The incremental cost-effectiveness ratio (ICER) of adalimumab+MTX versus MTX monotherapy was assessed by modelling, via microsimulation, the associated direct (and, in a scenario, indirect costs) and quality-adjusted life-years (QALYs) over 30 years from a UK (UK) National Health Service (NHS) perspective.

The PREMIER study is the principal evidence source used to derive model parameter values. Detailed results of this pivotal randomised clinical trial conducted over 2 years in MTX-naïve patients with early aggressive RA, comparing adalimumab as monotherapy (n=274) or combined with MTX (n=268) versus MTX monotherapy (n=257), have been described elsewhere.15 Additionally, economic data were derived from the health economic companion study to PREMIER, Study DE032.30 The characteristics of patients in the companion study (N=664) were similar to those included in the full PREMIER trial (N=799) in terms of age, gender, disease duration, prior DMARD use, disease activity, HAQ score and discontinuation.15 30 Response rates were slightly higher in DE032 (72% in adalimumab+MTX vs 50% in MTX in DE032 vs 62% in adalimumab+MTX vs 46% in MTX in the full PREMIER data). Since the relationships between modelled variables (eg, the effect of disease activity and Total Sharp Score on HAQ, and the estimation of resource use as a function of HAQ) are independent of treatment effect, data from all Study DE032 patients were used in the quantification of these relationships.

The microsimulation model generates individual patient histories. These histories tell the story of each individual patient's disease pathway in terms of his/her therapy and its consequences on disease activity and joint damage, and the subsequent consequence of these on survival, costs and quality of life. While side effects and mortality may occur any moment, treatment decisions are made in 26-week cycles, for 1000 patients initiated on adalimumab plus MTX therapy or MTX monotherapy (figure 1).

The histories are generated using analyses which have been carried out on the information available in the PREMIER clinical trial.15

Each patient history starts when they enter the model (figure 1) with a given score on the “28-joint Disease Activity Score” (DAS28): which reflects the number of tender joints (0–28), number of swollen joints (0–28), a pain score (0–100 mm visual analogue scale), and a value for the C reactive protein (CRP) concentration. This is a short-term reflection of disease activity. Additionally, each patient is given a score on the modified Total Sharp Score (mTSS), which assesses radiological damage among 44 joints and which reflects long-term irreversible damage.

For each patient, the starting scores on the four DAS-28 domains are computer generated by drawing at random from a multivariate normal distribution which was estimated using the sample means and covariance of the logit of the four domains. The effect of therapy was modelled by shifts in the distribution means, reflecting improvements. These shifts are again generated using a multivariate normal distribution as estimated by the sample means and covariance as observed in PREMIER. Naturally, while the shifts in the means capture the improvements in the scores, the covariance matrix captures the positive correlation between the four domains. As a consequence of the fact that this correlation is not perfect, one will find that patients improve on one, but not necessarily on another, domain of the DAS score.

The initial score for the mTSS is generated by drawing from the sample distribution at the start of the model. This variable assesses radiological damage among 44 joints and captures joint destruction or deformity which may result from irreversible joint erosion and joint space narrowing, the long-term effects of RA. The effect of treatment on mTSS was modelled as a linear function of treatment (tx) and time (months). The effect for subsequent treatment strategies are assumed to be the same as those applied for MTX (please refer to supplementary material for additional information).

The DAS28 variables are then used in an ordered logistic regression to predict ACR responses at 26 weeks. After 26 weeks, patients who failed to meet the 50% improvement criteria of the ACR50 switched to the next treatment in the sequence of three other treatment strategies, which was identical for both treatment arms: first MTX+hydrochloroquine, then leflunomide monotherapy and then gold monotherapy. This sequence of treatments was chosen because it includes therapies that target RA in different ways. Each treatment is thus considered a suitable alternative for patients who failed on the previous treatment in the sequence. The order of the treatments in the sequence was chosen such that the cost of treatments increases with each failed option. Further, the sequence of treatments that we apply has been used previously in economic modelling of biologics in RA.24

Patients reaching ACR50 on a given therapy remained on that therapy until discontinuation due to adverse events (AEs). If an ACR50 response was not reached in a given cycle, patients discontinued their current treatment and switched to the next therapy in the sequence. Patients failing to reach ACR50 on any of the four treatment strategies were assumed to switch back to, and remain on, the therapy that provided best response (which was estimated in the same way as when prescribed the first time) until discontinuation due to AEs. AE-related discontinuations were estimated to have occurred in 2% of patients annually regardless of therapy on the basis of the PREMIER randomised clinical trial.15 Assumptions about the efficacy of secondary DMARDs after insufficient response to either primary treatment were obtained from a Swedish study of patients with established RA.31 Details of model estimates are provided in the online supplementary appendix.

Disease activity and mTSS were linked to an individual's HAQ using an equation that was based on an analysis of PREMIER and adjusted for the observation that the relationship between mTSS and HAQ was more pronounced in patients when in remission (please refer to supplementary material for additional information) (3).

The probability of death at any point in time was related to the patient's age and HAQ score using the relationship reported by Sokka et al,32 for the data from PREMIER were not suitable to estimate survival rates.

Economic data (from the PREMIER health economic subset), which included employment status (employed, unemployed, self-employed, homemaker, student, retired, other), work-related information (average working time, work days missed and self-reported visual analogue scale measuring the degree to which RA affected normal work or homemaking performance), and healthcare resource use (number of hospital visits and duration of hospital stays and number of visits to a general practitioner (GP)), were mathematically linked to a patient's HAQ score at any point in time. Again, these links are described elsewhere.33

Further details on the relationships described above can be found in the online supplementary appendix.

Utilities, derived from the Health Utilities Index Mark 3 (HUI3), were estimated based on data from PREMIER. A logistic curve was applied, with utility as the dependent variable and HAQ score as the explanatory variable (see online supplementary appendix).

Treatment costs were based on estimates from the September 2013 British National Formulary.34 The cost of adalimumab therapy was based on 40 mg treatment every other week by self-injection, whereas the cost of MTX assumed a treatment strategy of 7.5 mg weekly, administered orally. The annual costs of MTX and adalimumab therapies were estimated to be £22.92 and £11 269.18, respectively, taking into account the PREMIER-based proportion of patients who received dose escalations. The annual costs of other DMARD therapies were estimated to be £31.33, £248.14 and £700.96 for hydrochloroquine, leflunomide and gold, respectively. Unit costs for hospital and GP visits were taken from the Personal Social Services Research Unit Costs of Health and Social Care, and are inflated to 2012 prices using the Hospital and Community Health Services Index.30 35 36 A hospital visit cost £290.00 per hospital bed day, and a GP visit cost £29.83. The cost of a day missed from work was estimated to be £89.60,37 considering the UK average weekly wage.

The primary outcome was the discounted ICER, which was assessed in the base case analysis and various scenario and probabilistic sensitivity analyses. Unless otherwise specified, the costs and QALY outcomes were discounted at 3.5%, as per current National Institute for Health and Care Excellence (NICE) research guidelines.38

The following alternative scenarios were considered: (1) the effects of the mTSS on HAQ (ie, the irreversible effects of untreated RA) were ignored; (2) mortality effects were ignored; (3) treatment response was defined as 70% (ie, ACR70); (4) the impact of an additional HAQ point on utility was assumed to be halved to test the impact of choosing an HRQL measure other than the HUI3 to model utility; (5) the model analytical horizon was shortened to 2, 5 and 10 years (alternating between including and ignoring the irreversible damage); and (6) discontinuation rates of 0% and 4% were tested.

A probabilistic sensitivity analysis (PSA) combined bootstrapping with random draws from uncertainty distributions. By bootstrapping the data from the randomised clinical trial using normal and Poisson distributions, we obtained uncertainty margins surrounding the annual increase in mTSS; the distribution of initial response; the relationship between clinical response variables and HAQ; and the relationship between HAQ and utility, loss of work, days in hospital, doctor visits and mortality.

All remaining estimates were from sources other than PREMIER. For these estimates, uncertainty distributions were defined as follows and were used to draw from each bootstrap. The base case annual incidence of AEs was multiplied with a factor that followed a Programme Evaluation and Review Technique (PERT) distribution, a version of the β distribution, with mean=1, minimum=0.5, and maximum=1.5. A similar approach was chosen for the parameter modelling the dependency between HAQ and mTSS and those for the efficacy and AEs of later therapies. Less uncertainty was assumed concerning the unit costs of days missed from work and costs of hospital stays and GP visits (PERT distributions for multiplication with mean=1, minimum=0.8 and maximum=1.2). The PSA generated 250 simulations each with 1000 patients per treatment arm. Results are presented via a cost-effectiveness acceptability curve.

This research was conducted by Pharmerit International with the financial support of AbbVie Inc. AbbVie Inc. participated in the design and conduct of the study and in the interpretation of the data.

The clinical outcomes at 2 years mimic the data observed in PREMIER despite the fact that, in contrast to the trial, the model assumed that patients would begin to switch medications after 6 months and could try up to three other agents before reverting to the treatment on which they experienced the best response. Unless otherwise specified, all results exclude indirect costs.

Figure 2 presents the distribution over time of the 1000 patients on the various treatment sequences for each arm of the model. As depicted on figure 2, there is a precipitous drop of patients on initial MTX monotherapy after 26 weeks because 46% of these patients did not achieve ACR50. This drop is less pronounced for combination therapy because 75% of patients achieved ACR50. These estimates are slightly higher than the response rates seen in the patient-level data. This may be explained by the use of data from all Study DE032 patients to inform the estimates of intervariable relationships. Thereafter, patients switched from their initial therapy only because of AEs. Response rates for the next treatment lines were estimated to be much lower, and a sizeable percentage of patients ended up in the last option of best previous treatment (17% MTX patients, 5% combination patients). The average mTSS increases almost linearly from a base line value of 18 to an average value of 106 with combination therapy and an average value of 157 with MTX.

The differences in clinical response presented in figure 2 result in differences in HAQ scores and associated modelled consequences in terms of resource utilisation (GP visits, hospital days), survival and utilities, which are illustrated in figure 3. Patients started with a relatively high average HAQ score of approximately 1.5, which decreased to an average of approximately 0.7 for patients who started on combination therapy (black line, top left panel, figure 3) and to approximately 0.9 for patients who started on MTX monotherapy (grey line, top left panel, figure 3). These curves reflect a weighted average of responders and non-responders. MTX monotherapy was also associated with lesser HRQL (average difference of 0.21 on the utility scale between the two treatment arms) and greater resource use in terms of both GP visits and hospitalisations. Table 1 presents an overview of the estimated differences in clinical and costs outcomes after 30 years.

Discounted life expectancy was estimated to be 12.62 versus 9.94 for combination therapy versus MTX monotherapy, respectively, an incremental gain of 2.68 life years in the combination treatment arm. Discounted QALYs were 6.83 versus 3.79, respectively, a gain of 3.04 QALYs in the combination treatment arm. Patients who started on combination therapy were estimated to remain on their initial therapy for an average of 13.32 years compared with 6.62 years for patients who started on MTX monotherapy. The associated discounted cost of medication were estimated to be £108 805 and £2 589, respectively, corresponding to a net cost of £106 217 favouring MTX. However, the more effective combination therapy was also associated with savings in terms of hospitalisations and GP visits, such that the total net cost for combination therapy was estimated to be £98 558. Accordingly, the ICER excluding indirect costs was estimated to be £32 425. When indirect costs were included in the analysis, the ICER decreased to £27 238.

Table 1 also shows the balance between costs and effects when four key alternate scenarios are modelled. Specifically, excluding the effects of irreversible damage on HAQ led to more modest increases in patient lifespan and incremental QALYs, which had a dramatic effect on the cost-effectiveness ratio; it increased to >£70 000. When survival benefits were ignored, the ICER improved to approximately £23 000. This improvement was driven primarily by increased costs associated with more patients surviving in the MTX arm, which resulted in decreased incremental costs. When therapy was stopped and replaced with the next treatment in the sequence in patients who did not achieve an ACR70 response, fewer patients received biologics in the combination therapy arm and less time was spent on this therapy. However, this change did not have a marked effect on the ICER: the costs decreased but the savings and the benefits also decreased, with the ICER changing from £32 425 to £30 198/QALY. Assuming an attenuated relationship between HAQ and utility (as may be expected if a utility measure other than the HUI3 is used to model patient-related quality of life) increased the ICER to £38 611/QALY owing to fewer incremental QALYs observed.

Using analytic horizons of 2, 5 and 10 years resulted in ICERs of £95 947, £56 014 and £37 948, respectively.

When shorter time horizons were assumed, the full benefit of treatment in terms of preventing or slowing the progression of RA was not realised. This is demonstrated by the modest differences in QALYs seen between the treatments in the shorter term. Further, when the effects of mTSS on HAQ were also excluded, the ICERs for the 2, 5 and 10 years horizon analyses increased to £190 481, £126 756, and £90 249, respectively. The relative contribution to the ICER of the irreversible damage component is thus 49.6%, 55.8% and 58% at 2, 5 and 10 years, respectively. When the discontinuation rate was set to 0% and 4%, the ICERs were £32 494 and £32 315, respectively.

Figure 4 presents the cost-effectiveness acceptability curve. The probability that combination therapy is cost-effective was estimated to be 100% at willingness-to-pay thresholds of £45 000 and above when the base case model assumptions were applied (figure 4). The estimates, conditional on the model used in the base case here, were extremely robust.