Cost-effectiveness analysis of treatment sequences containing tofacitinib for the treatment of rheumatoid arthritis in Spain

The efficiency of the treatment sequences containing tofacitinib was calculated by adapting a cost-effectiveness analysis model, previously used in other countries’ healthcare settings [11‐13], to the Spanish context. Main characteristics of this model have been published [11, 12]. Its consists of a microsimulation, the structure of which is based on models used previously to assess the efficiency of other bDMARs in RA [14‐16]. The patient-level simulation is a widely-used technique for RA modelling due to the clinical management characteristics, duration and use of sequential treatments. Unlike traditional Markov models which analyse cohorts of patients, in microsimulation’ models individual patients move through the different stages of the model one-by-one during the simulation. This allows time to be calculated in each treatment line and the associated costs to be applied to the first treatment cycle (such as induction therapy or patient education costs). Moreover, it also allows a patient history to be compiled to determine future events, time on treatment, costs and quality of life.

Our analysis included 100,000 patient-level simulation iterations, based on which the average costs and health outcomes of the treatment sequences assessed were obtained. The Incremental cost-effectiveness ratio (ICER) was determined among the treatment sequences assessed through an incremental comparison of costs and outcomes, applying an annual discount rate of 3% [17].

The analysis period covered each patient’s lifetime, whose pathway through the model was assessed in six-month cycles (Fig. 1). This period was deemed to be the most suitable interval for performing follow-up that allowed to detect changes in the disease’s course and to compile the resources consumption, as reported monitoring frequency in clinical trials and routine clinical practice assessment [18].

Disease progression was determined using the Health Assessment Questionnaire (HAQ) score. Mortality and quality of life are related to the HAQ score, rather than being associated with a specific state of health.

During the simulation, patients are allowed to stop their current treatment and move to next treatment line established in the sequence within the first six months of treatment (based on HAQ score variation), due to subsequent loss of response (> 6 months of treatment) or following the onset of serious adverse events (SAEs).

Patient’s HAQ score and their score-adjusted likelihood of suffering a SAE, are determined in each six-month cycle, conditioning the continuity of the therapy and ensuring that patients who remained on treatment are those benefitting from therapy. If a patient’s HAQ score changes enough to be considered an inadequate initial response or a loss of treatment response, the patient moves on to the next treatment line and returns to the start of the model diagram with the new treatment and an initial HAQ score equivalent to the pre-treatment score, thereby preventing patients with the greatest frequency of treatment changes from presenting accumulated benefits.

Treatment response was set to be assessed at three different intervals: short-term (first 6 months), medium-term (6–36 months) and long-term (> 36 months of treatment). The variation in HAQ score required for switching varies according to time. Treatment response is more pronounced in the first six months and LTE studies have demonstrated that improvements in HAQ become less sensitive to therapy over time (Fig. 1).

After starting a new therapy, to consider an adequate response at month six the improvement in the HAQ score has to be at least 0.35, and subsequently any decrease in the HAQ score would mean changes of less than 0.35 every six months. These thresholds were estimated by using a linear regression function for conversion of a 1.2 point improvement in the DAS28 (Disease Activity Score) to HAQ scores. Data from a trial comparing etanercept and MTX (TEMPO trial) [19] were used to establish the relationship between DAS28 and HAQ, based upon the expert opinion which stated that the minimum improvement in DAS28 required when a patient initially starts a new therapy is 1.2.

The study included RA patients who had active moderate-to-severe RA and who were eligible for treatment with tofacitinib, according to the authorized indication [9]. In one group, RA patients DMARDs-IR, while the other one included RA patients TNFi-IR population [20].

The profile of both populations (Table 1) was defined by the age and duration of RA of patients in the Biobadaser database [22]. Moreover, the average weight of the Spanish population aged over 45 [23] and the initial HAQ score of the DMARDs-IR population in clinical trials with tofacitinib were considered. These were similar to HAQ value in the Spanish CREATE registry [21] (patients with inflammatory rheumatic disease starting treatment with bDMARDs) based on the initial DAS28 score (HAQ = − 0.185 + (0.289 × DAS28)) (Table 1).

This model did not compare two therapies individually, but it assessed therapeutic sequences with up to five possible treatment lines. All the therapies in all the sequences include combination with MTX.

The analyses included two scenarios for the DMARDs-IR population, and two scenarios for the TNFi-IR population, where the patients had received MTX in first-line therapy and a TNF inhibitor in the second line, before starting the sequences assessed in the analysis.

Among the wide variety of combinations of therapies, the preferred sequences as most likely to be used in clinical practice were defined by the expert panel (Fig. 2).

These sequences were compared with an alternative sequence after replacing by tofacitinib (5 mg twice daily [BID]) the initial therapy of the sequences, in order to assess, the efficiency of positioning tofacitinib at initial therapy of the treatment sequence.

For TNFi-IR population, sequence was selected to include mechanism of actions others than anti-TNFs. The efficiency of tofacitinib was assessed by replacing subcutaneous tocilizumab by tofacitinib or by positioning tofacitinib before the administration of tocilizumab.

The active sequences including tofacitinib as initial line therapy were compared with the comparator sequences containing other bDMARDs. Dosages for these bDMARDs derived from the correspondent Summaries of Product Characteristic.

As it was done in previous evaluations with the global model [11, 12], the clinical efficacy data, in terms of HAQ score, for the initial treatment response, were obtained from a meta-analysis derived from a systematic literature review of studies on the efficacy of different treatments for moderate-to-severe RA [24, 25]. For therapies in which no efficacy data were found in terms of the HAQ score, efficacy was determined using the ACR values obtained from the literature, by means of the conversion proposed in National Institute for Health and Care Excellence (NICE) Technology Appraisal 198 [26].

Limited data about HAQ change in TNFi-IR population were identified in the systematic literature for the network meta-analysis [24]. When available the effectiveness data on the TNFi-IR population were applied to the population of patients who had changed treatment from a TNFi. Otherwise, we adopted the information available on csDMARD-IR patients.

Medium-term responses were obtained from the literature [27‐33] and there was assumed to be no long-term progression in patients receiving treatment with DMARDs (Table 2).

The model assesses the safety profile of tofacitinib in comparison to bDMARDs. Given the heterogeneity of their SAEs reported, we opted to model them all together, considering the onset of severe infections (SI) as being representative of the most common SAEs in RA patients, and applying the frequencies published [35], in line with the approaches previously done [11, 34].

The proportion of patients that switched to the following treatment on the defined sequence if an SAE occurred was established at 74%, for all the therapies included, based on the analysis of individualized data from the studies identified in the indirect comparison carried out [24], and in the same way that previous studies with the global model [12].

The functional status of the patients and RA severity are directly related to increased mortality in RA [36] so additional benefits will be associated to disease progression reduction, resulting from a reduction in mortality.

In this microsimulation model [11], mortality rate by age and gender due to any cause in the Spanish population was obtained from the Spanish National Institute of Statistics [23] and it was adjusted based on the corresponding HAQ value, using the ratio described in previous publications [37].

The effect on health-related quality of life was considered in the model by estimating the quality-adjusted life years (QALYs). The correlation between the HAQ score and utilities was estimated [38] (Table 2), moreover and similarly to previous modellings [11] over a four-week period, which would equate to a QALY loss of 0.012 in patients with SAEs.

Our analysis was developed from the perspective of the Spanish NHS. Due to lack of accurate evidence, the societal perspective was not assessed. The direct healthcare costs included were: drug-acquisition costs, administration costs, disease management costs according to severity and SAE costs.

Acquisition costs were estimated using the ex-factory price (EFP) [39] including the corresponding mandatory deduction outlined in Royal Decree 08/2010, the reference price established or the lowest EFP among the available biosimilars. These costs were calculated for two different periods: cost in the first six months of treatment (including induction dose where appropriate) and half-yearly cost from the sixth month of treatment, considering the dosages established in the summaries of product characteristics which are shown in Table 3. In the case of rituximab, cycles of two infusions are recommended, with the need for retreatment assessed 24 weeks after the last cycle. To reflect clinical practice in Spain, the administration of further rituximab cycles was considered every nine months [41] by weighting the cost to the six-month cycle period.

Administration-related costs included the cost of parenteral administration at the day hospital for intravenous drugs, or the cost of patient’s education in self-administration of subcutaneous medicines. The unit costs of these resources were obtained from eSalud [40], a Spanish database covering healthcare costs in the country.

The costs of managing RA according to severity were estimated using the published evidence [4]. The costs of managing a SAE were equated to managing a SI, and it was derived from the average costs per diagnosis-related group and autonomous community fees for the types of infection selected which, in line with the available evidence [42] and after validation by a panel of clinical experts, included these SI (without complications): septicaemia, acute bronchitis, pneumonia, pulmonary tuberculosis, kidney and urinary tract infections and respiratory inflammation.

All costs were expressed in euros (2018 values) (Table 3).

Considering the stochastic feature of the microsimulation models, a probabilistic sensitivity analysis (PSA) were carried out with the implementation of two nested loops. An inner loop reflecting the first-order uncertainty in which mean cost and mean QALYs are calculated from the 1000 simulated patients sample, by applying a normal distribution to cohort characteristics inputs (age, HAQ initial score, weight and RA duration). The outer loop allows for the second-order variation by randomly varying parameters around the HAQ relationships (utilities, mortality, costs) by randomly drawing the parameter from a normal distribution. The first loop was repeated for different parameters set up to obtain 100 cost-effectiveness estimations.