Ustekinumab for treating active psoriatic arthritis

Clinical effectiveness

3.1 Evidence on the clinical effectiveness of ustekinumab was taken from 2 clinical studies – PSUMMIT 1 and 2. Both were randomised, double‑blind, placebo‑controlled, phase III studies in adults with active psoriatic arthritis despite current or previous treatment. The studies were almost identical, except for the previous treatment: PSUMMIT 1 (n=615) included people who had previously had disease‑modifying antirheumatic drugs (DMARDs) with or without non‑steroidal anti‑inflammatory drugs (NSAIDs) only, whereas PSUMMIT 2 (n=312) also included people who had previously had tumour necrosis factor (TNF) alpha inhibitors. People in both trials generally had long‑standing moderate to severe active psoriatic arthritis with impaired physical function and high numbers of tender and swollen joints. In both PSUMMIT 1 and 2, approximately 70% of patients had skin disease, and 80–90% of patients had received prior DMARD therapy. Of the 180 people in PSUMMIT 2 who had previously had TNF‑alpha inhibitors (referred to in this document as 'TNF‑alpha inhibitor‑exposed'), more than half had received at least 2 biological drugs. In both studies, patients were randomised to ustekinumab 45 mg or 90 mg (administered at 0 and 4 weeks, then every 12 weeks thereafter) or placebo. People in the placebo group switched to have ustekinumab 45 mg after 16 weeks (if they had less than 5% improvement in both tender and swollen joint counts) or 24 weeks (all others), and people whose disease did not respond to the 45‑mg dose of ustekinumab switched to 90 mg after 16 weeks. People in the studies were followed for up to 100 weeks in PSUMMIT 1 and 52 weeks in PSUMMIT 2.

3.2 The primary end point in the PSUMMIT 1 and 2 trials was the American College of Rheumatology (ACR) 20 response rate at week 24. This is defined as an improvement of 20% or more in swollen and tender joint counts, and an improvement of 20% or more in 3 of 5 assessments of pain, disease activity and physical function. Secondary end points included measures of joint symptoms (including modified Psoriatic Arthritis Response Criteria [PsARC] and ACR 50/70), skin lesions (Psoriasis Area and Severity Index [PASI]), soft tissue symptoms, radiographic response, and disability and quality of life (Health Assessment Questionnaire Disability Index [HAQ‑DI], Dermatology Life Quality Index [DLQI] and 36‑item Short‑Form Health Survey [SF‑36]).

3.3 In both PSUMMIT 1 and 2, ustekinumab was associated with statistically significantly higher rates of ACR 20 response at week 24 than placebo. ACR 20 response rates in PSUMMIT 1 were 42.4%, 49.5%, 46.0% and 22.8% for ustekinumab 45 mg, ustekinumab 90 mg, ustekinumab 45 mg and 90 mg pooled, and placebo respectively (p<0.0001 for ustekinumab compared with placebo). Ustekinumab showed similar effectiveness compared with placebo regardless of prior exposure to TNF‑alpha inhibitors. ACR 20 response rates in PSUMMIT 2 for ustekinumab 45 mg, ustekinumab 90 mg, ustekinumab 45 mg and 90 mg pooled, and placebo respectively were:

3.4 Ustekinumab also demonstrated similar efficacy regardless of concomitant methotrexate use. ACR 20 response rates in PSUMMIT 1 for ustekinumab 45 mg, ustekinumab 90 mg, ustekinumab 45 mg and 90 mg pooled, and placebo respectively were:

3.5 Longer‑term analyses of the primary outcome suggested that response rates with ustekinumab were maintained over 52 weeks. Response rates in the placebo arm increased after week 24 because of people switching from placebo to ustekinumab.

3.6 The results from secondary outcome analyses at week 24 generally supported the conclusions from the ACR 20 results. The findings were observed across joint, radiographic, skin, soft tissue and health‑related quality‑of‑life end points, although the results varied between outcomes and between trials, and not all outcomes reached statistical significance in all analyses. For example, for all randomised patients in both PSUMMIT 1 and 2, PsARC response rates with ustekinumab (45 mg and 90 mg pooled) and placebo were 58.0% and 35.2% respectively; PASI 75 response rates (people who had at least 75% improvement in PASI score) with ustekinumab (45 mg and 90 mg pooled) and placebo were 57.6% and 8.8% respectively. For all randomised patients in PSUMMIT 1, the median HAQ‑DI changes from baseline with ustekinumab (45 mg and 90 mg pooled) and placebo were −0.25 and 0.00 respectively (p<0.001). For all these examples, results were similar in individual trials and for individual doses. Longer‑term analyses of PASI 75 responses suggested that response rates with ustekinumab were maintained over 52 weeks. Long‑term analyses of other secondary outcomes were provided as academic in confidence and therefore cannot be reported here.

3.7 In the absence of head‑to‑head randomised controlled trials, the company presented a mixed treatment comparison using a random‑effects model fitted with Bayesian methodology to explore the clinical effectiveness of ustekinumab compared with TNF‑alpha inhibitors (adalimumab, etanercept, golimumab and infliximab) in people who had not previously had TNF‑alpha inhibitors (referred to in this document as 'TNF‑alpha inhibitor‑naive'). The company did not carry out a mixed treatment comparison for the TNF‑alpha inhibitor‑exposed population because PSUMMIT 2 is the only trial to have included this population. The mixed treatment comparison focused on PsARC, PASI 75 and PASI 90 responses to treatment at weeks 12–16 and 24, which are consistent with the clinical parameters in the economic model. For ustekinumab, patient‑level data were extracted from PSUMMIT 1 and 2 for a weight‑based dosing subgroup in which patients who weighed 100 kg or less had ustekinumab 45 mg, and patients who weighed more than 100 kg had ustekinumab 90 mg. For the TNF‑alpha inhibitors, data were taken directly from 7 randomised, double‑blind, placebo‑controlled studies carried out in people with active psoriatic arthritis. The company reported that the findings showed that ustekinumab and TNF‑alpha inhibitors have better outcomes than placebo in most analyses. The results of the mixed treatment comparison for PsARC and PASI were marked as academic in confidence and cannot be reported here. It noted that, for the analysis of joint symptoms (PsARC), probabilities of response for ustekinumab were lower than for the TNF‑alpha inhibitors, although the 95% credible intervals for ustekinumab 45 mg overlapped with those of adalimumab, golimumab 50 mg and infliximab. The company reported that in the analyses of skin symptoms, there may be higher probabilities of response with infliximab (PASI 75 and PASI 90), golimumab 100 mg (PASI 75) and adalimumab (PASI 90) compared with other biological drugs, although the credible intervals were overlapping.

3.8 The company presented adverse event data from the PSUMMIT studies, 5‑year extensions of 4 studies of ustekinumab in psoriasis, the Psoriasis Longitudinal Assessment and Registry and the British Society for Rheumatology Biologics Register. The incidence of adverse events in the PSUMMIT studies was similar in the ustekinumab treatment arms compared with the placebo arms. For all randomised patients in PSUMMIT 1 and 2, the incidences were: ustekinumab 45 mg, 48.4%; 90 mg, 49.4%; 45 mg and 90 mg combined, 48.9%; and placebo, 47.9%. There were no disproportionate increases in adverse event rates over time. The most common adverse reactions seen with ustekinumab in the PSUMMIT trials included nasopharyngitis, upper respiratory tract infection, headache, arthralgia (joint pain), nausea and diarrhoea. The overall rates of study discontinuation because of adverse events were low (and higher with placebo than with ustekinumab); the rates were 3.4%, 1.5% and 1.5% for placebo, ustekinumab 45 mg and ustekinumab 90 mg respectively, in the placebo‑controlled period. The psoriasis extension studies and register data reported no clear dose–response effect or cumulative exposure effect for ustekinumab, and suggested that the rates of serious adverse events were comparable between ustekinumab and TNF‑alpha inhibitors.

Cost effectiveness

3.9 The company presented a de novo economic analysis that assessed the cost effectiveness of ustekinumab for treating adults with active psoriatic arthritis for whom the response to previous DMARD therapy has been inadequate. Ustekinumab was compared with TNF‑alpha inhibitors and conventional management in people who were TNF‑alpha inhibitor‑naive, and with conventional management only in people who were TNF‑alpha inhibitor‑exposed. The patient populations were based on the populations in the company's mixed treatment comparison and PSUMMIT 1 and 2. It was assumed that all patients who weigh less than 100 kg would have ustekinumab 45 mg, and all those who weigh more than 100 kg would have ustekinumab 90 mg. The model comprised a short‑term decision tree followed by a long‑term Markov model with a lifetime (52‑year) time horizon. It was similar to the models used in previous NICE appraisals of treatments for psoriatic arthritis (etanercept, infliximab and adalimumab for the treatment of psoriatic arthritis and golimumab for the treatment of psoriatic arthritis). In the decision tree phase, people had initial biological therapy for either 12 weeks (TNF‑alpha inhibitors) or 24 weeks (ustekinumab). At this point, people who had a PsARC response (defined as an improvement in at least 2 of the 4 criteria, 1 of which must be joint tenderness or swelling score, with no worsening in any of the 4 criteria) continued with biological therapy, and those who did not switched to conventional management. All patients then entered the Markov phase of the model. People having a biological therapy continued to have it until they switched to conventional management, either because the biological therapy lacked efficacy or led to adverse events (at a rate of 16.5% per year for all treatments), or they died. No second biological drug was permitted. The model considered costs from an NHS and personal social services perspective, and all costs and health effects were discounted at a rate of 3.5% per year.

3.10 For the TNF‑alpha inhibitor‑naive population, ustekinumab was compared with 4 TNF‑alpha inhibitors and conventional management, using clinical effectiveness evidence from the mixed treatment comparison. For the TNF‑alpha inhibitor‑exposed population, ustekinumab was compared with conventional management only, because at the time of the submission there were no randomised controlled trials of TNF‑alpha inhibitors in this population. Analyses of this population were based on clinical effectiveness evidence from the TNF‑alpha inhibitor‑exposed sub‑population of PSUMMIT 2.

3.11 The model captured health‑related quality of life through joint symptoms, disability and skin symptoms (PsARC response, HAQ‑DI score and PASI score). People who had a PsARC or PASI response were assumed to have a fixed improvement in HAQ‑DI or PASI score respectively. This improvement was maintained until a switch to conventional management, at which point the score returned to its baseline value (rebounded). People who did not have an initial response were assumed to have a smaller improvement in HAQ‑DI or PASI score until withdrawal of active treatment. Throughout periods of conventional management, people's disease progressed according to the natural history of psoriatic arthritis, modelled as a linear increase (worsening) in HAQ‑DI over time and a constant PASI score. HAQ‑DI and PASI scores were then mapped to EQ‑5D using an equation used in previous NICE technology appraisal guidance for etanercept, infliximab and adalimumab for the treatment of psoriatic arthritis and golimumab for the treatment of psoriatic arthritis. Costs and disutilities associated with adverse events were not included in the model. Healthcare resource use was estimated based on NHS reference costs, and included resource use associated with biological and conventional treatments (including initial and follow‑up consultations and blood tests) and resource use as a function of health state (including hospitalisations, surgical interventions and concomitant medications). The acquisition costs for TNF‑alpha inhibitors took into account the patient access scheme for golimumab. Administration costs were included for intravenous infliximab only, because all other biological drugs were assumed to be given by subcutaneous injection at no cost to the NHS.

3.12 A number of iterations of the economic model were presented by the company: the first in its original submission (referred to in this document as the 'original' model), the second corrected after the clarification stage (referred to in this document as the 'post‑clarification' model), and the third corrected during consultation (referred to in this document as the 'post‑consultation' model). A further iteration, incorporating the patient access scheme and 1 amended assumption, was presented by the company in its submission for the rapid review of NICE technology appraisal guidance 313 (referred to in this document as the 'rapid review' model; see section 3.30). The original model was used to develop base‑case, deterministic and probabilistic sensitivity analyses, and a series of scenario analyses; the results of the base‑case and scenario analyses were replaced by the post‑clarification and post‑consultation models and so are not reported here. The post‑clarification model incorporated corrections requested by the ERG during clarification, including amendments to the probability distributions for some variables and to the costs associated with psoriatic arthritis. The company used this model to develop an updated base case and updated scenario analyses. The post‑consultation model submitted during consultation corrected an error, identified by the company and a consultee during consultation, affecting the acquisition cost of golimumab 100 mg. The cost‑effectiveness evidence presented here for the TNF‑alpha inhibitor‑naive population is based on the results of the post‑consultation model, which replaced the previous models. The cost‑effectiveness evidence presented here for the TNF‑alpha inhibitor‑exposed population is based on the post‑clarification model (because the TNF‑alpha inhibitors were not included as comparators in this population, and therefore the golimumab error did not apply).

3.13 In the TNF‑alpha inhibitor‑naive population (post‑consultation model, probabilistic results), ustekinumab was associated with total costs of £70,249 and a total of 6.23 quality‑adjusted life years (QALYs). Adalimumab was associated with costs of £64,487 and 6.42 QALYs, and therefore dominated ustekinumab (that is, adalimumab was more effective and less expensive). Adalimumab, in turn, was associated with an additional £31,476 in costs and 1.76 QALYs compared with conventional management, giving an incremental cost‑effectiveness ratio (ICER) of £17,868 per QALY gained. The company also presented pairwise comparisons between ustekinumab and conventional management (post‑consultation model, probabilistic results): ustekinumab provided 1.57 additional QALYs compared with conventional management, at an additional cost of £37,239, giving an ICER of £23,723 per QALY gained. The deterministic sensitivity analyses (original model) showed that the results were most sensitive to the change in HAQ‑DI score over time associated with the natural history of psoriatic arthritis, the proportion of people who had a PsARC response and the HAQ‑DI change associated with PsARC response.

3.14 In the TNF‑alpha inhibitor‑exposed population (post‑clarification model, probabilistic results), ustekinumab provided an additional 1.41 QALYs compared with conventional management, at an additional cost of £41,199, to give an ICER of £29,132 per QALY gained compared with conventional management. The probabilistic sensitivity analysis (original model) indicated there was a 0% and 67% probability of ustekinumab being cost effective compared with conventional management if the maximum acceptable ICERs were £20,000 and £30,000 per QALY gained respectively. The deterministic sensitivity analyses (original model) showed that the results were most sensitive to the HAQ‑DI score change with the natural history of psoriatic arthritis and the HAQ‑DI change associated with a PsARC response.

3.15 The company presented a series of scenario analyses for both the TNF‑alpha inhibitor‑naive and ‑exposed populations (post‑clarification model). These explored structural assumptions in the model around the treatment continuation rule (timing and criteria), progression of psoriatic arthritis, and utility and clinical effectiveness estimates. In the TNF‑alpha inhibitor‑naive population, all scenario analyses showed that ustekinumab was more expensive and less effective than adalimumab. Ustekinumab was associated with probabilistic ICERs compared with conventional management ranging from £21,628 to £31,469 per QALY gained. In the scenario in which treatment response was assessed at week 24 for all treatments, ustekinumab was associated with additional costs of £38,222 and an additional 1.28 QALYs compared with conventional management, giving a probabilistic ICER of £29,808 per QALY gained for ustekinumab compared with conventional management. In scenario analyses for the TNF‑alpha inhibitor‑exposed population, ustekinumab was associated with probabilistic ICERs compared with conventional management ranging from £27,606 to £40,019 per QALY gained. In the scenario in which treatment response was assessed at week 24 for all treatments, ustekinumab was associated with additional costs of £43,064 and an additional 1.12 QALYs compared with conventional management, giving a probabilistic ICER of £38,516 per QALY gained for ustekinumab compared with conventional management.

Evidence Review Group's critique and exploratory analyses of the company's submission

3.16 The ERG carried out exploratory analyses to test whether the clinical effectiveness of ustekinumab was influenced by prior TNF‑alpha inhibitor treatment or ustekinumab dose. It stated that there is no convincing evidence of a substantial difference in the effectiveness of ustekinumab between people who have and people who have not previously had TNF‑alpha inhibitors, and that treatment effects were not statistically significantly different between ustekinumab doses.

3.17 The ERG identified a number of limitations in the evidence available from the PSUMMIT studies. The switch from placebo to ustekinumab at weeks 16 and 24 provides a short‑term comparison for a chronic condition such as psoriatic arthritis. Analyses of TNF‑alpha inhibitor‑exposed patients included only the 180 patients who had previously had varying numbers of TNF‑alpha inhibitors for varying durations. The ERG emphasised that the analyses of PSUMMIT 2 did not distinguish between people who had previously had 1, 2, 3 or more TNF‑alpha inhibitors, and so did not differentiate between people who had tried only some of the available TNF‑alpha inhibitors and people for whom TNF‑alpha inhibitors as a class had failed. The ERG considered that the data on patients whose disease was truly TNF‑alpha inhibitor refractory were scarce. It was also noted that for many of the secondary outcomes (DLQI, SF‑36 and radiographic scores), baseline scores were not reported, making interpretation of the results difficult.

3.18 The ERG considered that, despite some heterogeneity between trials, the mixed treatment comparison was appropriate to carry out and the results were robust. It did not consider that the weight‑based dosing subgroup matched the marketing authorisation, and noted that this led to exclusion of a large amount of data. However, the ERG noted that an additional analysis including all patients from PSUMMIT 1 and 2 provided fairly similar results to the weight‑based analysis. The ERG noted that overall, the mixed treatment comparison found that ustekinumab had the lowest or one of the lowest response rates for PASI 75, PASI 90 and PsARC.

3.19 The ERG noted that the company's economic model was similar to those used in previous NICE technology appraisal guidance on etanercept, infliximab and adalimumab for the treatment of psoriatic arthritis and golimumab for the treatment of psoriatic arthritis, although it had a longer time horizon (52 years compared with 40 years). The ERG considered many of the key assumptions used in the model to be broadly acceptable, including change in PASI score with biological treatment, rebound effect on treatment withdrawal, withdrawal rates in the TNF‑alpha inhibitor‑naive population, exclusion of disutilities and costs for adverse events, the equation used to map HAQ‑DI and PASI to EQ‑5D, resource use, drug and health state costs, and the approach to deterministic sensitivity analysis. The ERG cautioned that the results of the model should be interpreted with care, specifically the pairwise ICERs for ustekinumab compared with conventional management in the TNF‑alpha inhibitor‑naive population; it considered that these ICERs represent a scenario in which ustekinumab is the only alternative to conventional management, which is unrealistic.

3.20 The ERG highlighted weaknesses in the clinical effectiveness parameters used in the model. It noted that the company used a mixture of results from the mixed treatment comparison for the effectiveness of TNF‑alpha inhibitors and PSUMMIT results for the effectiveness of ustekinumab to obtain HAQ‑DI score changes, and considered that there were limitations to this approach. In addition, PsARC response rates for ustekinumab based on the weight‑based dosing subgroup resulted in a loss of data. The ERG considered both of these issues in exploratory analyses (see sections 3.24 and 3.25). Furthermore, the ERG queried the model assumption that people having conventional management did not have any improvement in PASI, whereas in clinical practice skin symptoms often respond well to DMARDs.

3.21 The ERG noted the simplifying assumption in the model that people switched to conventional management after failure of the intervention being analysed, and did not have subsequent biological therapies. Thus, the costs and benefits associated with subsequent lines of biological treatment were not included in the model. The ERG stated that in clinical practice in the UK, most people whose disease has failed to respond to 1 TNF‑alpha inhibitor would be considered for subsequent‑line TNF‑alpha inhibitor treatment.

3.22 The ERG emphasised the uncertainties about the natural history progression of psoriatic arthritis scores during conventional management. This is a key driver of the model. The assumptions underlying the gradual increase in HAQ‑DI scores during conventional management were consistent with previous submissions. However, the ERG noted that the estimate for the rate of progression was prepared from limited data in 2009 but not updated. It also queried whether the assumptions about rebound and progression of arthritis symptoms taken from models of TNF‑alpha inhibitors were applicable to ustekinumab, given its different mechanism of action.

3.23 The ERG highlighted that the TNF‑alpha inhibitor‑exposed population has not been considered in previous appraisals, and noted some uncertainties in the model for this population. By comparing ustekinumab with conventional management, the company made no distinction between people whose disease had not responded to 1, 2, 3 or more TNF‑alpha inhibitors. That is, it did not differentiate between people who had tried only some of the available TNF‑alpha inhibitors and people for whom TNF‑alpha inhibitors as a class had failed. The ERG stated that, in clinical practice in the UK, most people whose disease has failed to respond to 1 TNF‑alpha inhibitor would be considered for subsequent‑line TNF‑alpha inhibitor treatment. The ERG therefore considered the company's model to have severe limitations, noting that ustekinumab should be compared with other TNF‑alpha inhibitors in the TNF‑alpha inhibitor‑exposed population. The ERG further noted that much of the evidence informing the model was drawn from people who had not had prior TNF‑alpha inhibitor therapy. In particular, estimates for the natural history progression of HAQ‑DI (a key driver of the model), mortality rates and treatment withdrawal rates were based on TNF‑alpha inhibitor‑naive populations. It queried whether these assumptions were applicable to the TNF‑alpha inhibitor‑exposed population.

3.24 The ERG carried out exploratory analyses in both the TNF‑alpha inhibitor‑naive and TNF‑alpha inhibitor‑exposed populations. These explored the sensitivity of the company's model to assumptions about weight‑based dosing, HAQ‑DI rebound and natural history progression, the time horizon, the timing of treatment response assessment, and the inclusion of phototherapy. In the TNF‑alpha inhibitor‑naive population, the ERG's exploratory analyses showed that, in the incremental analysis (comparing ustekinumab, TNF‑alpha inhibitors and conventional management), ustekinumab remained dominated in all modelled scenarios. Probabilistic ICERs for ustekinumab compared with conventional management ranged from £22,455 to £55,029 per QALY gained. In particular, the ERG's analyses showed that assessing the response to treatment at week 24 for both ustekinumab and conventional management increased the ICER by £6987 per QALY gained, compared with the base case. The ERG presented a preferred base case for the TNF‑alpha inhibitor‑naive population, based on what it considered to be the most plausible assumptions. This consisted of the company's post‑clarification model (see section 3.12), with additional corrections added by the ERG (including amendments to the health state costs, probability distributions and baseline PASI and HAQ‑DI scores), applying the weight‑based dosing assumption to ustekinumab 90 mg only, and using HAQ‑DI scores drawn from an update of the mixed treatment comparison developed by the ERG for the NICE technology appraisal guidance on golimumab for the treatment of psoriatic arthritis. In this analysis, ustekinumab was dominated by adalimumab. Compared with conventional management, ustekinumab was associated with additional costs of £37,123 and an additional 1.6 QALYs, giving a probabilistic ICER compared with conventional management of £23,246 per QALY gained.

3.25 The ERG also presented exploratory analyses in the TNF‑alpha inhibitor‑exposed population. In these exploratory analyses, ustekinumab was associated with probabilistic ICERs compared with conventional management ranging from £28,670 to £69,139 per QALY gained. The ERG carried out an exploratory sequencing analysis for the TNF‑alpha inhibitor‑exposed population, to examine the cost effectiveness of ustekinumab compared with TNF‑alpha inhibitors, when used as second‑line treatments after failure of first‑line TNF‑alpha inhibitors. The ERG presented 3 scenarios for the sequencing analysis: 2 in which first‑line treatments failed because of lack of effectiveness (the first based on evidence from PSUMMIT 2, and the second based on the ERG's estimates) and a third in which first‑line treatments failed because of adverse events. In the first scenario, ustekinumab was associated with ICERs of £32,818 per QALY gained (compared with adalimumab, when etanercept was used as first‑line treatment) to £37,738 per QALY gained (compared with etanercept, when golimumab or adalimumab were used as first‑line treatment), and in the other 2 scenarios it was dominated by other treatments. However, the ERG stressed that this exploratory analysis was based on numerous assumptions and was subject to considerable uncertainty. The ERG did not present a preferred base case for the TNF‑alpha inhibitor‑exposed population because of the uncertainty remaining in the model.

Company's additional analyses provided during consultation and Evidence Review Group's critique

3.26 In response to consultation, the company submitted additional evidence exploring the cost effectiveness of the 45‑mg dose of ustekinumab alone, based on the post‑consultation model. The company noted that ustekinumab 45 mg could potentially be considered for all patients. In the base‑case analysis for the TNF‑alpha inhibitor‑naive population, ustekinumab 45 mg was the lowest‑cost and least effective biological treatment. In the TNF‑alpha inhibitor‑exposed population (base case, probabilistic analysis), ustekinumab 45 mg was associated with an ICER of £21,789 per QALY gained compared with conventional management. The company also reproduced the ERG's exploratory sequencing analysis and presented scenario analyses to explore the impact of key assumptions.

3.27 The ERG submitted a critique of these analyses. It noted that, in principle, the scenario in which all patients have ustekinumab 45 mg is reasonable to explore, but highlighted that there is uncertainty about the validity of this scenario in clinical practice. The ERG applied the 45‑mg dosing assumption to its preferred base case for the TNF‑alpha inhibitor‑naive population (see section 3.24), and noted that the results were generally similar to those presented by the company. For the TNF‑alpha inhibitor‑exposed population, the ERG confirmed that the company had correctly reproduced its exploratory sequencing analysis, although it emphasised that this analysis was highly uncertain because it was based on numerous assumptions.

Further evidence

3.28 Based on a comment received during consultation from a company that manufactures a comparator drug, further evidence was identified by the technical team that provided long‑term analyses of the change from baseline in HAQ‑DI and radiographic scores with ustekinumab compared with placebo (presented in 2 abstracts and a press release: Kavanaugh et al. 2013, McInnes et al. 2013 and Johnson and Johnson 2013). In a pre‑specified pooled analysis of the radiographic scores in PSUMMIT 1 and 2, the mean changes from baseline to week 24 were 0.40, 0.39 and 0.97 (ustekinumab 45 mg, ustekinumab 90 mg and placebo respectively). The mean changes from baseline to week 52 were 0.58, 0.65 and 1.15 (ustekinumab 45 mg, ustekinumab 90 mg and patients randomised to placebo respectively), showing that ustekinumab inhibited radiographic progression compared with placebo and that this inhibition continued to week 52. Data from PSUMMIT 1 alone were consistent with the pooled analysis. However, in PSUMMIT 2 alone there was no statistically significant difference in radiographic progression between ustekinumab and placebo; the company noted that the studies were not individually powered to detect differences in radiographic progression. A long‑term analysis of HAQ‑DI scores in PSUMMIT 1 showed that the mean changes from baseline to 52 weeks were −0.34, −0.43 and −0.37 in patients randomised to ustekinumab 45 mg, ustekinumab 90 mg and placebo respectively, and the mean changes from baseline at 100 weeks were −0.36, −0.45 and −0.36 (ustekinumab 45 mg, ustekinumab 90 mg and patients randomised to placebo respectively). Long‑term analyses of HAQ‑DI scores in PSUMMIT 2 were not available at the time NICE technology appraisal guidance 313 was prepared.

Rapid review of NICE technology appraisal guidance 313: patient access scheme

3.29 In NICE technology appraisal guidance 313, ustekinumab was not recommended for treating active psoriatic arthritis. After publication, the company agreed a patient access scheme with the Department of Health (see section 2.3) and submitted revised analyses to be considered in a rapid review of the guidance. Under the original patient access scheme the company provided 2x45 mg pre-filled syringes, for patients who needed the higher dose of 90 mg, at the same total cost to the NHS as for a single 45‑mg pre-filled syringe. The patient access scheme was withdrawn in January 2017 because the company now provides a 90‑mg vial at the same cost as the 45‑mg vial.

3.30 The company submitted a revised economic analysis (the 'rapid review' model) based on its post‑consultation model, incorporating the patient access scheme and an altered assumption about the effect of conventional management on skin symptoms (based on the Committee's considerations in NICE technology appraisal guidance 313). It presented analyses for both the TNF‑alpha inhibitor‑naive and ‑exposed populations. For the TNF‑alpha inhibitor‑exposed population, the company also presented a sequencing analysis. This analysis was developed from the ERG's exploratory sequencing analysis (see section 3.25) and used the scenario in which the first TNF‑alpha inhibitor failed because of lack of efficacy and clinical effectiveness data were taken from the PSUMMIT 2 study. The company considered that including the patient access scheme considerably improved the cost effectiveness of ustekinumab.

3.31 The patient access scheme was incorporated by reducing the unit cost of ustekinumab 90 mg to £2147. The company estimated the additional costs associated with the patient access scheme to be £33 per patient. It considered that these costs were very small and so would not affect the appraisal; therefore, it did not include them in the economic analyses.

3.32 The company modelled the effect of conventional management on skin symptoms in the same way as it had modelled the effects of biological drugs in the original model – that is, assuming a fixed improvement in PASI score based on PASI response. As part of this change, the company also amended the rebound assumption for people who withdraw from biological therapy, such that the PASI score rebounded to a score based on the effect of conventional management (rather than the baseline score). The PASI scores and response rates for conventional management were taken from the placebo arms of the company's mixed treatment comparison and the PSUMMIT 1 and 2 studies.

3.33 In the company's base case for the TNF‑alpha inhibitor‑naive population (rapid review model, probabilistic results, incremental analysis), conventional management was the lowest cost option, followed by ustekinumab then adalimumab. Ustekinumab was therefore the least costly biological drug, and was associated with total costs of £59,105, a total of 6.09 QALYs and an ICER compared with conventional management of £23,164 per QALY gained. Adalimumab had a pairwise ICER compared with conventional management of £21,765 per QALY gained.

3.34 In the company's base case for the TNF‑alpha inhibitor‑exposed population (rapid review model, probabilistic results), ustekinumab was associated with total costs of £62,724 and a total of 4.08 QALYs. It was associated with an ICER of £25,675 per QALY gained, compared with conventional management. In the sequencing analysis, ustekinumab was associated with deterministic ICERs ranging from £21,241 per QALY gained (compared with etanercept, when adalimumab, golimumab or infliximab are used first line) to £25,921 per QALY gained (compared with conventional management, when etanercept is used first line).

3.35 For both the TNF‑alpha inhibitor‑naive and exposed populations, the company presented a deterministic sensitivity analysis and scenario analyses consistent with those it presented in the original submission. In both populations, the results were most sensitive to the change in HAQ‑DI score over time associated with the natural history of psoriatic arthritis. In scenarios based on the TNF‑alpha inhibitor‑naive population, ustekinumab was associated with deterministic ICERs compared with conventional management of £21,411 to £29,580 per QALY gained. In equivalent scenarios based on the TNF‑alpha inhibitor‑exposed population, ustekinumab was associated with deterministic ICERs compared with conventional management of £23,229 to £33,578 per QALY gained. In scenarios in which the response to all treatments was assessed at week 24, ustekinumab was associated with ICERs of £27,914 per QALY gained (TNF‑alpha inhibitor‑naive population) and £32,608 per QALY gained (TNF‑alpha inhibitor‑exposed population), compared with conventional management.

Evidence Review Group critique of the company's rapid review submission

3.36 The ERG noted that the company had appropriately incorporated the patient access scheme into the latest version of the economic model from NICE technology appraisal guidance 313 (the post‑consultation model). It agreed with the company that the additional costs associated with the patient access scheme did not significantly alter the cost effectiveness of ustekinumab.

3.37 The ERG highlighted that, in the company's model for the TNF‑alpha inhibitor‑naive population, ustekinumab was extendedly dominated in all scenarios. An intervention is 'extendedly dominated' when it is more costly and less effective than a combination of 2 comparators; that is, the ICER for the intervention is higher than that of the next more effective comparator when both are compared with another less effective comparator. In the base case, ustekinumab was extendedly dominated by adalimumab and conventional management, because the ICER for ustekinumab compared with conventional management was higher than that of adalimumab compared with conventional management.

3.38 The ERG commented on the company's inclusion of the effect of conventional management on skin symptoms. It considered that the company's approach was mostly reasonable. However, the ERG highlighted that the PASI score to which people were assumed to rebound when they stop biological treatment was not the same as the average PASI score for people having conventional management. It noted that this difference resulted from differences in PASI response rates between week 12 (as applied to the conventional management arm) and week 24 (as applied to the biological therapy arms). The ERG commented that the effect of this assumption differed between the TNF‑alpha inhibitor‑naive and TNF‑alpha inhibitor‑exposed populations.

3.39 The ERG identified a number of errors in the company's economic model, relating to disease‑related costs, medical resource use, the accrual of QALYs in the second year of the model and the application of discounting. It noted that the errors tended to underestimate the costs and QALYs associated with psoriatic arthritis, and hence tended to underestimate the cost effectiveness of more effective treatments. Consequently, the ERG noted that correcting these errors caused the cost effectiveness of ustekinumab to improve relative to conventional management, but worsen relative to TNF‑alpha inhibitors.

3.40 The ERG presented a scenario analysis to explore the patient access scheme combined with the Committee's preferred assumptions from NICE technology appraisal guidance 313. This analysis was developed from the company's rapid review model with the errors corrected, a 40‑year time horizon and with the response to both ustekinumab and conventional management assessed at week 24. In this scenario (probabilistic analysis), in the TNF‑alpha inhibitor‑naive population ustekinumab remained extendedly dominated (by a combination of conventional management and adalimumab) and had an ICER of £21,857 per QALY gained compared with conventional management. In the TNF‑alpha inhibitor‑exposed population (probabilistic analysis), the ICER for ustekinumab was £25,292 per QALY gained, compared with conventional management. In the sequencing analysis based on this scenario, ustekinumab was associated with an ICER of £25,393 per QALY gained compared with conventional management, when golimumab, adalimumab or etanercept are used first line. The ERG noted that the time horizon had a small effect on the ICER, whereas the effect of the week‑24 assessment time point was larger.

3.41 Full details of all the evidence are available.