Etanercept, infliximab and adalimumab for the treatment of psoriatic arthritis: a systematic review and economic evaluation

Efficacy after 12 weeks’ treatment

The individual trial results (Table 3) and pooled estimates of effect (Table 4) demonstrate a statistically significant benefit of etanercept for all joint disease and HAQ score outcomes. There was no statistical heterogeneity for any outcome.

Across the two trials at 12 weeks almost 85% of patients treated with etanercept achieved a PsARC response, which is the only joint disease outcome measure that has been specifically defined for PsA. In addition, around 65% of patients treated with etanercept achieved an ACR 20 response, demonstrating a basic degree of efficacy in terms of arthritis-related symptoms. Around 45% of patients treated with etanercept achieved an ACR 50 response, and around 12% achieved an ACR 70 response, demonstrating a good level of efficacy. The subgroup analyses conducted on the Mease et al. 52,97,99,105,107,110 data revealed that the effect of etanercept was not dependent on patients’ concomitant use, or not, of MTX. The PASI results from Mease et al. 78 indicate some beneficial effect on psoriasis at 12 weeks; however, the data were too sparse (38 patients in total) to establish statistical significance. The statistically significant reduction in HAQ score with etanercept compared with placebo indicates a beneficial effect of etanercept on functional status.

Efficacy after 24 weeks’ treatment

At 24 weeks, the treatment effect for all joint disease outcome measures was statistically significantly greater with etanercept than with placebo, although these data were available only for one trial (see Table 3). As at 12 weeks, the subgroup analyses conducted on the Mease et al. 52,97,99,105,107,110 data revealed that the effect of etanercept was not dependent upon patients’ concomitant use, or not, of MTX. The size of treatment effect did not appear greater at 24 weeks than at 12 weeks.

At 24 weeks, the TSS annualised rate of progression was statistically significantly lower in patients treated with etanercept than in patients receiving placebo. This treatment difference did not vary with or without concomitant MTX use. However, this duration of follow-up is to be considered short and barely adequate for this outcome.

At 24 weeks, the treatment effect on psoriasis favoured etanercept with relative risks (RRs) for PASI 75 of 7.05 [95% confidence interval (CI) 1.68 to 29.56], PASI 50 of 2.65 (95% CI 1.46 to 4.80) and PASI 90 of 1.88 (95% CI 0.36 to 9.90). The result for PASI 75 and PASI 50 was statistically significant despite there being only 66 patients on etanercept evaluable for psoriasis. 52,97,99,105,107,110

Longer-term follow-up

The results for long-term follow-up are summarised in Table 5. The data are uncontrolled and therefore cannot be taken as reliable. In general, they do indicate that the improvements in patients’ joint and skin symptoms and HAQ score achieved during the controlled phase of the trials are maintained in the medium term. At 1 year, the mean annualised rate of progression TSS for all patients was –0.03 [standard deviation (SD 0.87)] indicating that on average no clinically significant progression of joint erosion had occurred. Limited 2-year data indicated little change in mean TSS, although data on patient numbers or variability were not reported.

Summary of the efficacy of etanercept in the treatment of psoriatic arthritis

• There is evidence from double-blind placebo-controlled trials of a good level efficacy for etanercept in the treatment of PsA. Conclusions to be drawn from these data are limited by the small sample size and the short duration of one of the trials.

• There is evidence from two RCTs that etanercept treatment improves patients’ functional status as assessed using the HAQ score.

• There is limited evidence from the two RCTs that etanercept treatment has a beneficial effect on the psoriasis component of the disease.

• Uncontrolled follow-up of patients indicate that treatment benefit is maintained for at least 50 weeks; however, these data may not be reliable.

• There are radiographic data from controlled trials for etanercept in PsA that demonstrate a beneficial effect on progression of joint disease at 24 weeks. This is a very short time over which to identify a statistically significant effect of therapy and indicates a rapid onset of action of etanercept. Data from uncontrolled follow-up indicate that, on average, disease progression may be halted for at least 1 year; however, these data may not be reliable.

Efficacy after 14–16 weeks’ treatment

At 14 weeks, both trials reported a significant improvement in the PsA-specific PsARC measure for patients receiving infliximab, relative to those receiving placebo (pooled RR 3.44, 95% CI 2.53 to 4.69 – Table 8). There was evidence of statistical heterogeneity (I² = 68%) between the two study estimates, due to the different placebo response rates (13.5% vs 27%). PsARC response on infliximab was around 77% in both trials.

The pooled RR for ACR 20 at 14 weeks was 5.47 (95% CI 3.43 to 8.71), with an overall response of 61% in infliximab-treated patients, demonstrating a clear degree of efficacy of infliximab in terms of arthritis-related symptoms. As very few patients receiving placebo achieved an ACR 50 or ACR 70 response, the pooled RRs clearly favoured infliximab in terms of these outcomes, although the limited number of observations mean that there is considerable uncertainty around these pooled estimates, as reflected by their CIs (see Table 8). Despite the potentially large relative effects, it should also be noted that only the minority of infliximab-treated patients achieved an ACR 50 or ACR 70 response at 14 weeks (36% and 17%, respectively). Data from the IMPACT trial79–81,89,96,109,111,113–115,117,118 indicated no significant difference in ACR 20 response at 16 weeks between patients with and without concomitant MTX, although the number of patients in each of these groups was small.

As with the ACR outcomes, few patients receiving placebo demonstrated skin improvements over 14–16 weeks in terms of a PASI response; the pooled RR for PASI 50 was 10.58 (95% CI 5.47 to 20.48), demonstrating a clear degree of efficacy of infliximab in terms of skin-related symptoms. PASI 75 and PASI 90 response measures favoured infliximab even more strongly, although it should be noted that PASI outcomes were recorded only for those patients with a score of at least 2.5 at baseline. Forty-two per cent of patients receiving infliximab achieved the highest level of skin response (PASI 90), although again there is considerable uncertainty around the estimates (see Table 7).

The statistically significant pooled percentage change from baseline in HAQ score with infliximab compared with placebo [mean difference –60.37 (–75.28 to –45.46)] indicates a beneficial effect of infliximab on functional status.

Efficacy after 24 weeks

The IMPACT 2 trial82,90,91,95,98,106,112,116 maintained randomisation for 24 weeks. The data for all measures of joint disease, psoriasis and HAQ are similar to those observed at the earlier 14-week follow-up, suggesting that the benefits of infliximab are maintained up to 24 weeks of treatment (see Table 7). Change from baseline in modified van der Heijde–Sharp score significantly differed between infliximab and placebo groups, indicating that infliximab may inhibit progression of joint damage at 24 weeks (see Table 7).

Longer-term follow-up

The data for longer-term follow-up (50 or 54 weeks) from the two IMPACT trials are summarised in Table 9. These data are uncontrolled and may therefore be unreliable. Also, the duration of treatment varied between participants, as some will have crossed over from placebo treatment. However, the data broadly indicate that the levels of efficacy achieved with infliximab in terms of joint disease, psoriasis and HAQ after 14–24 weeks’ treatment might be maintained in the medium term.

In terms of radiographic assessment, there was no significant change from baseline in the total modified van der Heijde–Sharp score for those infliximab-treated patients followed up at 50 or 54 weeks in the two studies, suggesting infliximab may inhibit progression of joint damage. However, as with other post-24-week outcomes, there was no placebo group for comparison.

Summary of the efficacy of infliximab in the treatment of PsA

• There is evidence from two double-blind placebo controlled trials of a good level of efficacy for infliximab in the treatment of PsA, with beneficial effects on joint disease, psoriasis and functional status as assessed by HAQ.

• Conclusions to be drawn from these data are limited by the short duration of the controlled trials; controlled data to evaluate long-term effects are not available.

• Uncontrolled follow-up of patients indicate that short-term benefit is maintained for at least 50 weeks; however, these data may not be reliable.

• Radiographic data from uncontrolled follow-up of infliximab trials suggest that the drug may delay the progression of joint disease in PsA, although these data are not of high quality.

Efficacy after 12 weeks’ treatment

At 12 weeks, both trials reported a significant improvement in the PsA-specific PsARC measure for adalimumab relative to placebo (pooled RR 2.24; 95% CI 1.74 to 2.88), with an overall response rate of around 59% for adalimumab. The pooled RR for ACR 20 at 12 weeks was 3.65 (95% CI 2.57 to 5.17), demonstrating a clear degree of efficacy of adalimumab in terms of arthritis-related symptoms. There was no statistically significant heterogeneity between any of the pooled outcomes. The pooled RRs for ACR 50 and ACR 70 also clearly favoured adalimumab, although as with other estimates of these outcomes their related CIs were wide (Table 12). Again, the large relative differences on these higher-response thresholds reflect some response with biologic therapy versus virtually none with placebo (e.g. 18% vs 0.5% for ACR 70). Data from the larger trial indicated little evidence of any differential ACR response at 12 weeks between patients with and without concomitant MTX. 51,88,92,93,100–104

Only one trial reported 12-week PASI response measures: in patients with psoriasis of at least 3% BSA at baseline. 51,88,92,93,100–104 Response was significantly greater for adalimumab than placebo at all three PASI thresholds (PASI 50/75/90 – see Table 11). As with the ACR outcomes, there was little evidence of any differential PASI response between patients receiving and not receiving concomitant MTX, although the number of patients in each subgroup was small.

The statistically significant pooled absolute mean change from baseline in HAQ score with adalimumab compared with placebo (mean difference –0.27; 95% CI –0.36 to –0.18) indicates a beneficial effect of adalimumab on functional status.

Efficacy after 24 weeks’ treatment

The ADEPT trial51,88,92,93,100–104 maintained randomisation for 24 weeks. The data for all measures of joint disease, psoriasis and HAQ were all similar to those observed at the earlier 14-week follow-up, suggesting that the benefits of adalimumab are maintained for up to 24 weeks of treatment (see Table 12).

In addition, this trial51,88,92,93,100–104 reported a statistically significant difference in mean change in TSS score from baseline (–0.2 vs 0.1, p < 0.001), favouring adalimumab over placebo in terms of delayed progression of joint disease. However, this duration of follow-up is to be considered short and barely adequate for this outcome.

The smaller of the two trials allowed patients to enter an open-label follow-up period from weeks 12–24. 83 The pattern of reported joint disease outcomes appears similar to those reported at the end of the 12-week randomised period; however, estimates based on these non-randomised data cannot be considered reliable.

Longer-term follow-up

The larger adalimumab trial followed patients in an open-label fashion, measuring several outcomes at 48 weeks and at 2 years (Table 13). 51,88,92,93,100–104 Both ACR response rates and mean HAQ scores at weeks 48 and 104 appeared to have remained stable relative to the randomised observations of these outcomes at weeks 12 and 24. Similarly, rates of PASI response reported at 48 weeks appeared largely consistent with the earlier randomised observations. Disease progression as measured by TSS was reported at weeks 48 and 144, with higher mean values than observed at 24 weeks, although the open-label observational nature of these open-label data makes it difficult to reliably determine any clear changes in TSS over time.

Summary of the efficacy of adalimumab in the treatment of psoriatic arthritis

• There is evidence from two double-blind placebo-controlled trials of a good level efficacy for adalimumab in the treatment of PsA, with beneficial effects on joint disease and functional status as assessed by HAQ.

• There is limited evidence from a single RCT that adalimumab treatment has a beneficial effect on the psoriasis component of the disease in patients with PsA, as measured by PASI.

• Conclusions to be drawn from these data are limited by the short duration the controlled trials; large-scale controlled data to evaluate long-term effects are not available.

• Uncontrolled follow-up of patients indicate that treatment benefits in terms of joint disease and HAQ measures may be maintained at up to 2 years; however, these data may not be reliable.

• Radiographic data from a single controlled trial for adalimumab in PsA demonstrate a beneficial effect on progression of joint disease at 24 weeks. This is a very short time over which to identify a statistically significant effect of therapy and indicates a rapid onset of action of adalimumab. Data from uncontrolled follow-up are inadequate to determine whether any potential delay in disease progression persists at 1–2 years’ follow-up.

Psoriatic Arthritis Response Criteria response

The results of the evidence synthesis for PsARC response are in the form of probability of response (Table 14). The mean probability of a PsARC response was estimated to be 71% for etanercept, 79% for infliximab and 59% for adalimumab, compared with 25% for placebo. While the credible intervals for all three biologics overlap each other, none overlap placebo.

Changes in Health Assessment Questionnaire

The results of the evidence synthesis of HAQ conditional on response are presented as absolute changes in HAQ. These are calculated separately for the patients achieving a PsARC response (Table 15) and those who did not achieve a PsARC response (Table 16).

Statistically significant reductions in mean HAQ score were achieved with all four treatments compared, i.e. the credible intervals did not include zero. However, patients who responded to placebo achieved an improvement in the HAQ score of –0.244, which is below the minimum clinically significant threshold for PsA of –0.3. 150 Patients who responded to etanercept and infliximab achieved similar mean changes in HAQ (–0.630 and –0.657, respectively), whereas responders to adalimumab achieved a lower mean change in the HAQ score of –0.477, although credible intervals overlap those of the other two treatments.

For all three biologics the changes in HAQ for those patients who did not respond to treatment were below the minimum clinically significant threshold. Placebo non-responders were used as a baseline in the synthesis.

Psoriasis Area and Severity Index

The results of the evidence synthesis for a PASI response are in the form of probability of response (Table 17). The mean probability of a PASI 75 response was estimated to be 18% for etanercept, 77% for infliximab and 48% for adalimumab, compared with 4% for placebo. The credible intervals for infliximab and etanercept do not overlap each other and none for the biologics overlap placebo.

American College of Rheumatology model

The results of the evidence synthesis for a ACR response are in the form of probability of response (Table 18). The ACR 20 is generally accepted to be the minimal clinically important difference that indicates some response to a particular intervention in terms of arthritis-related symptoms. The mean probability of an ACR 20 response was estimated to be 61% for etanercept, 68% for infliximab and 56% for adalimumab, compared with 14% for placebo. The credible intervals for all three biologics overlap each other, but none overlap those for placebo.

Summary of evidence synthesis results

Across all outcomes – PsARC, ACR and PASI – infliximab is associated with the highest probability of response. The response in joint disease (PsARC and ACR) is greater with etanercept than with adalimumab, whereas the response in skin disease (PASI) is greater with adalimumab than with etanercept, although these differences are not statistically significant. In those patients who achieve a PsARC response to treatment the highest mean reductions in HAQ are seen with infliximab and etanercept.

Comparison of evidence synthesis results

Each of the three company submissions combined evidence derived using Bayesian evidence synthesis methods. A brief comparison of these methods and the methods used by the assessment team are presented in Table 19 and discussed below.

Two of the company submissions – Abbott151 and Schering-Plough152 – conducted evidence syntheses to derive estimates that would allow the relative efficacy of the drugs to be compared (Table 19). (Since the production of this report, Schering-Plough has merged with Merck.) Wyeth153 chose not to conduct this synthesis themselves but to use the results of a previously published single technology appraisal (STA) relating to Abbott’s adalimumab. 74

Full details of the evidence synthesis model used by Wyeth153 were not provided in the Wyeth submission. 153 Further, the methodology of the evidence synthesis from which these results were obtained was not presented in the original report. 151 The synthesis was conducted by Abbott151 on the request from the Evidence Review Group (ERG) and only the results were presented in the ERG report. For this reason no summary/critique of the methods can be presented. The following section gives a comparative overview of the evidence synthesis results obtained by Schering-Plough,152 Abbott151 and by the Assessment Group in this report.

Psoriatic Arthritis Response Criteria response For PsARC response, all of the evidence synthesis models used a fixed-effects meta-analysis to synthesise the evidence. Both the Assessment Group and Schering-Plough152 identified and included six RCTs in their synthesis. Abbott,151 with slightly broader inclusion criteria, identified and included 10 RCTs. Abbott151 included RCTs in which the drug golimumab was administrated to the comparator arm of the RCT and, although no results were presented for this comparator, the other estimates do ‘borrow strength’ from these data. Although including the same six RCTs, both the Assessment Group and Schering-Plough152 estimated PsARC response using slightly different data. The Assessment Group used the closest follow-up outcome to 12 weeks, whereas Schering-Plough152 used the latest available end points (Table 20). This meant that, with the exception of the adalimumab data, the data inputs were principally the same. Abbott151 took a more complex bivariate approach, which enabled them to model the joint distribution of ACR/PsARC response at 12 weeks. Taking a bivariate approach allows the correlation between outcomes, if present, to be accounted for. However, if the correlation is zero then any bivariate joint modelling will arrive at the same estimates as two independent models. Given the lack of transparency of the Abbott151 evidence synthesis, it was not possible to unpick and decipher the subtleties of their model. The Assessment Group, following clinical advice, have used PsARC at 12 weeks to determine response to treatment. This follows clinical practice.

As can be seen from the results presented for the probability of response to the biologics under appraisal (and placebo) (Table 21), all of the mean estimates obtained were very similar, despite the different modelling assumptions and evidence used. There does appear to be some difference in the level of uncertainty, as presented by the confidence/credible intervals, but generally the means were close and the ranking consistent. The Abbott151 evidence synthesis model was extremely difficult to interpret; however, the analysis enabled the estimation of the joint probability of an ACR response and a PsARC response at 12 weeks. The 24-week results of the PsARC and ACR were then estimated individually, conditional on the 12-week response. Schering-Plough152 based their evidence synthesis on a previous HTA report,73 which linked two meta-analyses, one estimating PsARC the other HAQ conditional on PsARC.

Health Assessment Questionnaire conditional on a Psoriatic Arthritis Response Criteria response The economic models developed by both the Schering-Plough152 and the Assessment Group required an estimate of the expected change in HAQ in the first 3 months for treatment responders and non-responders, as measured by PsARC. HAQ conditional on a PsARC response was modelled by both the Assessment Group and Schering-Plough. 152 The two modelling approaches were based on fixed-effects meta-analysis. The Schering-Plough152 approach uses two linked meta-analyses, which estimated the probability of response and then the mean reduction in HAQ score conditional on that response. The Assessment Group estimated the probability of PsARC response in one meta-analysis and then used this result to inform a second HAQ model. Both synthesis models used the same clinical trials to inform the HAQ–PsARC estimates. However, Schering-Plough152 used the latest available end points for HAQ, in contrast with the Assessment Group, who elected to use the 12- to 16-week HAQ data to reflect short-term benefits. Long-term benefits are considered explicitly in the economic model.

The results obtained (Table 22) were generally similar, with the drugs maintaining the same ranking. The differences may reflect the slightly differing modelling approaches or the difference in data used. The Assessment Group included only the five trials that reported HAQ outcomes for responders and non-responders. To enable them to include all six trials, Schering-Plough152 assumed that for the one trial where the data were not stratified by responder/non-responder78 the HAQ change for the PsARC non-responder was equivalent to the average HAQ change in the non-responders, as seen in other trials, and that the HAQ change for the PsARC responders could be inferred to match the reported mean HAQ change. The Assessment Group opted not to make this assumption, as it was not clear that it was appropriate or that it would have a significant impact on the results obtained. The Assessment Group took the decision to use only data that reported in a manner that facilitated modelling. The Schering-Plough152 report clearly states that six trials were considered; however, the detailed appendix and model code both appear to consider a seventh trial of the biologic golimumab. Although they state that this was used only to inform relationships between variables, the coding and appendix do not make this clear.

Abbott151 did not model HAQ conditional on response, although HAQ for the economic modelling section of their report they did state that relationships between ACR response rate and HAQ improvement, and PASI response and PASI improvement were developed in order to obtain estimates of HAQ and PASI improvement for responders and non-responders for each treatment.

This analysis estimated the expected change in HAQ in the first 3 months, conditional on treatment response. PsARC is not a baseline variable and therefore conditioning the analysis on PsARC response may be potentially biased. The analysis assumes there are no confounding factors (unrelated to treatment received) that change during the trial and affect both PsARC response and, independently, the change in HAQ.

Psoriasis Area and Severity Index 50/75/90 response The PASI outcomes were synthesised by Abbott,151 Schering-Plough152 and the Assessment Group. Schering-Plough152 elected to use absolute PASI change as their main outcome, on the basis that this was the most appropriate outcome for the economic modelling. As a result, the estimates obtained are not comparable with the Assessment Group or Abbott151 results, both of which elected to use probability of achieving each PASI outcome (50/75/90) as their main outcome. This was achieved using two different modelling approaches. The Assessment Group elected to use an ordered multivariate logit model, whereas Abbott151 chose to use a bivariate probit model. The logit and probit models are similar; both allow the different thresholds of PASI (50/75/90) to be modelled simultaneously, the ordered nature of the data to be maintained and an estimate of patients’ percentage reduction in PASI score from baseline to be obtained. The results estimated and presented in Table 23 are similar. As previously stated, the Abbott model151 was complex and (as felt by the assessment team) difficult to fully understand. As such it is not clear if data from all 10 included trials were used in the Abbott PASI model. 151 The data inputs for the Assessment Group model are reported in Appendix 5. Owing to a lack of reporting in some trials, the Assessment Group model included data from five trials, one of which provided data on only two of the outcomes (PASI 50/75).

American College of Rheumatology 20/50/70 response Schering-Plough152 did not synthesise for this outcome. Both the Assessment Group and Abbott151 did, but again elected to use two differing modelling approaches, ordered logit and bivariate probit. The comparative results are presented in Table 24. The results are again similar, with the ranking of the drugs being maintained.

Abbott’s model151 produced estimates of 24-week ACR response conditional on the 12-weeks ACR response rate. The 12-week response rate was modelled as a joint distribution of 12-week PsARC and ACR response rates. The code and explanation of this modelling was not clear and therefore it was not possible to fully interpret all of the modelling conducted. As the Abbott economic model151 included both PsARC and ACR there was a need for them to estimate the correlation between these two outcomes. The correlation was estimated using the available evidence. However, it was unclear as to the number of trials informing the Abbott ACR synthesis151 and the correlation estimate. The Assessment Group have presented an ordered logit model, using data from all six trials. The estimates obtained were not used in the Assessment Group economic model, so it was not necessary to make any assumptions on the correlation between PsARC and ACR outcomes.

The annotated winbugs code, assumptions and data have been presented for all models used by the Assessment Group. Although it can be difficult to justify some of the differences in modelling assumptions taken by the various groups, the Assessment Group have tried to reflect clinical reality, minimise generalising assumptions and allow the results obtained to reflect the evidence obtained as part of the clinical review.

Etanercept

Randomised controlled trials in psoriatic arthritis Two placebo-controlled RCTs evaluated etanercept in patients with PsA. The first, which followed 60 patients for 12 weeks, reported more infections in the etanercept group than the placebo group for respiratory tract infection (27% vs 13%, respectively), pharyngitis (17% vs 10%), rhinitis (17% vs 13%) and sinusitis (10% vs 7%). Influenza was more commonly reported in the placebo group (0% vs 20%). 78 However, given the small numbers of patients in each group, these differences could be attributable to the play of chance. No deaths or withdrawals due to adverse events were reported for either group. Data on cancer and TB were not clearly reported.

A second, larger placebo-controlled RCT by the same authors, followed 205 patients over 24 weeks. 52,97,99,105,107,110 One patient in the placebo group died following surgical complications, and one patient from each group withdrew from the study. There were no reported cancers. Similar rates were observed between the etanercept and placebo groups for upper respiratory tract infection (URTI) (21% vs 23%), sinusitis (6% vs 8%) and urinary tract infection (6% vs 6%), although again, this efficacy study was not powered to detect a difference between groups in terms of adverse events. TB was not reported.

Non-randomised studies/large randomised controlled trials in other conditions Thirteen non-randomised studies, in which more than 500 patients received biologic agents, reported adverse event data for etanercept. The majority of treated patients had RA, although outcomes for PsA, juvenile idiopathic arthritis, ankylosing spondylitis and patients with other chronic inflammatory conditions were also reported (Table 26). Average length of follow-up ranged from 48 weeks to 7 years.

The total reported rate of infections ranged from 9.6% to 54.4% (reported by five studies), with serious infections (requiring hospitalisation) ranging from 2.6% to 16.2% (nine studies). Only three studies clearly reported cancer, with rates ranging from 1% to 5.7%. Seven out of eleven studies reporting rates of TB in patients receiving etanercept found no cases. The remaining four studies reported rates ranging from 0.03% to 1.4%. Four studies reported rates of withdrawal due to adverse events, ranging from 4.6% to 13.6%. Where reported, mortality ranged from 0% to 3.1% (five studies).

Two of these studies compared adverse event rates in patients receiving etanercept against control. 99,122 One cohort study122 reported significantly more infections in patients with RA receiving etanercept than control patients (22.6 vs 6.8 infections per 100 patient-years, p < 0.01; 6.4 vs 2.3 serious infections per 100 patient-years, p < 0.01). However, a second study, an analysis of collated trial data on the use of etanercept reported no significant difference in overall infection rates between etanercept and control (placebo or MTX) across a range of conditions (54.4% vs 41.4%, p > 0.05). 99

Infliximab

Randomised controlled trials in psoriatic arthritis Two placebo-controlled RCTs evaluated infliximab in patients with PsA. 79–82,89–91,95,96,98,106,109,111–118 One RCT followed 104 patients over 16 weeks, reporting more respiratory tract infections in placebo-treated patients than in infliximab-treated patients (9.8% vs 1.9% respectively), though reported rates of bronchitis (7.8% vs 5.8%) and rhinitis (3.9% vs 5.7%) were similar between groups. 79–81,89,96,109,111,113–115,117,118 However, the very small numbers of events reported preclude any meaningful interpretation of these differences. No deaths or withdrawals were reported for either group.

The second RCT followed 200 patients over 24 weeks and reported similar rates between infliximab and placebo groups for URTI (10% vs 14%), pharyngitis (5% vs 4%) and sinusitis (5% vs 4%), although as with other RCTs, the study was not powered to detect differences in adverse events. 82,90,91,95,98,106,112,116 One patient in the placebo group developed basal cell carcinoma of the skin, although no deaths or withdrawals due to adverse events were reported.

Non-randomised studies/large randomised controlled trials in other conditions Eighteen non-randomised studies and two RCTs in indications other than PsA reported adverse event data for infliximab. Outcomes were reported for patients with PsA, juvenile idiopathic arthritis and ankylosing spondylitis, although the vast majority of patients had RA (Table 27). Average length of follow-up ranged from 22 weeks to 6 years.

The total reported rate of infections ranged from 8.7% to 26.6% (reported by four studies). Where detailed separately, the most common infections were URTIs, ranging from 10.8% to 38.5% (three studies). Serious infections (requiring hospitalisation) ranged from 0.8% to 13.8% (12 studies). Eight studies reported total cancers, with rates ranging from 0.16% to 5.1%. Sixteen studies reported rates of TB in patients receiving infliximab, 11 of which reported rates less than 0.5%, with the overall range being 0% to 4.6%. Where reported, mortality ranged from 0.06% to 2% (seven studies). Four studies reported rates of withdrawal due to adverse events, ranging from 5.3% to 12.8%.

Four of the studies compared adverse event rates for patients receiving infliximab against some form of control group. 119,122,139,143 Two of these were RCTs of infliximab versus placebo plus MTX in RA,139,143 of which one found no difference in serious infections between groups at 22 weeks (3.3% vs 1.7%, p > 0.05),139 and one reported significantly more serious infections associated with infliximab at around 54 weeks (5.3% vs 2.1%, p < 0.05). 143 Two cohort studies compared adverse event rates between infliximab and control patients: one reported significantly higher rates of overall infections (28.3 per 100 patient-years vs 6.8 per 100 patient-years, p < 0.01) and serious infections (6.2 per 100 patient-years vs 2.3 per 100 patient-years) among patients with RA receiving infliximab,122 the second reported no significant differences in serious infections (1.6 per 100 patient-years vs 1.1 per 100 patient-years) or cancer (0.4 per 100 patient-years vs 0.5 per 100 patient-years) or mortality (0.3 per 100 patient-years vs 0.2 per 100 patient-years). 119

Adalimumab

Randomised controlled trials in psoriatic arthritis The smaller of the two RCTs evaluating adalimumab (102 patients over 12 weeks) reported more overall infections in placebo-treated than adalimumab-treated patients (32.7% vs 17.6%, respectively), with the infection classified as ‘serious’ for a single patient in each group. Reported rates of URTI were 8.2% and 13.7%, respectively. 83 As with other RCTs, small numbers of events reported limit meaningful interpretation of these differences. No deaths were reported for either group, and the small proportions of withdrawals were comparable.

The larger trial, which randomised 315 patients over 24 weeks, reported similar rates between adalimumab and placebo groups for URTI (12.6% vs 14.8%, respectively) and nasopharyngitis (9.9% vs 9.4%). 51,88,92,93,100–104 Serious infections were reported in three patients; two receiving adalimumab and one receiving placebo. No deaths were reported.

Non-randomised studies/large randomised controlled trials in other conditions Eight non-randomised studies and two RCTs in indications other than PsA reported adverse event data for adalimumab. Outcomes were reported for patients with PsA, juvenile idiopathic arthritis and ankylosing spondylitis, although, as for the other agents, most patients had RA (Table 28). Average length of study follow-up ranged from 12 weeks to 5 years.

The total reported rate of infections ranged from 9.1% to 45.3% (three studies), with serious infections ranging from 0.4% to 7.3% (nine studies). Four studies reported total cancer, with rates ranging from 0.1% to 1.1%. Eight studies reported rates of TB in patients receiving infliximab, ranging from 0% to 0.4%. Four studies reported rates of withdrawal due to adverse events, ranging from 5.8% to 10.7%. Where reported, mortality ranged from 0.2% to 0.9% (three studies).

Two of these studies were RCTs of adalimumab in conditions other than PsA. 133,140 One RCT of adalimumab alone or in combination with MTX against MTX alone in patients with RA, reported no difference between adalimumab monotherapy and MTX monotherapy in terms of overall infections (110 per 100 patient-years vs 119 per 100 patient-years), serious infections (0.7 per 100 patient-years vs 1.6 per 100 patient-years), or cancer (0.9 per 100 patient-years in each group). However, significantly more serious infections were observed for combined adalimumab/MTX therapy than for adalimumab monotherapy (2.9 per 100 patient-years vs 0.7 per 100 patient-years, p < 0.05). 140 The second RCT reported that, after 56 weeks of treatment in patients with Crohn’s disease, no significant differences were found between adalimumab and placebo in terms of overall (45.3% vs 36.8%) or serious infection rates (2.7% vs 3.4%). 133

Studies reporting more than one agent

No RCTs exist that provide a head-to-head comparison between any of the three agents of interest, and substantial clinical heterogeneity precludes any meaningful comparison of rates between the different uncontrolled studies summarised above. However, limited information on the relative rates of certain adverse events between agents was reported by 10 of these uncontrolled studies (Table 29).

Patients with RA predominated and the average length of study follow-up (where reported) ranged from 1 to 5 years. One prospective cohort study reported a total rate of infections of 21.3% (6.1% serious) and 26.6% (5.8% serious) for etanercept and infliximab, respectively. 122 Three more studies reported rates of serious infections for all three agents: etanercept (5.8%, 11.2%, 4.5%), infliximab (8.9%, 13.8%, 8.1%) and adalimumab (5.1%, 7.3%, 6.6%). 129,136,147

Rates of TB were reported in seven studies of patients receiving etanercept (0%–1.4%) and infliximab (0%–4.6%), four of which also included patients receiving adalimumab (0%–0.3%).

One large prospective cohort study of reported that 0.76% of patients treated with biologic agents developed cancer during follow-up. 148 None of the studies provided adequate data on rates of withdrawal, and no studies provided separate mortality data for each agent.

Summary of serious adverse events across all three agents

Table 30 summarises the rates of serous adverse events, where reported, among the included non-randomised studies and large RCTs. This indicates that the rates of serious adverse events cover a broadly similar range across the three different biologic agents. However, it should be noted that all of these estimates are derived from a highly heterogeneous group of studies in terms of participants (e.g. inflammatory condition, disease severity), study design (e.g. length of follow-up) and treatment regimens (e.g. dose and frequency). Consequently, reliable estimates of the relative rate of serious adverse events for each drug cannot be made.

Withdrawal rates due to adverse events were typically < 10% for all drugs, with the highest reported single estimate being 13.8% for one etanercept study. This would suggest that the majority of patients can tolerate biologic treatment in the medium term, although again these estimates are derived from a highly heterogeneous group of studies, therefore poorer tolerability in specific patient groups cannot be ruled out.

Study design and quality

All six included studies were randomised, double-blind controlled trials. Based on the quality assessment using the pre-specified criteria, all the included trials were rated as ‘good’ quality. Concealment allocation and blinding were adequate in almost all included trials. All of the trials appeared to deal with withdrawals appropriately by using intention-to-treat (ITT) analyses. The completeness of follow-up was fairly good in all trials, with losses to follow-up of < 20%, thereby minimising attrition bias. 172 All the trials reported the use of a power calculation to determine the sample size. Five of them had an open-label extension after the randomisation period. However, it should be noted that the maximum randomised follow-up period across these trials was only 24 weeks.

Though there were some differences relating to patients’ characteristics at baseline across the trials, participants were generally similar in terms of disease activity and severity, and were likely to represent a population with moderate to severe PsA requiring further treatment. This was reflected by the lack of evidence for statistical heterogeneity in most efficacy analyses in this review. However, although the majority of patients in the trials had previously received at least one DMARD, no trial specified the failure to respond to at least two DMARDs (patients whom the current BSR guidelines consider eligible for biologic treatment) as a recruitment criterion. Therefore, trial participants were not precisely representative of patients receiving these agents in practice, and were likely to have had less severe disease, having often received biologic therapy after failing a single DMARD.

There were inconsistencies in the choice of primary outcome between included studies. Most studies used the ACR 20 as the primary outcome measure, while one trial used the PsARC as the primary outcome. However, it should be noted that ACR 20 is not frequently used in routine clinical practice to measure response to a biologic treatment.

Outcomes relating to joint disease

There were limited efficacy data from RCTs for the three biological agents. For each agent, there were two RCTs with around 200 or fewer patients receiving active treatment. However, all six trials were of good quality and provided clear indication of a response to treatment at 12–16 weeks, with continued efficacy at 24 weeks for each biologic agent.

Point estimates of effect sizes were generally moderate to large, implying that these treatment effects could be clinically significant. Moreover, although a very small number of studies were pooled for each estimate, the CIs indicate reasonable precision of these estimates. However, pooling the long-term efficacy data from trials was impossible due to lack of data.

In general, there was no significant heterogeneity in the treatment effect for almost all of the efficacy outcomes, with the PsARC in infliximab being the only exception. The radiographic data from RCTs of etanercept and adalimumab in PsA demonstrated a beneficial effect on joint disease progression at 24 weeks. Follow-up this early is often considered insufficient to detect radiological changes, although if the 24-week effect is reliable it would indicate a rapid onset of action in terms of joint disease for these agents. The open-label extensions of these RCTs also provided data on radiographic assessment at long-term follow-up, indicating that the effect on joint disease progression may persist over time. However, the reliability of these longer-term data was compromised by the lack of a control group.

Functional status (Health Assessment Questionnaire)

All three agents appeared to have beneficial effects on functional status as measured by HAQ. The estimates with relatively high precision indicated that all of the biologic therapies significantly improved the functional status of patients with PsA at around 3 months’ follow-up. The clinical significance of these effects was not entirely clear, for example, adalimumab was associated with a significant absolute mean reduction of HAQ score from baseline of –0.27 (95% CI –0.36 to –0.18). However, only changes > –0.3 have been considered as clinically meaningful improvement in PsA. 150

In this systematic review, the benefit of the biologic treatment compared with placebo on joint disease outcomes was consistent with the previous systematic review, which investigated the efficacy of etanercept and infliximab in the treatment of PsA. 73 In general, both of the systematic reviews used the same rigorous methodology and revealed similar magnitudes of the treatment effect of etanercept and infliximab. The current review also assessed effects of the recently licensed biologic agent ‘adalimumab’ and demonstrated its beneficial treatment effects compared with placebo.

Outcomes relating to skin disease (psoriasis component)

Skin outcomes (i.e. PASI response) were less commonly reported than joint response measures. Where reported, these results were generally statistically significant, although CIs were wide – possibly due to the small sample size of patients evaluable for psoriasis in the trials. Overall, biologic treatment appears to have a broadly beneficial effect on skin disease in patients with PsA. Evidence of response from trials in patients with psoriasis lay outside the scope of this evaluation. 173,174

Relative efficacy of the biologics

As data for the direct head-to-head comparison between these biologic agents were not available from trials, the relative efficacy of these biologic agents in the treatment of PsA was evaluated using Bayesian indirect comparison methodology.

The results of this evidence synthesis highlighted the superior efficacy of biologics over placebo across the outcomes evaluated. Infliximab appears to be the most effective among the three biologics. Patients treated with infliximab had a higher probability of responding to treatment regarding both the skin and arthritis aspects of disease. Additionally, we have estimated that infliximab allows improvements in the functional and psychological impact of the disease, measured by HAQ. However, patients who responded to etanercept achieved similar mean changes in HAQ (–0.6275 for infliximab and –0.6235 for etanercept) with placebo non-responders being used as a baseline in the synthesis. For all three biologics the changes in HAQ for those patients who did not respond to treatment were below the suggested minimum clinically significant threshold,150 and only those for infliximab achieved statistical significance. A comparison of the indirect comparison undertaken by the Assessment Group with those of the manufacturers shows similar mean estimates of treatment effect despite the rather different methods used.

Study design and quality

For the evaluation of adverse events of these biological agents, this review included a range of study types including RCTs, trial open-label extensions and observational studies. The quality of studies therefore varied across these different study designs; in particular, observational studies were subject to confounding, thereby threatening the internal validity of their findings. In addition, the definition of serious adverse events was also unclear in most studies.

Outcomes relating to serious adverse events

Previous systematic reviews have focused on short-term follow-up and reported conflicting findings on the risk of serious infections and cancer associated with biologic treatment. Our current systematic review contributes an evaluation of potential serious adverse events of biologic treatment in the longer term, incorporating the risk of activation of latent TB. Although the estimates of the rates of these adverse events varied widely, the findings from our review did raise a concern that treatment with etanercept, infliximab and adalimumab might be associated with an increased risk of serious infection, malignancy and activation of latent TB. The adverse event analyses demonstrated that etanercept, infliximab and adalimumab were associated with a broadly similar range of incidences of these events. However, there was considerable uncertainty around these estimates, in part due to the high degree methodological and clinical diversity between the included studies. In addition, the adverse event data were derived primarily from patients with RA or other indications, so the generalisability of these findings to patients with PsA remains unclear. Overall, the limited evidence prevents firm conclusions about the comparative safety of the three biologic agents being drawn from our systematic review.

Overview

A probabilistic decision-analytic model was developed to estimate the costs and QALYs of the three biologics over a lifetime (40 years) compared with palliative care only. The model has similarities with the earlier York Assessment Group model, but a number of changes have been implemented, necessitating a full description of the model here. The model aims to be consistent with licensed indications and current BSR149 and BAD173 guidelines for the use of biologics in PsA (Box 1).

Patients with severe skin psoriasis and PsA who have failed, or cannot use, MTX may need to be considered for biologic treatment, given the potential benefit of such treatment on both components of psoriatic disease

The parameters of the model were obtained from published literature, manufacturers’ parameter estimates, the results of the evidence synthesis in Chapter 3, Efficacy of all three biologics and a structured elicitation of expert opinion. The model adopts the perspective of the UK NHS and Personal Social Services. The price year is 2008–9 and the annual discount rate is 3.5%. 184 The population is assumed to be 47 years old, with at least 7 years since diagnosis of PsA, based on the average characteristics of participants in the RCTs (see Table 1). The body weight is assumed to be between 60 and 80 kg, based on the mean adult weight in the UK general population (women 69.7 kg, men 83.5 kg185). Patients are assumed to have failed at least two DMARDs. In the base case, patients are assumed to fulfil BSR criteria (see Box 1). In the base case the HAQ at the start of the model is 1.05, based on the average in the RCTs (see Table 1). Although the mean HAQ when patients start biologics in the BSR register was 1.8,186 clinical opinion suggests that, in current practice, clinicians are more likely to offer biologics early in the course of the disease.

Clinical opinion suggests that about 50% of patients starting biologics have mild or minimal psoriasis (< 3% BSA or a PASI score of < 2.5), 25% have mild-to-moderate psoriasis (a baseline PASI score of between 2.5 and 10), and 25% have moderate-to-severe psoriasis (a PASI score > 10) (Ian Bruce, Arc Epidemiology Unit, University of Manchester, UK, 20 November 2009, personal communication). Approximately 50% of patients in the RCTs had < 3% BSA psoriasis or a baseline PASI < 2.5 (see Table 1), indicating the trials are broadly representative of skin involvement in general practice. We assume patients in the base case have mild-to-moderate psoriasis with a PASI score of 7.5. The effect of biologic treatments in other patient subgroups is explored in scenario analyses.

Model structure

The model is a cohort model, assuming a homogeneous baseline population. The model has a Markov structure (see Figure 2). Patients enter the model either (i) when commencing therapy with etanercept, infliximab or adalimumab or (ii) with no therapy (assumed to be palliative care only).

FIGURE 2.

Structure of the decision model, assuming patients continue beyond 3 months if they achieve a PsARC response.

Initial response at 3 months

Table 33 shows the parameters used in the base-case model. Initial response of the drug is defined in the model as PsARC for joints and PASI 75 for psoriasis, based on the BSR149 and the BAD guidelines173 (Box 2). These parameters were estimated by the evidence synthesis (see Chapter 3, Efficacy of all three biologics).

For patients on TNF antagonist treatment with psoriasis and PsA, treatment may be continued if there has been a sufficient response in at least one of these components (see BSR guidelines for definition of disease response in PsA).

The BAD guidelines highlight that the recommended time points for assessing the initial response vary between drugs and between guideline-making bodies. The licences for psoriasis recommend an assessment at 14 weeks for infliximab, at 12 weeks for etanercept and at 16 weeks for adalimumab. Current NICE guidelines for psoriasis recommend an assessment at 10 weeks for infliximab. In the current appraisal we do not make these distinctions and assume that an assessment is made for all drugs at ‘around 3 months’ or between 12 and 16 weeks. The assessment of effectiveness in Chapter 3 (see Assessment of effectiveness) did not find any appreciable differences in the biologics’ response rates for joint disease or psoriasis between approximately 12 weeks and 24 weeks.

In the decision model, the change in HAQ compared with baseline is conditional on whether a PsARC response was achieved. These parameters were estimated by the evidence synthesis in Chapter 3, Efficacy of all three biologics. It is uncertain whether the change in HAQ is the same for all PsARC treatment responders, or depends on the particular biologic treatment followed. In the opinion of our clinical advisor, either scenario could be plausible (Ian Bruce, personal communication). In the base-case model, we allow the change in HAQ for treatment responders to depend on PsARC response and the biologic treatment, and consider the alternative scenario as a sensitivity analysis. According to the evidence synthesis in Appendix 5, the mean change in HAQ in the first 3 months for PsARC responders, across all biologic drugs, is –0.5688 [standard error (SE) 0.0315] and the mean change in HAQ for PsARC non-responders, across all biologic drugs, is –0.1697 (SE 0.0338).

During the initial 3-month trial period the model assumes that patients on biologics have some improvement in HAQ even if they do not reach the PsARC threshold. These parameters were estimated by the evidence synthesis in Chapter 3, Efficacy of all three biologics. Patients who do not achieve the required level of response during the first 3 months and are withdrawn from therapy are assumed to return to the same HAQ score after withdrawal as patients who had palliative care only.

The model assumes that patients who achieve a PASI 75 response will gain at least a 75% improvement in psoriasis compared with baseline PASI. The calculation of the expected improvement in PASI for PASI 75 responders is described in Appendix 18. Patients who do not achieve a PASI 75 response will also have some proportionate gain in PASI while they continue taking a biologic, although this will be less than a 75% improvement (see Appendix 18).

A proportion of patients in the placebo arms of the RCTs achieved a PsARC response and an improvement in HAQ. Part of the response in both the placebo and treatment arms of RCTs may be due to non-pharmacological aspects of medical care that would be common to both arms (sometimes called a ‘placebo’ or ‘expectancy’ effect). It is uncertain whether this effect would be reproducible in general practice. 189 In the base case we assume that part of the predicted response for treatment observed in the trial is attributable to the controlled trial setting and would not be reproducible in general practice. The change in HAQ in patients using biologics is reduced by the mean change in HAQ across the placebo arms of the RCTs. A similar adjustment is made for the expected change in PASI in patients using biologic therapy. Appendix 9 gives further details of the conceptual framework and adjustments made for the possible placebo/expectancy effects. An alternative scenario assumes that the response rate to treatment in the RCTs is fully generalisable to general practice and no adjustment for placebo/expectancy effects is made.

Because there are two response variables (PsARC and PASI), there are four possible outcomes at 3 months: skin response only, joints response only, response of both and response of neither (Figure 2). The base-case model assumes that the responses to psoriasis and arthritis might be correlated. Appendix 10 reviews the evidence on the correlation between these responses and how the decision model calculates the probabilities of each of the four outcomes at 3 months. An alternative scenario assumes that the responses to psoriasis and arthritis are independent.

The BSR guidelines recommend that biologics are withdrawn if a PsARC response is not achieved at 3 months. This rule is used in the base-case analysis of the model. However, in patients who have significant skin and joint disease, some patients may achieve PsARC but not PASI 75, or achieve PASI 75 but not PsARC. In these cases, one could specify that patients should continue biologic therapy irrespective of the psoriasis response (BSR guideline), or those that respond to either can continue (BAD guidelines) or (in principle at least) only those that achieve both should continue. These alternative continuation rules are explored in sensitivity analyses.

The model assumes that no patients withdraw due to adverse events in the first 3 months. This is because the RCTs estimate responses on an ITT basis, whereby withdrawals for any reason are considered treatment failures and counted as non-response. Including withdrawals during the first 3 months in the model would, therefore, be double-counting.

Long-term outcomes and withdrawal from biologic therapy

If the decision is made to continue with the biologic therapy beyond 3 months, it is assumed that patients maintain their initial improvement in HAQ while on that therapy. This is based on evidence from an opinion elicitation exercise from clinical experts, and supported by data on HAQ and HRQoL from biologics registers. 186,190 Appendix 11 describes the opinion elicitation methods and results used to inform the model. It is assumed that patients maintain the improvement in PASI while on biologic therapy. This assumption has been made in other decision models (see Systematic review of existing cost-effectiveness evidence).

There is an ongoing risk of withdrawal from biologic therapy. Withdrawal might occur for lack of continuing efficacy (‘secondary non-response’), adverse events or other reasons. The rate of withdrawal after 3 months is assumed to be independent of the HAQ and PASI score in the model, to be independent of whether the initial response was for both psoriasis and arthritis or just arthritis, and to be constant over time. The rate is estimated from a meta-analysis of registry data from several countries to be –1.823 (SE 0.2044) on the log scale, or exp(–1.823 + 0.5 × 0.2044²) = 0.165 per year (see Appendix 12). Although the registries present withdrawal rates by drug, these data are not randomised and patient cohorts starting on different biologic therapies are unlikely to be similar. 191 Therefore, the decision model assumes the same withdrawal rates for all biologics. Appendix 12 gives further details. As the withdrawal rate is constant over time after the first 3 months, patients who achieve an initial PsARC response will on average remain on biologic drugs for just over 6 years in the model (1/0.165 = 6.06 years).

Patients withdraw from biologic to palliative care only. On withdrawal, it is assumed that mean PASI returns to its initial score at baseline (rebound equal to initial gain). There is considerable uncertainty about change in HAQ associated with withdrawal (rebound). Previous modelling work assumed rebound of HAQ follows either of two alternative scenarios, with no data to inform which scenario is the more likely: rebound equal to initial gain, and rebound equal to natural history (NH). 177 These scenarios are explained in more detail in Appendix 11. The current model is informed by the expert opinion elicitation exercise conducted with five experts, described in Appendix 11. All experts suggested that not all the initial gain in HAQ is lost following late withdrawal of patients who initially responded to biologic therapy at 3 months. This scenario, that the HAQ rebound might be less than initial gain, has not been considered in any of the previous models of PsA, nor, to our knowledge, in any model of RA. Given the difficulty and limitations of eliciting expert opinion and the novelty of these findings, the current model assumes that rebound is equal to initial gain in the base case, and explores other scenarios (rebound less than initial gain and rebound equal to NH) in sensitivity analyses.

Outcomes for patients on palliative care

The PASI is assumed not to change on average compared with baseline for patients undergoing palliative care. HAQ is assumed to progressively worsen in such patients at a constant rate, estimated by an analysis requested from Deborah Symmons and colleagues at Manchester University for this appraisal using data from the Norfolk Arthritis Register (NOAR) (see details in Appendix 14).

Illustration of progression of HAQ in the model

Figure 3 illustrates the progression of HAQ over time for three different patient histories in the model. For a patient whose arthritis is controlled by biologic therapy, HAQ score is initially reduced (improves) and then maintained over time. For a patient who does not start biologic therapy, HAQ increases (deteriorates) over time to a maximum score of 3. For a patient who withdraws at 5 years, HAQ ‘rebounds’ (quickly increases) to the baseline level after withdrawal and then increases at the same rate as those who never started biologic therapy. However, in this scenario (‘rebound equal to initial gain’) the 5-year delay in progression obtained while on biologic drugs is permanently maintained after withdrawal.

FIGURE 3.

Illustration of the progression of arthritis for a patient successfully maintained on biologic, a patient without biologic and a patient who withdraws at 5 years. Note: a greater HAQ score indicates worse disability.

Utility

Health utility is measured as a function of HAQ and PASI. This relationship was estimated from analyses provided by the manufacturers, who carried out linear regressions of EQ-5D utility versus HAQ and PASI in participants in key RCTs (see Appendix 17). The base-case utility function is:

Other utility functions, supplied by the manufacturers, were used as sensitivity analyses.

Figure 4 illustrates the change in utility over time for different patients in the model. For a patient who is maintained on biologic therapy, utility is initially improved as a consequence of the reduction in HAQ and PASI, the latter depending on the proportion of patients who respond to psoriasis, given a response of arthritis (see Figure 2 and Appendix 10). This utility gain is assumed to be maintained over time. For a patient who did not start biologic therapy, utility deteriorates over time to a minimum value that is < 0, indicating that the general population would consider HRQoL with the severest arthritis symptoms and uncontrolled psoriasis to be worse than death. For a patient who withdraws at 5 years, utility ‘rebounds’ to the baseline level after withdrawal and then deteriorates at the same rate as those on NH. The area between these curves (area ‘A + C’ in Figure 4) represents the difference in lifetime QALYs between a patient who withdraws at 5 years and a patient who never uses biologic therapy.

FIGURE 4.

Illustration of utility (HRQoL) of a patient successfully maintained on biologic, a patient without biologic and a patient who withdraws at 5 years. Note: EQ-5D utility takes a maximum value of 1, indicating full health; values of < 0 correspond to health states that are considered worse than death by the general population.

Time horizon for maintaining treatment effects

It is uncertain whether the effectiveness of biologic therapy is maintained in the very long term. Previous models considered a scenario where it is assumed that all patients withdraw from biologic therapy at 10 years, and all gains in HAQ with respect to NH are lost at this point. 177 Figure 4 illustrates the effect on utility of this ‘10-year time horizon for treatment effects’ scenario compared with the base case that assumes that treatment effects are maintained over the lifetime.

The difference in lifetime QALYs for a patient who is maintained successfully on a biologic, compared with NH, is area A + B + C + D. However, if is assumed that treatment effects last for only 10 years, the difference in QALYs over 10 years between being on a biologic and NH is only area A + B. For a patient who withdraws from a biologic at 5 years, the difference in lifetime QALYs compared with NH is area A + C. The difference in QALYs between assuming a 10-year time horizon and assuming a 40-year time horizon for a patient who withdraws from therapy at 5 years is area ‘C’. Biologic therapy appears much more effective if it is assumed that treatment effects in those who withdraw and those who do not withdraw are maintained over the long term. The base-case model assumes that the benefits of biologic therapy are maintained for a lifetime. Time horizons for treatment remaining effective for up to 10 years and up to 20 years are considered in sensitivity analyses.

Health service costs

The acquisition costs of the drugs and of their administration and monitoring were obtained from BSR recommendations and pharmaceutical list prices65 (see Appendix 13). The base case assumes that four vials of infliximab are administered and that vial sharing is not permitted.

Health-care costs increase with severity of both arthritis36 and psoriasis. 37 The health service costs of treating arthritis were measured from a UK-based study that estimated the effect of HAQ on costs in patients with RA41,59 (see Appendix 15). The NHS costs used for treating mild-to-moderate psoriasis in patients who do not use biologics or who do not respond to biologics were obtained from NHS unit costs of phototherapy187 and a UK RCT. 192 No UK studies based on prospective IPD were identified to estimate the health service costs of treating moderate or severe psoriasis in patients who do not use biologics or who do not respond to biologics. In the model these costs were obtained from a Dutch RCT and adjusted to UK price levels188 (see Appendix 16).

All-cause mortality

All-cause mortality was estimated from UK life tables. A Gompertz function was fitted to these data (see Appendix 19). The base case uses a published estimate of the additional mortality risk in PsA. 29 The effect of biologics on mortality in PsA is uncertain. The US VA study of MTX in psoriasis and patients with RA found that MTX was associated with significantly reduced incidence of vascular disease. 193 Long-term control of chronic inflammation may reduce mortality. However, long-term use of biologics might increase other mortality risks. The decision model assumes that there is no difference in mortality rates between treatments, or between biologic treatments and no treatment.

Subgroup analyses

The base-case model assumes a cohort of patients with PsA with baseline HAQ of 1.05, the mean of HAQ across the RCTs (see Table 1), and mild-to-moderate psoriasis (baseline PASI of 7.5). The model considered other cohorts in subgroup analyses:

• A more severe baseline HAQ of 1.8, which is the mean HAQ of patients entering the British Society for British Society for Rheumatology Biologics Register (BSRBR). 186

• No skin involvement, with PASI of 0. Clinical opinion suggests 50% of patients with PsA starting biologics in clinical practice would have mild or no skin involvement (Ian Bruce, personal communication).

• A baseline PASI of 12.5, corresponding to moderate-to-severe psoriasis. 194,195 Clinical opinion suggests that 25% of patients with PsA starting biologics in clinical practice would have a baseline PASI > 10 (Ian Bruce, personal communication).

The review described in Chapter 3 did not find any evidence with which to assess whether treatment effects might differ by baseline severity, and, consequently, these analyses assume no change in relative treatment effects and focus just on variation between subgroups in baseline severity.

The base-case model assumes patients have failed at least two DMARDs, but are naive to biologics at baseline. The model was also used to estimate the cost-effectiveness of biologics used as a second course of therapy, if the first biologic is withdrawn. For example, if etanercept has been tried and failed, then the next alternative in sequence is adalimumab, infliximab or no biologic therapy. The reason why the patient failed the first course of therapy is potentially important information in deciding on the second course. Therefore, we consider two subgroups: one who failed the first biologic because of adverse events, and another who failed because of lack of efficacy. No RCTs have evaluated outcomes in these subgroups, and we estimate treatment response and withdrawal rates for these subgroups from observational data from the BSR register, which showed that if a patient failed first-line therapy for lack of efficacy, then the risk of failing the second-line therapy for lack of efficacy increased by 2.7 (95% CI 2.1 to 3.4). If a patient failed first-line therapy because of an adverse event then the risk of failing the second-line therapy for adverse events increased by 2.3 (95% CI 1.9 to 2.9). 196 Appendix 20 describes how these data were used to estimate the probability of initial response and later withdrawal for biologic therapies used as second line.

Analytic methods

The uncertainty in each parameter was represented using a probability distribution. The probabilities in Table 33 were assigned beta distributions. If p∼Beta(α,β) then α = E(p) × E(p) × (1 – E(p))/Var(p) and β = E(p) × (1 –  E(p)) × (1 –  E(p))/Var(p). The rate of change of HAQ while not on treatment was assigned a gamma distribution to ensure that values are strictly positive. If x∼Gamma(a,s) then a = E(x) × E(x)/Var(x) and s = Var(x)/E(x). All other uncertain parameters were assigned normal distributions with the mean and SE shown in Table 33. Probabilistic sensitivity analysis was carried out using Monte Carlo simulation.

The results of the model are presented in two ways. First, mean lifetime costs and QALYs for the three strategies are reported and their cost-effectiveness compared, estimating ICERs using standard decision rules. 197 Briefly, the alternative strategies are ranked by mean cost. Strategies that are more costly than another, but offer no greater expected benefit are known as ‘dominated’, and excluded. Strategies that are dominated by a linear combination of other strategies are considered subject to ‘extended domination’ and are also excluded. ICERs are then calculated for each of the remaining strategies, compared with the next best alternative. Although NICE does not specify a particular cost-effectiveness threshold, a strategy is more likely to be considered cost-effective if the ICER were < £20,000 per QALY, and less likely to be considered cost-effective if the ICER were > £30,000 per QALY. 184 Second, the decision uncertainty is shown as the probability that each intervention is the most cost-effective for a given cost-effectiveness threshold.

A series of alternative scenarios is also presented to explore the effect of changing one or more parameters/assumptions in the model.

Estimated probabilities of response at 3 months in the base case

Based on the results of the evidence synthesis in Chapter 3 (see Results of review of clinical effectiveness), and an estimate of the correlation between PsARC and PASI 75 outcomes in biologic therapy from an RCT,51 the model estimated the probability that a patient would respond for psoriasis only, joints only, both outcomes or neither outcome with each biologic therapy. These outcomes are shown under two assumptions: positive correlation (base case) and independence (Table 34).

Results of the base-case cost-effectiveness analysis

The results of the base-case cost-effectiveness analysis are shown in Table 35, and univariate sensitivity analyses in Table 36. The base-case analysis suggests that infliximab is the most effective treatment (in terms of expected QALYs), followed by etanercept then adalimumab. Infliximab is also the most costly treatment, followed by etanercept then adalimumab. The ICER of etanercept compared with palliative care is about £18,000, and the ICER of infliximab compared with etanercept is about £44,000 per QALY. Of the three biologic therapies, etanercept has the highest probability of being cost-effective at a threshold of between £20,000 and £30,000 per QALY. Etanercept is the most cost-effective strategy in 44% of simulations of the base-case model, at a threshold ICER of £20,000 and in 48% of simulations at a threshold of £30,000 per QALY.

Adalimumab is extendedly dominated by palliative care and etanercept. This means that if NICE were considering adalimumab, the ICER relative to palliative care would be £26,470/1.409 = £18,786. However, the expected QALY per patient achieved with etanercept is greater than for adalimumab (7.00 vs 6.58), while the ICER of etanercept versus palliative care is £17,853. Therefore, it would not, on average, be cost-effective to recommend adalimumab because a greater QALY gain can be achieved from etanercept within the threshold of £20,000 per QALY.

Expected QALYs are low in this model. The total lifetime discounted health associated with palliative care is about 5.17 QALYs. This is because the base-case scenario assumes that utility declines fairly rapidly in patients with uncontrolled arthritis, and may be < 0 in later years (see Figure 4). For comparison, if HAQ and PASI could be reduced to 0 for the complete time horizon of the model (40 years), the model predicts that this cohort would expect 15 QALYs, given the rate of mortality, the intercept of the utility function and the discount rate. Figure 5 partitions the lifetime discounted QALYs gained by biologic therapies into those associated with improving arthritis and those associated with improving psoriasis, relative to palliative care. In the base case, utility gains as a result of improvement in arthritis are predicted to be much greater than utility gains as a result of improvement in the psoriasis component of PsA.

FIGURE 5.

Gains in lifetime discounted QALYs associated with treating arthritis and psoriasis in PsA with biologic therapies relative to palliative care.

The expected lifetime (40-year) discounted costs without biologics (palliative care only) are about £42,000 in the base case for a patient with PsA and mild-to-moderate psoriasis. This can be partitioned into £29,000 for the treatment of arthritis and £13,000 for the treatment of psoriasis. Figure 6 partitions the total lifetime discounted health-care costs of the strategies into costs associated with the acquisition, monitoring and administration cost of the biologic drugs, the cost savings associated with treating arthritis (i.e. the reduction in HAQ score) and the cost savings associated with treating psoriasis (i.e. the reduction in PASI score). All costs are shown relative to the costs of palliative care.

FIGURE 6.

Lifetime discounted costs of biologic drugs, and cost savings for arthritis and psoriasis relative to non-biologic treatments for PsA.

The lifetime discounted acquisition, administration and monitoring cost of infliximab is about £52,000; etanercept is about £33,000 and adalimumab is about £27,000. These prescribing costs are much greater than any offset health-care cost savings elsewhere. Infliximab is associated with the greatest gains in PASI and HAQ, and the greatest cost savings. Adalimumab has the second greatest gains in PASI and associated cost savings, and etanercept has the second greatest gains in HAQ and associated cost savings.

Results of sensitivity analyses

Table 36 shows the results of the univariate sensitivity analyses. Table 37 shows the cost-effectiveness of the alternatives in each of the scenarios, assuming that an ICER of £20,000 or less is likely to be cost-effective and a strategy with an ICER of ≥ £30,000 is unlikely to be accepted.