduced compared to its efficacy as first-line therapy⁴⁴, in agreement with data from patient registries⁴⁵. Only a few randomized studies currently exist which demonstrate the efficacy of a TNFα inhibitor as second- or third-line treatment⁴⁶. In view of these uncertainties, the reduction in efficacy of second-line therapy was the subject of a sensitivity analysis, with values varying between 0–15% (baseline value = 0%, no reduction). This adjustment was applied to the percentages of patients achieving the different ACR responses at 3 months with second-line treatment and is assumed to be identical, irrespective of the drug under consideration.

The biological agents used in cases of failure of two sequential TNFα inhibitors were rituximab and tocilizumab, which were also assumed to have equivalent efficacy, with an ACR50 response at 1 year of 28%, as reported in the guidance for use of these agents issued by the French health authorities (Commission de la Transparence)^19,47.

Estimation of effectiveness

Evaluation of the clinical benefit of the two different treatment strategies studied was based on the estimated change in functional health assessment questionnaire-disability index (HAQ-DI) score over the 2-year follow-up in each cohort. The improvement in this score was extrapolated from estimations of the change in HAQ-DI score observed in patients achieving ACR20, ACR50, and ACR70 response in a Phase III study of tocilizumab in RA⁴⁸ using the values published by Soini et al.⁴⁹ (Table 1). These values were used for each of the treatments considered, irrespective of line.

Estimation of costs

Economic source data used to establish the model are presented in Table 2. All costs were estimated from a public insurance third-party payer perspective. Economic criteria were limited to the total direct costs over the 2-year period, including the purchase of biological treatments as well as the costs of follow-up (except for other pharmacy costs). Indirect costs were not included in this analysis. Patients with RA in France are fully reimbursed for their medical costs. The economic perspective retained in the model is, therefore, a combination of retail costs and a societal perspective limited to direct medical costs.

Table 2. Unit costs used in the economic model.

Unit cost input			Source
Annual medication acquisition costs (retail prices per injection 2013)†			Sécurité Sociale: l’Assurance Maladie^64
Certolizumab pegol (200 mg)	Unit cost: €434	€12586/year (Year 1)
		€11284/year (Year 2)
Etanercept (50 mg)¶	Unit cost: €247	€12860/year
Adalimumab (40 mg)	Unit cost: €501	€13017/year (26 injections/year)
		€26035/year (52 injections/year)
Golimumab (50 mg)	Unit cost: €937	€11247/year
Rituximab (500 mg)	Unit cost: €1346	€10768/year
Tocilizumab (20 mg/ml)	Unit cost: €680	€12377/year
Annual medication administration costs‡			Sécurité Sociale:l’Assurance Maladie^64
Certolizumab pegol (200 mg sc)	10 injections for 3 months then 2 injections/month	€91.4/year (Year 1)€81.9/year (Year 2)
Etanercept (50 mg sc)¶	52 injections/year	€163.8/year
Adalimumab (40 mg sc)	26 injections/year	€81.9/year
	52 injections/year	€163.8/year
Golimumab (50 mg sc)	12 injections/year	€37.8/year
Rituximab (500 mg iv)	2 infusions 2 weeks apart  (1000 mg/infusion) twice a year	€1448/year
Tocilizumab (20 mg/ml iv)	13 injections/year	€4706/year
Total quarterly medication costs
Certolizumab pegol (200 mg sc)	€4372 (first 3 months) then €2623/3 months
Etanercept (50 mg sc)¶	€3256/3 months
Adalimumab (40 mg sc)**	€3930/3 months
Golimumab (50 mg sc)	€2821/3 months
Rituximab (500 mg iv)††	€6108 (first 3 months only)
Tocilizumab (20 mg/ml iv)‡‡	€4271/3 months
Other direct medical costs standardized by disease severity level$			Saraux et al.^50
DAS28 score >3.2	First 6 months	€1215 (SD €405)
	Subsequent 6 months	€1215 (SD €405)
DAS28 score 2.6–3.2	First 6 months	€905 (SD €263)
	Subsequent 6 months	€696 (SD €240)
DAS28 score <2.6	First 6 months	€771 (SD €199)
	Subsequent 6 months	€511 (SD €162)

† Unit costs for acquisition and administration of the biological treatments correspond to retail prices including taxes on 01/04/2013 (i.e. when the risk sharing agreement began).

‡ €3.15 per subcutaneous injection, and €362 per intravenous infusion session.

$ Physician visits, laboratory tests, hospitalizations, imaging, physiotherapy, nursing, adaptive aids, and transportation.

¶ Considering only 50 mg unit dose.

** Considering the mean cost of treatment for a population of patients of whom 80% receive adalimumab 40 mg once every 2 weeks and 20% receive adalimumab 40 mg once/week.

†† A course of two injections, once/year for responders. Switch after 6 months for non-responders (one single course of two injections).

‡‡ For a patient of 70 kg treated once/month at a dose of 8 mg/kg.

ACR, American College of Rheumatology; DAS28, disease activity score 28; EQ-5D, Score as a function of HAQ-DI score; HAQ-DI, Health assessment questionnaire-disability index.

With respect to acquisition and administration costs of biological agents, costs were expressed as retail prices, with all taxes included as of April 1, 2013. No discounting was applied. Unit costs (annual or three-monthly) for the acquisition and administration of the biological agents under consideration are shown in Table 2. For intravenous agents (rituximab and tocilizumab), the national hospital tariff for drug perfusion was used. For subcutaneous agents, the unit tariff for drug administration by a nurse was considered, but not the cost of nurse visits to a patient’s home. The costs of associated treatments (such as synthetic DMARDs or corticosteroids) have not been included in view of uncertainty about extent of use and their very low cost compared to the cost of biological agents.

Concerning other direct costs of follow-up, we applied the values published by Saraux et al.⁵⁰ for France in 2010, based on expert consensus and established using a “standard cost” approach. These cost values are consistent with those subsequently determined directly from a claims database analysis, following adjustment in order to eliminate resource consumption associated with the treatment of comorbidities, notably by comparison with a control group⁵¹. Costs were estimated by valorizing the theoretical consumptions during the first 6 months of follow-up and subsequent 6 month periods according to RA activity measured using the DAS28 score. This score is divided into three categories: DAS28 > 3.2, between 2.6–3.2, and <2.6. We took these values and hypothesized that there was a correspondence between level of RA activity measured with the DAS28 score and ACR20, ACR50, and ACR70 responses, respectively^52–54. This allowed us to define direct medical costs for patients who achieved an ACR70 response, those who achieved ACR50, but not ACR70 (ACR50-69), those who achieved ACR20 but not ACR50 (ACR20–49) and those who failed to achieve ACR20.

A cost-effectiveness analysis was performed comparing the two treatment strategies.

Impact of implementing a pay-for-performance agreement

We evaluated the impact of implementation of a pay-for-performance agreement, in which reimbursement by the payer is conditional on a pre-defined outcome for the patient. In the present case, costs of first-line treatment with CZP were reimbursed in patients whose treatment is interrupted early at 3 months due to an inadequate response (ACR50 response criteria not met) or intolerance.

Sensitivity analyses

Various sensitivity analyses were performed to address uncertainties associated with our hypotheses and variability in the parameters used and to calculate 95% confidence intervals on the main results. For input parameters whose variance was known, the range of variation around their base case value that was tested was equivalent to ±1 SD of the parameter in the base case. In the absence of information on variance, we used linear variations within a range of values considered as plausible by expert rheumatologists.

Table 3 summarizes the variables that were studied in the sensitivity analysis, the coefficients of variation used in the principal central analysis, and the corresponding laws of probability. Wilcoxon tests were used to test the significance of the difference in mean 2-year costs per patient.

Table 3. List of variables used in the sensitivity analysis and coefficients of variation.

Variable	Description	Distribution	Value used in the base case analysis	Standard deviation (SD) or coefficient of variation (CV)
HAQ at baseline		Linear	1.7	1.5–2 (CV)
Reduction in benefit of TNFα inhibitor in Line 2	Reduction in ACR50 response	Linear	1	0.75–1 (CV)
	No response	Normal	0.135	0.0328 (SD)
Gain in HAQ	ACR20-49*	Normal	0.442	0.0357 (SD)
	ACR50-69†	Normal	0.668	0.0509 (SD)
	ACR70	Normal	0.923	0.0624 (SD)
Other direct medical costs per annum according to ACR response^50	ACR20-49*	Normal	€2430	€810 (SD)
	ACR50-69†	Normal	€1601	€500 (SD)
	ACR70	Normal	€1282	€360 (SD)

* Patients who achieve an ACR20 response, but fail to achieve ACR50.

† Patients who achieve an ACR50 response, but fail to achieve ACR70.

ACR, American College of Rheumatology; EQ-5D, Score as a function of HAQ-DI score; HAQ-DI, Health assessment questionnaire-disability index.

Results

Description of the therapeutic pathway

The therapeutic pathways taken by patients starting CZP treatment were modeled for cohorts of 1000 patients (Figure 1). In the CZP-Ref cohort, patients who discontinue CZP go on to a second-line TNFα inhibitor and then, in the case of discontinuation of second-line therapy, to rituximab as third-line and to tocilizumab as fourth-line. In this strategy, the proportion of patients still treated with CZP was 94.5% at 3 months, 80.1% at 1 year, and 70.1% at 2 years.

Figure 1. Number of patients per line of treatment in the CZP-TTT strategy (right) and CZP-Ref strategy (left). [▪]: line 1 (CZP); [•]: line 2 (ETA, ADA, or GOL); [♦]: line 3 (rituximab); ▴: line 4 (tocilizumab). ADA, Adalimumab; CZP, Certolizumab pegol; ETA, Etanercept; GOL, Golimumab.

For comparison, in the CZP-TTT strategy, 55% of patients receiving CZP as first-line therapy are considered as non-responders at 3 months and start second-line treatment with another TNFα inhibitor. The remaining 45% are considered as definitive responders and remain on first-line therapy to the end of the 2-year period, with the exception of 2.5% of the original cohort who are considered to be non-persistent. In the CPZ-TTT strategy, the proportion of patients on second-line therapy gradually declines from 55% at month 3 to 39% at month 24, whereas, in the CPZ-Ref strategy, this proportion progressively accrues from 5.5% to 24.4% over the same period. The proportion of patients on third- or fourth-line therapy at 2 years was 5.6% in the CZP-Ref strategy and 15.4% in the CZP-TTT strategy.

Estimated clinical outcomes

The total mean time spent with an ACR50 response over the 2 years of follow-up was 0.98 with the CZP-Ref strategy and 1.23 years for the CZP-TTT strategy, corresponding to a gain of 3.04 months/patient. The total mean time spent with an ACR20 response was 1.41 years and 1.78 years, respectively. This corresponds, for a cohort of 1000 patients, to a gain of 4.4 months/patient associated with following the CZP-TTT strategy. For an ACR70 response, the total mean times were 0.55 and 0.43 years, respectively, corresponding to a gain of 1.35 months/patient³.

The levels of ACR response shown above translate as a benefit in terms of functional disability measured using the HAQ-DI score, as outlined in the Methods. Each level of ACR response achieved corresponds to a standard decrease in this score. From an HAQ-DI score at treatment initiation of 1.7, the predicted evolution of mean HAQ-DI scores for each of the two therapeutic strategies is presented in Figure 2. As treatment in non-responding patients was switched more rapidly in favor of potentially more effective biotherapy in the CZP-TTT strategy, mean HAQ score in this cohort continued to decrease from the third month. The mean difference in HAQ score between the two cohorts at 24 months was 0.117 points (95% CI = 0.1–0.136).

Figure 2. Evolution in mean HAQ-DI score over the 2-year follow-up period in the CZP-Ref (▪) and CZP-TTT (•) cohorts. CZP, Certolizumab pegol; HAQ-DI, Health assessment questionnaire-disability index; TTT, Treat-to-Target.

Economic impact

Table 4 shows the direct total costs for cohorts of 1000 patients over 24 months. These costs integrate treatment with biological agents over successive lines, including the costs of injections and follow-up. The CZP-TTT strategy is more expensive than the CZP-Ref strategy by €1912 (95% CI = 1498–2355) per patient. This difference is not due to the overall mean retail cost of CZP, which is much lower in the CZP-TTT cohort (€12 455) than in the CZP-Ref cohort (€19 055 per patient), but rather to the costs of second-line treatments and to a lesser extent subsequent lines (third- and fourth-line). The costs of second-line treatment are increased by a combination of two factors, namely the unit cost of a mix of treatments per 3-month period, which is greater than that of CZP, and the fact that patients spend more time on second-line therapy.

Table 4. Mean total direct costs per patient in each cohort over 2 years—costs of different biological agents used and other costs of follow-up (Euros).

Cohort	Line 1	Line 2	Line 3	Line 4	Other direct costs	Total (95% CI)
CZP-TTT	certolizumab pegol: €12455	Mix*: €11309	rituximab: €905	tocilizumab: €809	€3362	€28841 (27 557–30 199)
CZP-Ref	certolizumab pegol: €19055	Mix*: €3814	rituximab: €296	tocilizumab: €156	€3607	€26930 (25 340–28 572)
CZP-TTT + risk-sharing agreement	certolizumab pegol: €10420	Mix*: €11309	rituximab: €905	tocilizumab: €809	€3362	€26806 (25 522–28 164)

* Mix of second-line: 55% (etanercept), 36.4% (adalimumab), and 8.6% (golimumab).

CZP, Certolizumab pegol; TTT, Treat-to-Target.

Table 5. Results of the multivariate sensitivity analysis.

	CZP-TTT vs CZP-Ref
Incremental total costs per patient (24 months) without risk-sharing agreement
Mean (95% CI)	€1912 (1498–2355)
Incremental total costs per patient (24 months) with risk-sharing agreement
Mean (95% CI)	€124 (−538–278)
Difference in gain in HAQ-D1 between T0 and T0 + 24 months
Mean (95% CI)	0.117 (0.10–0.136)

CZP, Certolizumab pegol; HAQ-DI, Health assessment questionnaire-disability index; SD, Standard deviation; TTT, Treat-to-Target.

Cost-effectiveness results and impact of implementing a pay-for-performance agreement

In the context of the pay-for-performance agreement for CZP in the CZP-TTT cohort, the mean cost per patient is reduced to €26 806; thus, generating a mean saving per patient treated first-line of €2035 due to the pay-back associated with the early stop of CZP. Compared to the per capita cost of the CZP-Ref cohort of €26 930, this agreement more than annuls the mean cost difference (−€124) between the CZP-TTT and CZP-Ref strategies in the base-case analysis without pay-for-performance, a difference judged as not economically significant. The CZP-TTT strategy with a pay-for-performance agreement for CZP thus offers superior clinical effectiveness at no additional cost when compared to the CZP-Ref strategy.

Sensitivity analyses

The results of the different univariate sensitivity analyses are presented in Figure 3, which shows the contribution of the different variables analysed to the gain in HAQ-DI and incremental cost of CZP-TTT vs CZP-ref. The reduction in ACR50 response rate of the second-line TNFα inhibitor appears to be the most important contributor to the gain in HAQ-DI and the direct cost of follow-up of patients with an ACR20-49 response is the most important contributor to the variance of the incremental cost of the CZP-TTT strategy.

Figure 3. Univariate sensitivity analysis for the HAQ-DI gain and the incremental 2-year cost of the CZP-TTT vs CZP-R strategy. ACR, American College of Rheumatology; CZP, Certolizumab pegol; HAQ-DI, Health assessment questionnaire-disability index.

Discussion

This study was performed in order to simulate the clinical and economic consequences of implementing a Treat-to-Target strategy in the framework of the risk-sharing agreement for pricing of CZP. The study highlights the interest of original forms of financing of treatment with biological agents. In particular, pay-for-performance agreements between healthcare insurance and the pharmaceutical industry may be expected to contribute indirectly to an improvement of treatment effectiveness by enhancing awareness among rheumatologists about recent EULAR recommendations, as well as to optimizing healthcare expenditure²³. This type of agreement, which has rarely been used to date in France, and never in rheumatology, may be expected to improve the effectiveness of treatments in everyday practice.

In this respect, Treat-to-Target type treatment strategies, based on well-characterized early clinical predictors of treatment response, are particularly suited to setting up pay-for-performance agreements. The analysis showed that a Treat-to-Target strategy was associated with an increase in time spent with an ACR50 response of 3.13 months/patient, and a concomitant gain in functional outcome. Our conclusions are consistent with the findings of a recent real-life study evaluating the impact of a Treat-to-Target strategy performed in Great Britain⁵⁵.

The Treat-to-Target strategy was more expensive than the reference strategy in absolute terms, but this difference was entirely offset by the pay-for-performance agreement. It should, however, be emphasized that the additional cost of the Treat-to-Target strategy is entirely attributable to the higher costs of subsequent treatment lines, which are introduced more rapidly in the Treat-to-Target strategy, and notably of the longer time that patients are receiving more expensive second-line therapies. The 2-year cost of treatment by various sequences of biological DMARDs is determined by the combination of unit cost and treatment duration on each treatment line. Stopping a first-line treatment rapidly as recommended in the TTT strategy and switching to an expensive second line like tocilizumab, for example, as anticipated in the most recent (2014) French practice guidelines³⁸, and which appears to be more effective⁵⁶, will increase the total treatment cost substantially.

Data from randomized clinical trials on the relationship between short- and long- term failure rates for other TNFα inhibitors are fragmentary and difficult to interpret. However, assuming that all TNFα inhibitors are considered to be equally effective, there is no reason to think that the clinical benefit estimates of our analysis may not be generalized to Treat-to-Target strategies using other TNFα inhibitors. The costs of biological DMARDs used in our analysis were, as it is the rule, based on their list prices in 2013. This study has been performed in the context of cost regulation of pharmaceuticals by public payers; it is important to note that, in France, the mechanism for this consists mainly of price/volume agreements, with rebates that remain confidential. Actual net prices of drugs may in fact be inferior to published list prices, leading to a possible over-estimation of our estimates of drug costs.

With respect to consequences for care, this pay-for-performance agreement in France provides an incentive to physicians using CZP to monitor treatment at 3 months and modify treatment if necessary as recommended in the EULAR practice guidelines. This illustrates a rare case of where a risk-sharing agreement may contribute to the implementation of best clinical practice.

The model used was a relatively simple one and no doubt fails to take into account the complexity of therapeutic decision-making in RA. For example, we have assumed a single ACR50 assessment determining future treatment at 3 months. However, in real-life, physicians who implement a Treat-to-Target strategy are likely to have an iterative approach, re-evaluating response on a regular basis and adapting treatment accordingly. This approach would be expected to amplify the differences in effectiveness and cost between the Treat-to-Target and reference strategies. In contrast, in the reference strategy, some notion of effectiveness, albeit one that is difficult to quantify, is likely to underlie decisions to switch. This practice would tend to reduce the observed differences between the Treat-to-Target and reference strategies. We also assumed that patients who were ACR50 responders at month 3 in the Treat-to-Target strategy and continued on first-line therapy would maintain their benefit up to 2 years. However, it is likely that, for certain patients, the early treatment response may be lost and that, in other patients, tolerability issues may justify a change in treatment. Finally, the possibility that patients may discontinue treatment with TNFα inhibitors definitively, rather than switching, has not been taken into account.

Our model has made certain assumptions about the extent of use of different TNFα inhibitors over the course of the 2-year follow-up, and the order in which they are used. For example, with the publication of international practice guidelines recommending a treat to target strategy², treatment persistence at 1 year under first line biotherapy is decreasing, as rheumatologists choose to switch ineffective treatments earlier. These assumptions have implications for the cost evaluation. Nonetheless, the price differential between the different TNFα inhibitors is relatively low, and patterns of use of these biological DMARDs have been relatively stable over recent years. In addition, we assumed that all DMARDs would be administered by a nurse (in hospital for IV agents and at home for SC agents). However, most patients in France usually self-inject SC anti-TNFα agents, so this assumption leads to some inflation of cost for first- and second-line therapies. However, given that administration costs only account for ∼1% of total SC drug costs, such cost inflation will be marginal.

The model used in our study was based on ACR50 response as the main criterion for stopping treatment at 3 months. The choice of this outcome measure was guided by the use of the ACR response criteria as the primary outcome measure in clinical trials of CZP^57,58 and as the response variable in the meta-analysis of clinical trials of TNFα inhibitors in RA⁴¹. However, our model is a generic one, and the relative benefit of the Treat-to-Target strategy over the reference strategy is independent of the criterion used. Changing the criterion would only alter the proportion of patients to be switched at 3 months in the Treat-to-Target strategy, rather than the relative effectiveness of the two management strategies. For example, similar results would have been obtained using the DAS28 or the CDAI to determine the response, which are simpler and more straightforward outcome measures for use in everyday practice^59–61. The treatment goal in the 2013 EULAR practice guidelines² is discernible improvement by 3 months and remission or low disease activity at 6 months. Remission is defined either as scores on the tender joint count, swollen joint count, CRP and patient global assessment (0–10 scale) all ≤1, or as an SDAI score ≤3.3⁶², which is a very stringent treatment goal.

A limitation of our study is that indirect costs were not taken into account. The rationale for this was that the cost-utility analysis was performed from a health insurance perspective. In RA, indirect costs represent a larger proportion than direct costs⁶³ and healthcare initiatives to reduce the total cost of RA should, thus, take into account indirect costs as well. Nonetheless, indirect costs have been shown to be directly proportional to the level of functional disability and the intensity of symptoms measured, for example, with the HAQ-DI score. For this reason, it is likely that our conclusions would be strengthened further by taking into account indirect costs.

Conclusions

A “Treat to Target” strategy for use of a TNF-inhibitor as first-line therapy consisting of continuing therapy in patients with an adequate clinical response at 3 months, but stopping treatment in inadequate responders and switching them to second-line biotherapy, provides superior clinical benefit but yields higher costs. Implementing a pay-for-performance agreement for CZP, consisting of reimbursing the health Insurance for medication costs incurred by patients stopping treatment at 3 months, offsets the extra cost. This pay-for-performance agreement contributes to reinforcing the application of a treat to target strategy and may, thus, provide better disease control to patients with RA at no incremental cost.

Transparency

Declaration of funding

This study was funded by UCB Pharma.

Declaration of financial/other relationships

Cemka-Eval is a consulting company which received an unrestricted grant from UCB Pharma France to perform this analysis, with the support of a scientific committee of expert rheumatologists who received honoraria to participate in this project. FF has received a research grant from UCB Pharma. TP has declared links with, and PC has declared consultancy/advisory work for Abbvie, BMS, Pfizer, UCB Pharma, Roche, MSD, Novartis, and Janssen. TDC is currently or has been employed by UCB Pharma and Amgen SAS. JMJ is an employee of UCB Pharma. JME peer reviewers on this manuscript have no relevant financial or other relationships to disclose.

Acknowledgments

The authors acknowledge Costello Medical Consulting, UK, for editorial assistance, which was funded by UCB Pharma.