Abrocitinib, tralokinumab and upadacitinib for treating moderate-to-severe atopic dermatitis

Incidence and/or prevalence

As many as 1 in 5 children and 1 in 10 adults in the UK are estimated to have AD,3,10 with about 18% of cases of childhood AD categorised as moderate and 2% as severe. 3 Of adults with AD, it has been reported that 5% of cases are severe. 11 Of the people who need treatment for AD, 7% are estimated to have moderate-to-severe disease, and 27% of those receiving treatment will require systemic therapy to elicit sufficient symptom relief. 12,13

Significance for patients in terms of ill-health (burden of disease)

Although the impact of AD varies with age and disease severity, common across ages is that AD affects various aspects of day-to-day living, including emotional and mental well-being, and family and social interactions. 14 As a disease predominantly affecting children, AD can consequently have a substantial impact on parents and other family members due to potential changes in lifestyle management, such as diet and family routine, to help manage symptoms. The treatment regimen required to sooth symptoms can be time intensive, and children are likely to require assistance from older siblings or adults to apply the topical treatments at the intervals needed for optimal effectiveness. In addition to the physical symptoms of AD, many children and adults experience sleeplessness, anxiety, depression and other mental health problems related to their AD. 10,14 Adults with AD frequently report decreased work productivity. 14

People with AD may also face a financial burden arising from extra costs associated with purchasing cleaning and laundry detergents, and bathing products tailored to sensitive skin. 3 Other costs could be incurred from travelling to appointments for assessment or treatment, and for emollients and moisturisers potentially not provided by the NHS. A study focusing on adults with AD that encompassed nine European countries, including the UK, reported that AD was associated with an annual cost to the patient of about £800 (GBP). 15

Significance for the NHS

In 2006, around 24% of people in England and Wales visited their general practitioner (GP) with a skin disorder, which is equivalent to 12.9 million people. 16 Of those presenting to primary care with a skin disorder, 0.8 million (6.1%) are referred for specialist advice, with most (92%) attending appointments with dermatologists within the NHS. 16 After diagnosis and establishing a treatment plan, the majority of care occurs at home, with topical treatments forming the mainstay of care, and many of which are purchased by the patient. Thus, it has been reported that, despite skin disorders being common, the cost of skin disease to the NHS is modest. In 2005–6, the direct cost to the NHS in England and Wales for managing skin disease was reported to be around £1820 million. 16 With the introduction of biological therapies for AD, which is one of the more common skin disorders, it is likely that the cost to the NHS for managing skin disease will rise in coming years.

Available treatment options

Interventions to be assessed

Quantity of research available

As noted in the section Identification of studies, the EAG identified a systematic review reporting a NMA of systemic treatments for moderate-to-severe AD that searched records up to August 2019. 47 The EAG considers the review to have been carried out systematically and to follow accepted systematic review methodology. The systematic review appraised completed and ongoing studies for all systemic therapies used in the management of AD, and captured studies on all interventions and comparators of interest to the MTA reported here. The EAG re-evaluated the studies included by the review against the inclusion criteria presented in Table 3.

Searches of electronic databases to update the search of the identified review (search date 1 August 2019 to 8 July 2021) retrieved 1365 records (post deduplication) that were of possible relevance to the review (see Report Supplementary Material 1). First-pass appraisal of the 1365 unique records led to exclusion of 1244 records. Full publications for 121 references from the EAGs’ literature review were ordered, of which publications for two records (both conference abstracts) could not be obtained. 58,59 Manual searching of the identified systematic review forming the basis of the update search identified 18 supplementary records for full-text appraisal. 60–77 The EAG’s SLR retrieved records for an additional seven systematic reviews that evaluated one or more systemic interventions of interest to the MTA:78–84 one of the identified reviews is a ‘living’ systematic review and as such is continually updated. 79 Cross-referencing of bibliographies of the seven systematic reviews identified one additional record85 to those retrieved by the EAG and the original review, giving a total of 138 full text publications screened for inclusion in the review. Additionally, two sets of committee papers outlining recommendations from NICE for the use of dupilumab12 and baricitinib13 in the management of AD were also identified by searching the NICE website. The EAG’s literature search captured all studies presented in the committee papers for baricitinib and dupilumab. As noted in the section Definition of the decision problem, the EAG had access to submissions to the STA process from the individual companies producing abrocitinib, tralokinumab and upadacitinib, with documents available including the original submission and responses to requests for data from the EAG. The EAG’s literature search identified all relevant studies reported in the company submissions.

Of the 138 full articles evaluated, 38 publications describing 23 studies were relevant to the review (including 4 errata; Table 4): citation details for conference abstracts identified during the EAG’s literature review and related to full publications are provided for completeness. Six additional studies for which full-text publications were not available at the time of writing were identified from searches of trial registries (ClinicalTrials.gov and EU Clinical Trials Register). The six studies each evaluated one of abrocitinib, tralokinumab or upadacitinib, and relevant results were provided by the companies during clarification (see Table 4). Summaries of the studies included in the review are presented by key characteristics of studies (see Table 4). A list of publications screened but subsequently excluded (with reasons for exclusion) from the review is available in Report Supplementary Material 1.

The update search of the literature carried out on 29 November 2021 identified an additional 377 unique records. Given that only 5 months had passed since the EAG’s original search, the EAG limited its inclusion criteria to full publications of RCTs not previously found. Of the 377 titles and abstracts appraised, only one record identified a novel RCT. 86 The study compares dupilumab 300 mg Q2W versus placebo with clinical efficacy assessed at 16 weeks in Chinese patients. Patients were required to be aged ≥ 18 years old and have moderate-to-severe AD for ≥ 3 years, which could not be adequately controlled with topical medications or for which topical treatment was inadvisable. Additional criteria for eligibility were EASI score ≥ 16, IGA score ≥ 3 and ≥ 10% body surface area (BSA) affected by AD. Based on the full-text publication, it is unclear whether people who had previously received systemic therapy were included, and, thus, data are not available for those having inadequate response to systemic therapy prior, or who cannot tolerate or are contraindicated to systemic therapy. Additionally, the study enrols only adults, and focuses on monotherapy. As data are not available by prior line of therapy, the study would not inform the EAG’s analyses and has not been described in full.

Seven publications61–67 describing four studies of CsA versus placebo were identified by the systematic review forming the basis of the EAG’s literature review. 47 The published studies met the inclusion criteria of the EAG’s SLR for RCTs. However, because none of the studies of CsA reported data on the clinical outcomes of interest to the review and the studies were carried out 15 years ago or longer, the studies are not discussed further, and study characteristics are not reported. As no RCT for CsA was available to inform a NMA, the EAG carried out a SLR of observational studies evaluating CsA in the treatment of AD (described in greater detail in the section Systematic review of existing cost-effectiveness evidence).

The company submission for upadacitinib reported results from a search to identify observational studies on CsA that could inform a NMA. The company’s review identified one study describing an indirect comparison of CsA with dupilumab plus TCS to generate an estimate of comparative clinical effectiveness for EASI 75. 87 The EAG considered the company’s literature review to be robust and chose to carry out an update search to identify additional potentially relevant studies.

Searches of electronic databases to update the company’s search of the identified review (search date 1 January 2019 to 8 July 2021) retrieved 746 records (post deduplication) that were of possible relevance to the review (see Report Supplementary Material 1). First-pass appraisal of the 746 unique records led to exclusion of 743 records. Full publications for three references from the EAG’s literature review were reviewed,87–89 one of which was the study described in the company submission for upadacitinib (Ariens et al.). 87 One publication was identified as a Letter to the Editor detailing the drug survival of dupilumab and was excluded. 89 The second publication described a retrospective observational study with a primary outcome of drug survival of dupilumab and CsA when used to treat moderate-to-severe AD in adults (N = 251), with EASI 75 captured as a secondary outcome (Dal Bello et al.). 88 Both full-text publications report data on proportion of people achieving EASI 75 with dupilumab and with CsA and so both meet the EAG’s inclusion criteria. 87,88 However, in Ariens et al.,87 the authors applied regression models to adjust data for people receiving CsA to those of people receiving dupilumab 300 mg Q2W with TCS from LIBERTY AD CHRONOS. 70 The authors repeated the analysis in reverse, that is, adjusting results from those treated with dupilumab from CHRONOS to those given CsA in the registry. No type of adjustment was reported in the second publication identified by the EAG. 88 Ariens et al. 87 reported the adjusted proportion of people achieving EASI 75 after 12–16 weeks of treatment with CsA to be 52%, compared with 75% as observed in the dupilumab group from CHRONOS. By contrast, Dal Bello et al. 88 reported achievement of EASI 75 at 16 weeks by 44.3% (66/149) of those in the dupilumab group compared with 24.5% (25/102) of people treated with CsA. The EAG acknowledges that there is a considerable difference in estimates for dupilumab between the two studies. As data reported by Ariens at el. 87 for dupilumab are derived from the RCT, CHRONOS, rather than a registry, and CHRONOS is included in the network to generate estimates of first-line treatment, the EAG considered the adjusted estimate for CsA reported by Ariens et al. 87 to be more the appropriate choice to inform a NMA of RCTs.

Quality of research available

Most of the studies included in the assessment of clinical effectiveness are considered to be well-conducted and well-designed Phase III RCTs, and, as such, are at an overall low risk of bias (summary of quality assessment in Report Supplementary Material 1). All studies are described as randomised, aside from the single observational study that provides evidence on CsA. 87 For most studies, the method of randomisation and allocation concealment for the included RCTs was deemed to be adequate. However, details on the methods used to generate the randomisation sequence were not available for five studies and so risk of allocation bias was judged to be unclear for these trials (see Report Supplementary Material 1).

Limited details were available in the full publications on the methods implemented to initially conceal allocation and to subsequently maintain masking of treatment from clinicians and participants. However, additional information was available from other sources, including clinical trial registries, the company submissions to the STA process and committee papers for recommended treatment options (dupilumab and baricitinib). The tools used to assess severity of the signs and symptoms of AD, and therefore level of improvement, are subjective in nature and so most recorded outcomes are at increased risk of bias. Most of the included studies employed a double-blind process to mitigate against introducing bias into outcome assessment. However, for most studies described as double blind, it was unclear whether the outcome assessor was the treating clinician and, if not, whether the outcome assessor was masked to treatment. Follow-up at the end of treatment (12–16 weeks) was high across most studies, with several studies categorised as a low risk of attrition bias for the outcomes evaluated. No study was assessed as high risk of selective reporting, with results routinely reported for all pre-specified outcomes.

The quality of the observational study87 used to facilitate comparison of upadacitinib and dupilumab in the first-line setting was assessed using the case–control component of the Newcastle–Ottawa Scale (NOS) for assessing the quality of non-randomised studies. 116 Based on the NOS, the observational study scored a point for most aspects of selection, comparability and exposure. The EAG acknowledges that the NOS is not an ideal tool to assess quality as Ariens et al. 87 does not include a control group, with both groups receiving an active treatment. Data on patients receiving CsA were acquired from secure records and, with the exception of EASI 75, the groups had comparable baseline characteristics. As noted in the section Ciclosporin A, the lower baseline EASI score of those receiving CsA reflects the patient population likely to present in clinical practice with moderate-to-severe AD, whereas people enrolled in clinical trials typically have more severe AD. Overall, the EAG considers the data presented by Ariens et al. 87 to represent the most robust evidence available to inform the NMA.

As highlighted in the section Quantity of research available, despite the low overall risk of bias of the studies, much of the data informing the analyses of comparative clinical effectiveness are derived from post hoc subgroups, which introduces uncertainty, the level of which cannot be quantified, into the analyses.

EASI 75 network meta-analysis: first-line systemic treatments – adult population

Upadacitinib and abrocitinib are both proposed as a first-line systemic therapy for adults having inadequate response to topical treatments.

The most relevant comparator in the first-line setting is CsA; however, no relevant RCTs of CsA were identified that could be linked in a NMA to the upadacitinib or abrocitinib trials. In the broader search for evidence to inform a comparison with CsA, Ariens et al. 87 was identified. As described in the section Quantity and quality of research available, Ariens et al. 87 provides the results of a regression analysis of patient-level data for patients treated with dupilumab in the placebo-controlled RCT CHRONOS and patients treated with CsA in daily practice at the Department of Dermatology and Allergology, UMC Utrecht, the Netherlands. Data for EASI 50 + ΔDLQI ≥ 4 were not available from Ariens et al. 87 and so EASI 75 became the primary outcome for this population.

Concomitant use of TCS was permitted as needed for all patients treated with CsA. CsA effectiveness presented in Ariens et al. 87 was therefore adjusted to data for dupilumab in combination with TCS from CHRONOS. Consequently, the EAG considered the appropriate comparator to be upadacitinib in combination with TCS from the AD UP trial, and abrocitinib in combination with TCS from the trials JADE COMPARE and JADE DARE. A NMA assessing these interventions versus CsA without concomitant TCS was not possible.

Time of assessment in CHRONOS, AD-UP, JADE COMPARE and JADE DARE was 16 weeks, whereas EASI scores were available for the range between weeks 12 and 16 for CsA.

The primary analysis focused on the post hoc subgroup of patients in the upadacitinib and abrocitinib trials for whom the trial intervention was their first-line systemic therapy. Of the CsA-treated patients in Ariens et al.,87 70% had no history of previous treatment with oral immunosuppressive drugs, though outcome data for this specific subgroup were not available and this cohort was therefore compared with the full population of the CHRONOS trial treated with dupilumab. Of the dupilumab-treated patients in CHRONOS, 41% had previously received systemic immunosuppressants to treat AD. A sensitivity analysis was conducted using the full trial population for AD UP, 52–58% of which had previously received systemic therapy. The difference in prior systemic therapy introduces clinical heterogeneity into the analysis, which is likely to favour upadacitinib because those with prior treatment are more severe at baseline.

The dupilumab data from CHRONOS reported in the Ariens et al. 87 analysis and in the committee papers for TA534 differ slightly; an additional two patients had a response (EASI 75) according to Ariens et al. 87 compared with TA534. The difference may be due to different handling of missing data with a LOCF analysis performed in cases of missing follow-up EASI values in Ariens et al.,87 whereas the dupilumab data from CHRONOS in TA534 either included all observed data or censored patients who received rescue medication. In the primary analysis, the data from Ariens et al. 87 (CsA vs. dupilumab) and from CHRONOS (dupilumab vs. placebo) were therefore analysed as two separate studies. However, to assess the possible impact of including the same dupilumab arm twice (once in CHRONOS and once in Ariens et al. 87), a sensitivity analysis was conducted, where the CsA arm from Ariens et al. 87 was considered an additional arm of the CHRONOS study. A different sensitivity analysis was also conducted where patients requiring rescue medication were considered non-responders. These data were available for AD UP and CHRONOS but not for the observational CsA data.

The network of trials contributing to the NMA in the first-line adult population are presented in Figure 3.

FIGURE 3.

Network plot first-line adult population, combination therapy, EASI 75. Abro, abrocitinib; Dup, dupilumab; Upa, upadacitinib.

For the NMA in the first-line adult population, the RE and FE models for the primary analysis and the sensitivity analyses were similar in terms of goodness of model fit (similar DIC), but for the primary analysis the residual deviance for the REs model was closer to the number of unconstrained data points than the FE model (see Report Supplementary Material 1).

The results of the NMA showed that abrocitinib, upadacitinib, dupilumab and CsA were all more effective than placebo, that is, leading to more responders (patients reaching EASI 75). The difference versus placebo was statistically significant for abrocitinib (both doses), upadacitinib (both doses) and dupilumab, but not for CsA, irrespective of the use of fixed or REs model. Results from the NMA were in agreement with findings from standard pair-wise analyses, in which all interventions analysed were found to be statistically significantly more effective than placebo.

Both doses of upadacitinib were shown to be more effective than CsA, with a larger OR for upadacitinib 30 mg than for upadacitinib 15 mg. The point estimates were similar for the fixed and REs models and the results were statistically significant except for the lower upadacitinib dose using the REs model. Both doses of abrocitinib were also shown to be more effective than CsA, with a dose-dependent response. However, only when using the FE model for the comparison of abrocitinib 200 mg were the results statistically significant.

Analysing the CsA data from Ariens et al. 87 and dupilumab data from CHRONOS (as reported in TA534)12 as one multiarm trial resulted in similar results to the primary analysis but with narrower CrIs. Similarly, using the full-trial population in AD UP rather than focusing on the first-line population (patients who had not received CsA) had limited impact on the results compared with the primary analysis. The sensitivity analysis using data where patients who received rescue therapy in AD UP and CHRONOS were censored also gave similar results to the primary analysis.

The placebo response in AD UP and CHRONOS were similar and thus no baseline risk adjustment sensitivity analysis was conducted for the first-line adult population.

No comparator data for CsA used without concomitant TCS were identified and therefore a NMA comparing upadacitinib and CsA used as monotherapies was not possible.

EASI 50 + ΔDLQI ≥ 4 network meta-analysis: monotherapies as second-line treatment – adult population

NMAs of interventions used as monotherapies in the second-line setting (for patients who have failed on CsA) could be carried out for both EASI 50 + ΔDLQI ≥ 4 and EASI 75, and, although results are reported for both, the primary outcome for this appraisal is EASI 50 + ΔDLQI ≥ 4. See Report Supplementary Material 1 for EASI 75 results used in scenario analyses.

Although relevant trials for baricitinib used as a monotherapy were identified, the results of these were redacted and baricitinib could, therefore, not be included in the monotherapy networks assessing EASI 50 + ΔDLQI ≥ 4 or EASI 75. For upadacitinib, the dose-finding trial reported by Guttmann-Yassky et al. in 2020 could have informed the NMA for EASI 75, as it provides the relevant outcome data at 16 weeks for upadacitinib 15 mg, 30 mg and for placebo. However, the company did not provide the relevant subgroup data for this trial at the clarification stage, and it was therefore excluded from the analysis. Data for dupilumab were informed by the post hoc subgroup data from SOLO 1 and SOLO 2 presented in TA534. For both end points in the monotherapy NMA, results for SOLO 1 and SOLO 2 were reported pooled across both studies and so has been considered as a single study, referred to as ‘SOLO CAFÉ-like’.

The primary analysis for the monotherapy NMAs was based on using all observed data, regardless of rescue medication use to determine response, with a sensitivity analysis conducted where patients requiring rescue medication were considered non-responders. Patients were not allowed rescue therapy in the abrocitinib trials.

The relevant subgroup in the abrocitinib studies, the restricted population, which was used for the primary analysis, was small. A sensitivity analysis using the generalisable population, which includes people who have had any prior systemic immunotherapy (not limited to prior CsA), was therefore also conducted. Due to variation across studies in placebo response, for both EASI 50 + ΔDLQI ≥ 4 and EASI 75, sensitivity analyses were also conducted to assess the impact of adjusting for these differences.

The network of trials contributing to the NMA of monotherapies on EASI 50 + ΔDLQI ≥ 4 in the second-line adult population is presented in Figure 4.

FIGURE 4.

Network plot second-line adult population, monotherapy, EASI 50 + ΔDLQI ≥ 4. Abro, abrocitinib; Dup, dupilumab; Tralo, tralokinumab; Upa, upadacitinib.

For the NMAs of EASI 50 + ΔDLQI ≥ 4, the RE and FE models for the primary and all sensitivity analyses were similar in terms of goodness of model fit (similar DIC) (see Report Supplementary Material 1). However, for both RE and FE models, the residual deviance showed a relatively poor prediction of the data used in each analysis, especially for the sensitivity analysis using the generalisable abrocitinib population.

The primary analysis, which focused on response irrespective of rescue therapy use, showed that treatment with any of the interventions assessed (abrocitinib, dupilumab, tralokinumab or upadacitinib) led to a statistically significant improvement in EASI 50 + ΔDLQI ≥ 4 compared with placebo. Results from the NMA were in agreement with findings from standard pair-wise analyses, in which all interventions analysed were found to be statistically significantly more effective than placebo. However, for abrocitinib 200 mg, the NMA resulted in a substantially higher OR compared with the underlying trial data.

A dose-dependent numerical benefit was observed for upadacitinib 15 mg and upadacitinib 30 mg over dupilumab; however, the differences did not reach statistical significance. Similarly, abrocitinib showed a dose-dependent effect versus dupilumab. Neither analysis showed a statistically significant difference versus dupilumab. Tralokinumab therapy resulted in a lower response than dupilumab, although, as for the other interventions, the difference was not statistically significant.

Censoring patients receiving rescue therapy in the dupilumab, tralokinumab and upadacitinib trials led to a smaller benefit of abrocitinib 200 mg, upadacitinib 15 and 30 mg compared with dupilumab, whereas the benefit of dupilumab over tralokinumab therapy increased and it also became more beneficial than abrocitinib 100 mg, though none of the relative differences between the interventions and dupilumab were statistically significant.

The sensitivity analysis based on the generalisable population for abrocitinib show relatively similar results to the primary analysis, although the ORs for both doses of abrocitinib were more favourable and the 95% CrIs were narrower for the generalisable population as the sample sizes were larger.

There was variation in placebo response across the included trials, from no responders to a third of patients on placebo being responders at 12 or 16 weeks: treatment effectiveness was captured at 12 weeks in studies evaluating abrocitinib and at 16 weeks in all other studies. However, the largest variation in placebo response rates was in the abrocitinib trials, which had very low numbers of included patients (six patients in each of the placebo arms). The sensitivity analysis adjusting for heterogeneity in placebo response gave a lower DIC than the primary, unadjusted analysis, indicating a better model fit. However, the total residual deviance, for this analysis, was lower than the number of unconstrained data points, indicating that the model may be ‘overfitting’ the data, that is, the model predicts the underlying trial data extremely well (and hence a lower DIC) but is likely to be less generalisable to the population of interest than the unadjusted analysis using observed data. The EAG is concerned that the results are unreliable for use in the cost-effectiveness analyses and, in keeping with the companies’ approach, the EAG used the observed data to inform the primary cost-effectiveness analysis.

EASI 50 + ΔDLQI ≥ 4 network meta-analysis: second-line systemic treatments in combination with topical corticosteroid – adult population

For the NMAs of interventions used in combination with TCS in the second-line setting (patients who have failed on CsA), post hoc subgroups were used for all studies apart from the dupilumab trial CAFÉ and the tralokinumab study ECZTRA 7. However, the data for dupilumab were informed by the pooled results of CAFÉ and the relevant post hoc subgroup data from CHRONOS presented in TA534. The pooled data have been considered as a single study. Two baricitinib trials were relevant for inclusion in these analyses, BREEZE-AD4 and BREEZE AD7. The majority of the results from these trials were redacted but baricitinib could be included in the analysis of EASI 75 based on data from BREEZE-AD4.

Network meta-analyses were possible to perform for both EASI 50 + ΔDLQI ≥ 4 and EASI 75, and although results are reported for both, the key outcome for this appraisal is EASI 50 + ΔDLQI ≥ 4. See Report Supplementary Material 1 for EASI 75 results used in scenario analyses.

The primary analysis for the combination therapy NMAs is based on using all observed data, regardless of rescue medication use to determine response, with a sensitivity analysis conducted where patients requiring rescue medication were considered non-responders. Sensitivity analyses were also conducted using the generalisable, rather than restricted, population for abrocitinib. A baseline risk-adjusted sensitivity analysis was conducted but the models did not converge despite attempts to increase convergence by thinning the sampling and increasing the number of model iterations.

The network of trials contributing to the NMA of combination therapies on EASI 50 + ΔDLQI ≥ 4 in the second-line adult population is presented in Figure 5.

FIGURE 5.

Network plot second-line adult population, combination therapy, EASI 50 + ΔDLQI ≥ 4. Abro, abrocitinib; Dup, dupilumab; Tralo, tralokinumab; Upa, upadacitinib.

For the NMAs of EASI 50 + ΔDLQI ≥ 4, the RE and FE models for the primary and all sensitivity analyses were similar in terms of goodness of model fit (similar DIC) and residual deviance (see Report Supplementary Material 1).

The FEs primary analysis showed that treatment with abrocitinib, dupilumab, tralokinumab or upadacitinib led to a statistically significant improvement in EASI 50 + ΔDLQI ≥ 4 compared with placebo, in agreement with findings from pair-wise meta-analyses. Using a REs model, the 95% CrIs were wider and, although treatment with any of the interventions was favoured over placebo, the results for tralokinumab did not reach statistical significance.

When compared with dupilumab, there were no comparisons that were statistically significant using the REs model. With the FEs model, the results were statistically significant for upadacitinib 30 mg, in favour of upadacitinib, and for tralokinumab, in favour of dupilumab. The OR of upadacitinib 1 mg, abrocitinib 100 and 200 mg were closer to 1, favouring dupilumab for both of the lower doses and favouring abrocitinib for the higher dose.

Censoring patients receiving rescue therapy in the dupilumab, tralokinumab and upadacitinib trials had a limited impact on the comparison of abrocitinib 100 or 200 mg versus dupilumab. However, the benefit of upadacitinib 30 mg over dupilumab was substantially smaller than for the primary analysis, the benefit of dupilumab over tralokinumab was statistically significant, and although not statistically significant, dupilumab therapy was favoured over upadacitinib 15 mg.

The sensitivity analysis based on the generalisable population for abrocitinib showed similar results to the primary analysis for upadacitinib 15 mg, 30 mg and tralokinumab; however, the benefit of dupilumab therapy over tralokinumab therapy reached statistical significance. For abrocitinib 100 and 200 mg, the direction of effect compared with dupilumab was the same as when using the restricted population (primary analysis), but the treatment effect favouring abrocitinib 200 mg over dupilumab was larger and the treatment effect favouring dupilumab over abrocitinib 100 mg was also more pronounced.

Placebo response varied between 25% and 61% in the studies contributing to the composite outcome. However, the models for the baseline risk-adjusted sensitivity analysis did not converge despite attempts to increase convergence by thinning the sampling and increasing the number of model iterations. Therefore, no results are presented for this sensitivity analysis.

EASI 75 network meta-analysis: adolescents

The companies for abrocitinib and upadacitinib are seeking a recommendation by NICE for the use of their drugs for the treatment of AD in adolescents. Of the treatment options for AD available in the NHS, baricitinib does not hold a marketing authorisation for use in adolescents and CsA is only for people aged 16 years and over and is thereby not available for a large proportion of the adolescent population. Thus, neither baricitinib nor CsA is a relevant comparator for treatment of adolescents. Additionally, although dupilumab has only been assessed and recommended by NICE for an adult population, its marketing authorisation encompasses adolescents, and dupilumab is funded for use in adolescents under the NHS England Medicines for Children Policy as part of specialised commissioning. 119 An adolescent is eligible for treatment with dupilumab if they are seen within a specialised treatment centre and they meet the criteria set out within TA534, for use of dupilumab in adults. 119 Dupilumab is therefore the key comparator for the adolescent population.

Trial evidence was available for the use of upadacitinib and abrocitinib both as monotherapies (JADE MONO 1 and 2, MEASURE UP 1 and 2) and in combination with TCS (JADE TEEN and AD UP) in the adolescent population. However, comparator data for dupilumab were only identified as a monotherapy (AD ADOL) but no trial was identified assessing dupilumab in combination with TCS in an adolescent population. A NMA assessing abrocitinib and upadacitinib versus dupilumab, all in combination with TCS, was therefore not possible, but the results of the combination therapy trials for abrocitinib and upadacitinib are presented alongside the monotherapy NMA results below for completeness.

Data on the composite outcome EASI 50 + DLQI ≥ 4 were not presented in the dupilumab trial (AD ADOL) and therefore only EASI 75 could be assessed for this population.

The abrocitinib trials did not allow use of rescue medications, whereas the dupilumab and upadacitinib trials included in the NMA for the adolescent population did. However, the dupilumab trial only reported results where patients were censored if/when they received rescue therapy. So, the primary analysis for the NMA in the adolescent population is based on upadacitinib data where patients were included in the analysis even if they required rescue medication, dupilumab data where patients were censored when they received rescue medication and abrocitinib data where patients did not receive rescue medication. A sensitivity analysis was conducted using upadacitinib data where patients were censored when receiving rescue medication, similar to the dupilumab study.

Due to variation in the placebo response between the trials included in the analysis, a sensitivity analysis adjusting for heterogeneity in placebo response was also conducted.

The NMA results are focused on the doses of the interventions recommended for adolescents, which are abrocitinib 100 and 200 mg and upadacitinib 15 mg. The network of trials contributing to the NMA in the adolescent population are presented in Figure 6.

FIGURE 6.

Network plot adolescent population, monotherapy, EASI 75. Abro, abrocitinib; Dup, dupilumab; Upa, upadacitinib.

For the NMA in the adolescent population, the FE and RE models of the primary and sensitivity analyses were similar in terms of goodness of model fit (similar DIC, see Report Supplementary Material 1), but the residual deviance for the RE models were closer to the number of unconstrained data points in all analyses, which reinforces the EAG’s preference for the RE model.

The primary analysis shows that treatment with abrocitinib (either dose), dupilumab or upadacitinib 15 mg was associated with a statistically significant improvement in EASI 75 compared with placebo. Results from the NMA were in agreement with findings from standard pair-wise analyses for each of the trial, in which all the interventions analysed were found to be statistically significantly more effective than placebo. However, for abrocitinib 200 and 100 mg, the NMA resulted in substantially higher ORs compared with the underlying trial data.

When compared with dupilumab, treatments with abrocitinib 200 mg, abrocitinib 100 mg and upadacitinib 15 mg were all more effective than dupilumab, although these comparisons were not statistically significant.

Censoring patients receiving rescue therapy in the dupilumab and upadacitinib trials led to relatively similar ORs compared with the primary analysis and no statistically significant difference compared with dupilumab.

Placebo response varied from 0% to 22% in the studies contributing to the analysis for the adolescent population. The DIC for the sensitivity analysis adjusting for heterogeneity in placebo response was markedly lower than the DIC for the primary analysis, indicating a better fitting model. However, similar to the sensitivity analysis conducted for the second-line adult monotherapy analysis, residual deviance for this NMA was lower than the number of data points used in the analysis, potentially indicating ‘overfitting’ to the underlying data used in the model. As such, due to concerns around generalisability of the results of the sensitivity analysis, the NMAs based on the observed data were preferred to inform the cost-effectiveness analyses.

No data were identified on dupilumab used with concomitant TCS in an adolescent population. Therefore, a NMA comparing upadacitinib and abrocitinib with dupilumab, all in combination with TCS, was not possible. However, data on the efficacy of upadacitinib and abrocitinib in combination with TCS and compared with placebo in adolescents were available. The results showed that both upadacitinib and abrocitinib with TCS are statistically significantly more effective than placebo with TCS, in terms of EASI 75. However, the relative benefit over placebo was substantially smaller when the interventions were used in combination with TCS than as monotherapies.

Safety

The safety of the interventions and comparators during treatment up to 16 weeks are reported for the full-trial populations and not separated by line of therapy. The EAG has focused on serious adverse events (SAEs) and on specific AEs (irrespective of severity) in line with those included in TA534 and TA681; injection-site reaction (for dupilumab and tralokinumab), conjunctivitis and allergic conjunctivitis, upper respiratory tract infections, acne and oral herpes.

These are not necessarily the most common AEs for each of the treatments, but each has been found to be associated with at least one of the treatments and were considered the most important to include by the EAG’s clinical experts.

In terms of both SAEs and the specific AEs of any severity, the numbers were generally small, indicating that the short-term use of these interventions, as monotherapy and in combination with TCS, was well tolerated. The results are therefore not discussed separately for monotherapy and combination therapy, or for adults and adolescents. AE data from individual studies are provided in Report Supplementary Material 1.

Populations

As described in the section Decision problem, the population relevant to the MTA are those with moderate-to-severe AD, irrespective of previous treatment and of age, with a subgroup of interest being those for whom systemic therapies have been inadequately effective, not tolerated or contraindicated.

As per guidance from NICE, the EAG has considered each of the companies proposed positions in the treatment pathway as part of the assessment for the CEA. Thus, clinical data in the NMA were analysed by pre-specified subgroups based on the age and line of therapy and TCS use (see Methods of data synthesis). Consequently, the modelled populations are as follows:

• Adults who are eligible for systemic treatment (CsA) on inadequate response to topical treatments (referred to hereafter as the adult first-line systemic treatment population).

• Adults who achieve inadequate response to, cannot tolerate or are contraindicated to CsA – monotherapy and combination therapy (referred to hereafter as the adult second-line systemic treatment population).

• Adolescents, irrespective of prior therapy.

Trials assessing AD treatments in the adolescent population included a mix of patients at all lines of systemic treatment (see Method for reviewing effectiveness). Due to small numbers in the trials that include adolescents, a robust subgroup analyses by line of treatment for this population is not possible; therefore, adolescents are considered in the model irrespective of prior line of therapy. Furthermore, due to limitations in the clinical evidence, combination therapy data are unavailable for the adolescent population and monotherapy data are unavailable for the adult first-line systemic treatment population (see Assessment of clinical effectiveness). Further details on available treatment effectiveness data, interventions and proposed positions in the treatment pathway for the CEA are given in the sections Treatment effectiveness and Interventions and comparators, respectively.

Interventions and comparators

The interventions of interest as part of this MTA are abrocitinib, tralokinumab and upadacitinib. As per the final protocol,48 the EAG has considered each of the companies proposed position in the treatment pathway as part of the CEA. As such, the interventions are compared with the recommended treatment in the proposed position of the treatment pathway, outlined in Table 8. The recommended first-line systemic treatment for adults is CsA but does not hold a marketing authorisation for use in adolescents. However, dupilumab is funded for use in adolescents under the NHS England Medicines for Children Policy and is included as the comparator for this population. Recommended second-line systemic treatments for adults include dupilumab and baricitinib.

Each of the treatments is considered as a monotherapy and in combination with TCS in the cost-effectiveness analyses. However, as described in the section Method for reviewing effectiveness, results are only available from the NMA for combination therapy for the adult first-line systemic treatment population, while results for monotherapy are available for the adolescent population and for baricitinib for the adult second-line systemic treatment population (see Table 8).

Table 9 presents an overview of the treatment regimens and the EAG approach to inclusion of the different treatment regimens in the model.

Model structure

For the CEA, a hybrid economic model was developed comprising a short-term (1 year) decision tree component, to capture the treatment induction phase and treatment response assessments, followed by a long-term (lifetime), three-state MM. The comparator in the analysis for first-line systemic treatment is CsA and for second-line systemic treatment, comparators are dupilumab and baricitinib. All treatments in the model are evaluated as both monotherapies and in combination with TCS. The development of the model structure was informed by published models identified in the SLR, supplied company submissions and the approaches accepted for TA53412 and TA681. 13 Additionally, the EAG has developed two short-term decision tree models to reflect the differences in first- and second-line systemic treatments. Each component of the model is discussed in turn below.

Short-term decision tree – first-line systemic treatment (adults)

Abrocitinib and upadacitinib are the only treatments that have a proposed position in the first-line systemic treatment pathway and as such will be compared against CsA, which is only recommended for a maximum of 1 year, after which all patients discontinue to BSC.

Figure 7 presents the model schematic for first-line systemic treatment for adults. All patients enter the short-term model, starting either abrocitinib/upadacitinib or CsA (all in combination with TCS) and remain on treatment for 16 weeks (treatment induction phase). At week 16, treatment response is assessed, defined as achieving EASI 50 + DLQI ≥ 4 and responders remain on treatment until week 52. Non-responders at week 16 discontinue treatment and receive BSC.

FIGURE 7.

First-line systemic treatment short-term decision tree model structure – adults.

Between week 16 and 52, responders may lose response to treatment or discontinue treatment for other reasons and will enter the long-term MM in the BSC health state. Abrocitinib/upadacitinib responders who sustain their response between week 16 and 52 and are still on treatment enter the long-term MM in the maintenance health state. For patients on CsA, the maximum recommended treatment duration is 12 months; thus, they discontinue to BSC in the long-term MM. Sustained response for abrocitinib/upadacitinib at week 52 is based on conditional discontinuation data, defined as the proportion of patients discontinuing treatment at week 52 from those who achieve response at week 16.

In the short-term model, the BSC health state is composed of responders and non-responders and these proportions are informed by week 16 data response data (see Method for reviewing effectiveness and Treatment effectiveness). This approach was accepted in TA681 as an appropriate way to capture the waxing and waning nature of response to BSC treatment.

Short-term decision tree – second-line systemic treatment (adults)/adolescents (all lines of treatment)

Figure 8 presents the model schematic for second-line systemic treatment (adults) and adolescents (refer to the section Populations for further details about line of therapy for adolescents). Adult patients enter the short-term model starting treatment on one of the five second-line systemic treatments (abrocitinib, baricitinib, dupilumab, tralokinumab or upadacitinib). Adolescent patients enter the model on abrocitinib, dupilumab or upadacitinib. Patients remain on treatment for 16 weeks (treatment induction phase), after which point treatment response is assessed, defined as achieving EASI 50 + (C)DLQI ≥ 4. Responders at week 16 remain on treatment until week 52. Non-responders at week 16 discontinue treatment and receive BSC.

FIGURE 8.

Second-line systemic treatment short-term decision tree model structure (adults)/adolescents (all lines of treatment).

Between weeks 16 and 52, responders may lose response to treatment or discontinue treatment for other reasons and will enter the long-term MM in the BSC health state. Responders who sustain their response between weeks 16 and 52 and are still on treatment enter the long-term MM in the maintenance health state. Sustained response at week 52 is based on conditional discontinuation data, defined as the proportion of patients discontinuing treatment at week 52 from those who achieve response at week 16.

In the short-term model, the BSC health state is composed of responders and non-responders and these proportions are informed by week 16 data response data (see Method for reviewing effectiveness and Treatment effectiveness). As mentioned previously, this approach was accepted in TA681 as an appropriate way to capture the waxing and waning nature of response to BSC treatment.

Long-term Markov model

At the end of each of the short-term decision trees (start of year 2), patients enter a long-term three-health state MM. Health states in the model include maintenance, BSC and death (Figure 9).

FIGURE 9.

Long-term MM schematic.

Patients who have maintained a response at week 52 and still on treatment enter the MM in the maintenance health state and remain there until loss of response (via treatment waning) or if they discontinue treatment for any reason (all cause discontinuation) whereupon they transition to the BSC health state.

Patients who have discontinued treatment to BSC in the short-term decision tree enter the MM in the BSC health state and remain there until death. As with the BSC health state in the short-term decision tree model, the MM BSC health state is composed of responders and non-responders and these proportions are informed by week 16 data response data (see Method for reviewing effectiveness and Treatment effectiveness), in line with approach accepted in TA681. 13

At any time in the model, patients can transition to the death state. As treatment for AD is not expected to affect mortality, transitions to the death state are informed by general population mortality rates.

In the long-term model, an annual cycle length has been implemented and half-cycle correction applied.

Time horizon, perspective and discounting

The time horizon of the model is lifetime (up to a maximum age of 100 years). The perspective of the analysis is the NHS in England. Costs and QALYs have been discounted at 3.5%, as per the NICE reference case. Scenarios were conducted limiting the time horizon to 5 years after the mean age and 75 years of age for the adult analyses and 18 years of age for the adolescent analyses.

Treatment effectiveness

The primary treatment outcome assessed in the MTA model is treatment response at week 16, defined using a composite outcome of EASI 50 + DLQI ≥ 4. In TA53412 and TA681,13 the composite outcome was preferred by the committee as it was deemed to be sensitive to changes in treatment outcomes and more clinically relevant than EASI 75. As a result, the composite outcome was carried forward in each of the company models submitted as part of the MTA.

Log ORs from the NMA (see Method for reviewing effectiveness) were used to estimate week 16 treatment response probabilities used in the model. However, as noted in the section Quantity and quality of research available, there were several data limitations in the NMA that meant outcome data (composite or EASI 75) were unavailable for some of the populations. Table 10 presents an overview of the NMA outcome data available for each population.

The composite outcome from the NMA was obtained for the adult second-line systemic subgroup for both monotherapy and combination therapy analyses and EASI 75 was explored in scenario analyses. However, the EAG was unable to obtain composite outcome data for baricitinib as these data are redacted in TA681 (and the company declined to provide them to the EAG). Furthermore, EASI 75 data for baricitinib were only available as combination therapy for the adult second-line systemic population. Therefore, any comparisons with baricitinib in the adult second-line systemic treatment subgroup are presented as scenario analyses.

The composite outcome could not be obtained for treatments relevant to the adolescent and adult first-line systemic treatment subgroups for both monotherapy and combination therapy analyses. However, the EASI 75 outcome was available for the adolescent subgroup for the monotherapy analyses and adult first-line systemic treatment subgroup for the combination analyses, and as such this outcome is used for the base case in these populations.

The committee for TA534 and TA681 considered that in clinical practice, dupilumab and baricitinib will likely be used as combination therapies rather than monotherapies and this view is reflected by the EAG’s clinical experts when considering the three new treatments under consideration. Thus, for the adult first-line systemic treatment subgroup, the combination analyses are likely to be more relevant than the monotherapy analyses. Furthermore, upadacitinib is the only treatment where the company has proposed its use in the adult first-line systemic treatment subgroup. For the adolescent population, the EAG’s clinical experts explained that they would be treated in the same way as adults; thus, combination therapy is more relevant than monotherapy.

As the combination therapy analyses are more relevant for clinical practice, the EAG considered that missing monotherapy data are not critical for decision-making for the adult first-line systemic treatment subgroup. However, for the adolescent population, monotherapy analyses may potentially underestimate the effectiveness of the treatments when used in combination with TCS in clinical practice and may reflect a conservative view of cost effectiveness. The EAG has included treatment response outcomes one-way sensitivity analyses to capture the uncertainty around the estimates, but the upper bound estimate for the adolescent population may be useful to help understand what the cost effectiveness of combination therapy might be for this population.

Treatment waning

Over time, patients may lose response to treatment, whether on active treatment with biologics or BSC. In TA534, assumptions around treatment effect waning were included in the company’s economic model and accepted by the committee. The TA534 final appraisal document (FAD) states:

In the dupilumab maintenance state, the company assumed that 2% of the benefit would be lost in year 2, 5% in year 3, 7% in year 4, and 8% in year 5 and beyond. It used these estimates to adjust down the proportion of people who continued to have dupilumab (that is, those who lost the benefit of dupilumab moved to the best supportive care state and then accrued the utility associated with that state).

In the company submissions for abrocitinib, tralokinumab and upadacitinib and in TA681, treatment-waning assumptions were influenced by the approach in TA534 due to a lack of data in the key trials for the drugs. However, the assumptions and implementation of active treatment waning in TA534 have been interpreted in various ways by the companies for baricitinib (TA681), abrocitinib, tralokinumab and upadacitinib.

Active treatment-waning proportions in the abrocitinib and upadacitinib models were taken from TA534. In the abrocitinib model, by year 5 and beyond, 8% of patients on active treatment would lose response. However, between years 2 and 5 in the upadacitinib model, up to 8% of patients experienced treatment waning and from year 6 up to year 10, a further 1% per year were assumed to lose response, with no further waning beyond year 10. For the tralokinumab base case, it was assumed that between 2% and 3% of patients would lose treatment response annually up to year 4, with 1% losing treatment response annually from year 5 onwards.

The implementation of active treatment waning in the tralokinumab and upadacitinib models was similar to TA534, with patients discontinuing to BSC upon treatment waning. However, in the upadacitinib model, when patients on active treatment move to the BSC health state upon treatment waning, they first incur utility of BSC non-responders and then gradually return to the baseline utility following BSC non-responders’ treatment-waning rates (see Report Supplementary Material 1). In TA681 and the abrocitinib model, patients on active treatment were assumed to incur a utility loss. In TA681, patients returned to baseline utility upon treatment waning and in the abrocitinib model, non-responder utilities were applied to patients who lost response.

For the MTA model, the EAG has adopted the active treatment-waning approach accepted in TA534. Specifically, the EAG has assumed that in years 2, 3, 4 and 5 onwards, 2%, 5%, 7% and 8% of patients, respectively, will lose response to active treatment and discontinue to BSC. As soon as patients no longer achieve EASI 50 + DLQI ≥ 4, they are considered non-responders. The EAG acknowledges that there may be overlap between the proportion of patients losing response to treatment and long-term all-cause treatment discontinuation, as lack of efficacy is included as a reason to stop treatment. However, the size of the overlap between treatment waning and all-cause discontinuation is unknown and as such the EAG approach can be considered conservative. The EAG included long-term all-cause treatment discontinuation and treatment-waning proportions in one-way sensitivity analysis (OWSA) to determine if these parameters are key drivers of cost effectiveness (see Results). Furthermore, the EAG included a scenario analysis where no active treatment-effect waning was assumed.

Unlike TA534, TA681 and the company economic models, BSC is not a comparator in the EAG model but a single health state that patients transition to due to lack or loss of response to treatment or treatment discontinuation for other reasons. As described in the section Model structure, BSC is modelled as a weighted average of responders and non-responders to BSC to reflect the waxing and waning nature of AD and thus captures treatment effectiveness fluctuations. As such, the EAG has assumed no additional treatment waning for the BSC health state, which is in line with the ERG’s preferred approach in TA681 (see Report Supplementary Material 1). The committee for TA681 considered that the ERG’s approach represented different patients moving in and out of disease control over time, but treatment waning would be between the ERG’s approach (no waning, BSC modelled as a single health state of 50% responders and 50% non-responders) and the company’s approach based on TA534 (up to 97% of BSC patients lose response). However, the committee for TA681 did not give further direction to handle treatment waning in the BSC health state. To explore the uncertainty around BSC waning, baseline placebo response has been included in the OWSA (which informs the BSC health state), as well as scenarios exploring shorter time horizons.

Further information on the BSC treatment-waning assumptions adopted in TA534 and TA681, and assumptions used in the company models can be found in Report Supplementary Material 1.

Mortality

Treatments for moderate-to-severe AD are not expected to affect mortality. Thus, the EAG has used Office of National Statistics (ONS) National Life Tables for England and Wales to estimate age-adjusted all-cause mortality in the economic model.

Adverse events

Adverse events included in the EAG’s economic model are in line with those included in TA534 and TA681 and were considered the most important to include by the EAG’s clinical experts. In TA534, the most frequent and serious AEs reported in the dupilumab trials were included and these were injection-site reaction, allergic conjunctivitis, infectious conjunctivitis and oral herpes. TA681 also included the most frequent and serious AEs reported in the baricitinib trials, as well as those from TA534, with the only addition to the included AEs being upper respiratory tract infection. The company for tralokinumab only included the TA534 AEs in their economic model. In the abrocitinib and upadacitinib models, AEs with an incidence of > 5% in the intervention trials, dupilumab trials, TA534 and TA681 were included but no detail was provided for the severity of the AEs. Furthermore, the company for upadacitinib excluded oral herpes from included AEs as clinical advice suggested that patients with oral herpes would self-medicate with over-the-counter medication. However, in TA534, the cost associated with oral herpes was for a GP visit and the EAG considers it should be included in the CEA. Refer to Report Supplementary Material 1 for a comparison of AEs included in TA534, TA681 and the company models.

The EAG’s approach to AEs is generally in line with TA534, TA681 and the company models and includes serious AEs with an incidence of > 5% in any treatment arm. The EAG reviewed and extracted data on AEs from publications included in the clinical SLR, company submissions and appendices, company clarification responses and committee papers from TA534 and TA681. The available clinical study reports (CSRs) for abrocitinib, tralokinumab and upadacitinib studies were also searched for the AEs of interest where there were missing data. As the EAG included specific AEs that may not be applicable or captured for all treatments in the model, a NMA on individual AEs was not deemed to be appropriate.

The AEs included in the EAG’s economic model are injection-site reaction, allergic conjunctivitis, infectious conjunctivitis, oral herpes, upper respiratory tract infection and acne. In the upadacitinib model, skin infections were included, and in the abrocitinib model, folliculitis, headache, nausea, pharyngitis and nasopharyngitis were included, but these were excluded from the EAG model as the severity of these events could not be determined. These AEs are easily treated, and a cost to the NHS is rarely incurred. The EAG considered that the definition of infectious conjunctivitis in the companies’ models is based on the Medical Dictionary for Regulatory Activities (MedDRA) term ‘conjunctivitis’ (system organ class of infections and infestations), which is likely to reflect infectious conjunctivitis. To maintain consistency across the analysis, the EAG extracted data for conjunctivitis to inform the AE rate for infectious conjunctivitis.

Data on AEs for CsA were unavailable. In TA534, the committee accepted assuming zero AEs for CsA in the short term as treatment is only given for 1 year before patients move to BSC and this assumption has been carried forward in the EAG economic model.

Adverse events for BSC were based on placebo data from AD-UP. As mentioned in the section Baseline characteristics, baseline characteristics from the upadacitinib trials were deemed to be representative of the population who would be treated in clinical practice in England according to the EAG’s clinical experts. The EAG has assumed that the BSC AE rate for monotherapy is the same as combination therapy because in clinical practice BSC includes TCS.

Table 12 presents the 16-week AE rates included in the EAG economic model. Weekly and annual rates of AEs were calculated from the 16-week data for the short- and long-term model. The rates of AEs were used to estimate the costs to treat an AE only as it was assumed that the HRQoL data collected in the trials and used in the model would capture the acute impact of AEs. The EAG’s approach is in line with TA534 and TA681, as well as the approach adopted in the company models for abrocitinib, tralokinumab and upadacitinib. Refer to the section Costs of managing adverse events for AE costs used in the economic model.

Flares

During treatment for moderate-to-severe AD, patients may experience acute exacerbations of symptoms, called flares. The rate of flare can vary depending on the treatment received by a patient but treatments for flare are similar.

Flare rate was not an end point in the key studies described in the section Quantity and quality of research available. In TA534 and TA681, the receipt of rescue medication was accepted as a proxy for flare. Furthermore, the companies for tralokinumab and upadacitinib used receipt of rescue medications from their key trials to inform the rate of flare used in their economic models. However, the company for abrocitinib had data on protocol-defined flares from REGIMEN and this was used to inform their economic model.

The EAG requested 16-week data on the receipt of rescue medication (used as a proxy for flare, in line with TA534 and TA681) or rate of flare. Each company supplied the requested data (Table 13) and the data have been used in the EAG’s economic model. The flare rate for dupilumab was extracted from TA534 and data for baricitinib were obtained from Reich et al. 97 Similar to AEs, the flare rate for BSC was based on placebo data from AD-UP. However, unlike the data for AEs, flare data are split by first- and second-line systemic treatment. As mentioned in the section Baseline characteristics, baseline characteristics from the upadacitinib trials were deemed to be representative of the population who would be treated in clinical practice in England according to the EAG’s clinical experts. The EAG has assumed the BSC flare rate for monotherapy is the same as combination therapy because in clinical practice BSC includes TCS. Furthermore, due to a paucity of data, the EAG assumed flare rates for CsA were the same as upadacitinib.

For abrocitinib, receipt of rescue medication in JADE COMPARE, MONO 1 and MONO 2 was prohibited. However, protocol-defined flare data at 40 weeks from REGIMEN were available for abrocitinib (reported in the company submission) and the data were used to inform the annual rate of flare for both adults (second-line systemic treatment) and adolescents for monotherapy and combination therapy, in line with the company’s preferred approach. Additionally, the annual rate of flare for abrocitinib was used to calculate a weekly rate of flare to be used in the short-term part of the economic model.

The rate of flare was applied in the short-term model by converting the available treatment-specific data into weekly rates. For the long-term model, annual rates of flare were estimated from the 16-week data, except for dupilumab and abrocitinib where annual flare data were already available. Flare rates were used to estimate the costs to treat a flare only as it was assumed that the HRQoL data collected in the trials and used in the model would capture the acute impact of flares. The EAG’s approach is in line with TA534 and TA681, as well as the approach adopted in the company models for abrocitinib, tralokinumab and upadacitinib. Refer to the section Costs of managing flares for flare costs used in the economic model.

Health-related quality of life

For each of the drugs considered in the MTA, their key trials collected EQ-5D-5L data, which in the companies’ submissions were mapped to the EQ-5D-3L using the van Hout cross-walk algorithm,144 as per the NICE reference case. 145 Each of the companies developed regression models to estimate utility values to be used in their models. They also estimated baseline, responder and non-responder values, according to subgroup and measure of response. The companies for abrocitinib and tralokinumab estimated utility values by type of therapy (combination therapy or monotherapy). Disutility values associated with AEs and flares were not included in the companies’ models as per TA534 and TA681.

Treatment-specific utility values were adopted by the companies for abrocitinib and tralokinumab, as this approach was accepted in TA534 and TA681. Furthermore, in TA534, it was accepted that non-responders on dupilumab accrued the average utility of a dupilumab non-responder and BSC non-responder (0.82) at week 16 after starting treatment, and after week 52 accrued the utility value of BSC non-responders (0.77). The TA534 approach was adopted by the companies for abrocitinib and tralokinumab. However, in the abrocitinib model, utility values were used to capture the benefit of early response to systemic treatment prior to the week-16 assessment point. In the upadacitinib model, utility values were not treatment specific; thus, no weighting of systemic treatment non-responder and BSC non-responder values was implemented. However, the utility values were used to capture the benefit of early response to upadacitinib treatment prior to the week-16 assessment point.

Due to the variation in the approach to utilities across the companies’ models, TA681 and TA534, a consistent and conservative approach to utilities for all treatments is adopted in the MTA model. It is worth noting that unlike in TA534, TA681 and the companies’ models, BSC is not a comparator in the EAG’s model. As mentioned previously, BSC is modelled as an average of responders and non-responders to BSC treatment to capture the waxing and waning nature of response. The implication of BSC as a comparator in the previous models is that it was reasonable to assume that being on systemic treatment but not achieving response was still likely to result in an improvement in HRQoL over and above BSC in the short term, hence the assumption being accepted in TA534.

However, as the three new drugs are only being compared against currently available systemic treatments in the MTA model and BSC is only a health state and not an independent comparator, the EAG took a conservative approach to use treatment-specific baseline utilities for weeks 0–16, treatment-specific responder utility values for those who achieve and maintain response to treatment and the weighted average utility of responders and non-responders to BSC for non-responders to treatment and those who discontinue treatment. As such, the benefit of systemic treatment remains only for those who respond to treatment. Given the remit of the MTA and the approach to the BSC, the EAG considers this a necessary deviation from the approach accepted in TA534 and can be considered conservative.

The EAG investigated the utility regressions and resulting utility values from each of the company submissions to assess the suitability of the data for the MTA model but found that that definition of the populations used to estimate utilities in each of the companies’ models was not aligned with definitions used in the EAG analysis (see Methods of data synthesis and Populations). Furthermore, monotherapy- and combination therapy-specific utility values were not available in the upadacitinib company submission. Thus, from each of the companies, the EAG requested HSUVs based on the relevant subgroups from the key trials, by response category (composite and EASI 75), dose (where applicable) and type of therapy (monotherapy and combination therapy). Additionally, the EAG requested data on the number of observations at each assessment point to gauge the size of the data sets informing the utility regressions and aid choice of which utilities should inform the drug class estimates.

All three companies provided the requested utility data to the EAG in time for the development of the economic model. The utility data provided by the company for abrocitinib was subject to several issues which made the suitability for use in the MTA model limited. Utility data for abrocitinib in the adolescent population using the EASI 75 measure of response were not provided, which is relevant for the adolescent analyses. The company did provide utility data for adult second-line systemic population for both the composite and EASI 75 outcomes and monotherapy and combination therapy. The abrocitinib utility analyses use data from the full-trial populations of the relevant JADE trials and apply the baseline characteristics of relevant populations (generalisable and restricted) to generate utility values. However, as mentioned in the section Quantity of research available, a fundamental issue with the JADE trial programme is that most of the trial populations are not relevant to the decision problem (patients are predominantly naïve to systemic treatment). Thus, the patient numbers informing the post hoc subgroups that are relevant to this appraisal are small and may potentially result in unreliable estimates of utility.

The companies for tralokinumab and upadacitinib provided complete data for the populations that are relevant to proposed position in the pathway for their drugs. The utilities provided warranted further examination by the EAG considering other available data.

As part of the HRQoL SLR (see Results – Health-related quality-of-life), the EAG extracted utility data from TA534 and TA681 for dupilumab and baricitinib. The baricitinib values were available for the composite outcome and combination therapy for the adult second-line systemic treatment, but in the MTA analyses, outcome data for baricitinib are only available for EASI 75. Additionally, the committee for TA681 concluded that, given the flaws with the company’s utility values, the utility values from TA534 were preferable. From TA534, utility values for dupilumab were available for monotherapy and combination therapy for the adult second-line systemic treatment population, but no data were available for adolescents.

As such, to account for limitations associated with missing data, uncertainty due to small numbers and relevance of the populations for utility values, the EAG decided to adopt a drug class approach for utility values in the model. The drug class approach was considered to be appropriate as the drugs considered in the economic model fall into two classes: JAK inhibitors (abrocitinib, baricitinib and upadacitinib); and monoclonal antibodies (dupilumab and tralokinumab). The EAG considers that HRQoL is unlikely to be affected by different treatments within a drug class but there could be differences across different drug classes due to the mode of action and administration of treatment. Furthermore, as JAK inhibitors are associated with both high and low doses, type of dose given may be indicative of severity of AD. As such, company’s utility values for tralokinumab were used to inform the base case for the monoclonal antibody drug class and upadacitinib utility values were used for the JAK inhibitors. Furthermore, the EAG employed a simplification for JAK inhibitors, using available high and low dose and mapping values to high- and low-dose treatments. Notably, the analyses that include baricitinib are only considered in a scenario and not the EAG base case due to lack of the composite outcome for baricitinib.

With regard to CsA, utility values were assumed to be the same as upadacitinib 15 or 30 mg (depending on the comparison). In the adolescent population, as there are no monoclonal antibody utility values, the EAG assumed the tralokinumab monotherapy adult second-line systemic treatment utility values for dupilumab monotherapy.

The weighted utility values for BSC responders and non-responders were based on the upadacitinib placebo utility values for the relevant population as baseline characteristics and BSC response status reflect the upadacitinib trials. Refer to Report Supplementary Material 1 for an overview of the companies’ regression models used to estimate the utility values used in the EAG base-case analysis. The EAG also explored a scenario where utility values for TA534 (presented in Report Supplementary Material 1) were used for all the populations.

Disutilities associated with AEs have not been included in the EAG’s economic model in line with TA534, TA681 and the companies’ models. It has been assumed that due to the frequency of capturing EQ-5D data in the upadacitinib and tralokinumab trials, the impact of AEs on HRQoL will be captured in the data.

Utility values in the model are adjusted for age based on UK population norms using the multiplicative method detailed in Ara and Brazier 2010. 146

Resource use and costs

The following cost categories are included in the model:

• drug acquisition costs (see Drug acquisition costs)

• drug administration costs (see Drug acquisition costs)

• concomitant medication costs (see Concomitant medication costs)

• healthcare resource use costs (monitoring costs) [see Healthcare resource use costs (monitoring costs)]

• costs of managing flares (see Costs of managing flares)

• costs of managing AEs (see Costs of managing adverse events).

The economic analysis is conducted from an NHS and personal social services perspective and therefore only includes costs that would be incurred by the NHS and personal social services. Costs are reported in pound sterling for a 2019–20 cost year. Drug costs have been sourced from the British National Formulary (BNF) and electronic drug marketing tool (eMIT), while service costs have been sourced from the National Schedule of NHS Costs and Personal Social Services Research Unit (PSSRU).

List of assumptions

The EAG base-case assumptions are summarised in Table 25.

Deterministic results

List price ICERs are presented in Table 26 for the adult first-line systemic treatment, Table 27 for the adult second-line systemic treatment and in Table 28 for the adolescent populations. As abrocitinib and upadacitinib have different doses, the EAG has ordered these first in the presentation of results. The EAG notes that incremental QALYs were relatively small and incremental costs were relatively large for each treatment in each population resulting in the sensitive ICERs.

Probabilistic results

The EAG conducted a probabilistic sensitivity analysis (PSA) to assess the impact of the combined uncertainty from all parameters in the model. This was performed by sampling from distributions of the uncertain parameters 1000 times, to generate the equivalent number of sampled ICERs. Conditional discontinuation data, utility values, AEs and flare rates were varied using a beta distribution. Costs were varied using a gamma distribution. Variation for the week-16 treatment response was based on 1000 CODA samples from the NMA (see Methods of data synthesis for NMA methods).

List price probabilistic ICERs are presented in Table 29 for the adult first-line systemic treatment, Table 30 for the adult second-line systemic treatment and in Table 31 for the adolescent populations. For the cost-effectiveness acceptability curves (CEACs), refer to the Report Supplementary Material 1 document. It should be noted that for each population and intervention, PSAs were run separately due to the structure of the model and therefore the sampling from parameter distributions for the comparator provides slightly different mean estimates for each pairwise comparison. However, total costs and QALYs for the comparator are similar for the PSA results. Additionally, the EAG notes that incremental QALYs were relatively small and incremental costs were relatively large for each treatment in each population resulting in the sensitive ICERs.

One-way sensitivity analysis

One-way sensitivity analysis was conducted by varying key model parameters between the upper and lower values of the expected value used in the deterministic base case. The key model parameters include:

• week-16 response

• conditional discontinuation (used to inform the week-52 response and annual discontinuation)

• treatment waning

• utility values

• monitoring/healthcare resource use (frequency)

• AEs (frequency)

• flares (frequency)

• age.

The response at week 16 was varied for each treatment using the 95% CrI estimated by the NMA. The response to BSC at week 16 was also varied, as this parameter is used to weight costs and utilities when patients discontinue active treatment and transition to BSC.

Conditional discontinuation rates, AE rates and flare rates were varied individually for each treatment by their 95% confidence intervals (CIs). No estimates of precision were available for treatment waning, utility or monitoring/healthcare resource use, and therefore the SE was assumed to equal ±20% of the mean value. Utility values were varied individually for each treatment, while treatment waning and monitoring/healthcare resource use parameters were varied simultaneously for the intervention and comparator. Age data (including variation) for the post hoc subgroups were presented by treatment arm rather than an overall mean. As such, the EAG calculated a mean age for each subgroup (presented in the section Baseline characteristics) and this was varied by ± 20% of the mean value. Drug acquisition costs and service costs were not varied as these are assumed to be fixed values. Alternative discount rates and time horizons were explored in scenario analysis (see Scenario analyses).

The below paragraph summarises the results of the OWSA for the top two or three parameters for the adult first-line systemic treatment population, adult second-line systemic treatment population and adolescents. For tornado diagrams of the top 10 parameters, see Report Supplementary Material 1.

The key drivers in the adult first-line systemic treatment population are the week-16 response for BSC and the week-52 response for upadacitinib 15 and 30 mg. As noted in the section Week-52 treatment response outcomes, the week-52 response represents all-cause discontinuation for people whose condition responded to treatment at week 16 but withdrew from treatment at week 52. Additionally, the upadacitinib week-52 response is used to inform the abrocitinib week-52 response. The key drivers in the adult second-line systemic monotherapy and combination therapy populations also include the week-16 response and the week-52 response. In addition to response at weeks 16 and 52, the key driver for the adolescent population also included the BSC non-responder HSUV, and lower non-responder BSC utility values favour dupilumab. As noted in the section Health-related quality-of-life, the overall BSC utility value is derived by weighting the BSC non-responder and responder values by the proportion of responders to BSC.

Scenario analyses

Model validation

A senior health economist was responsible for the specification and development of the MTA model. A principal health economist was responsible for validating model assumptions and performing a detailed quality assurance of the MTA model. A third health economist, not involved in the MTA project, performed an independent review of the MTA model, including face validity checks and black and white box testing of the model.

The EAG’s clinical experts were involved with validating key assumptions in the model to ensure clinical validity of model inputs and outputs as well as peer review of the report.

Summary of key results

The purpose of this MTA was to assess the cost effectiveness of abrocitinib, tralokinumab and upadacitinib individually as monotherapy and in combination with TCS for treatment of moderate-to-severe AD. In the MTA, as requested by NICE, the cost effectiveness of abrocitinib, tralokinumab and upadacitinib has been evaluated for the proposed position in the treatment pathway for moderate-to-severe AD as presented by the companies in their submissions to the STA process. All results shown in this report are based on list prices for all drugs.

The companies for abrocitinib and upadacitinib have a proposed position for the entire indication, that is for adolescents irrespective of prior treatment and adults as both first- and second-line systemic treatment. Upadacitinib is available in two doses, 15 and 30 mg. Upadacitinib 15 mg is approved for use in adolescents and both doses are approved for adults. In the adolescent population, upadacitinib 15 mg dominates dupilumab. In the adult first-line systemic treatment population, compared with CsA with TCS, upadacitinib 15 and 30 mg in combination with TCS are associated with probabilistic ICERS of £80,342 and £152,134. In the adult second-line systemic treatment population, upadacitinib 15 mg as monotherapy dominates dupilumab and in combination with TCS is less costly and less effective than dupilumab (south-west quadrant ICER of £185,877). Upadacitinib 30 mg as monotherapy and in combination with TCS is more expensive and more effective than dupilumab, with probabilistic ICERs of £66,539 and £107,339, respectively, for the adult second-line systemic treatment population.

Abrocitinib is available in two doses, 100 and 200 mg, with both approved for use in adolescents and adults. In the adolescent population, abrocitinib 100 and 200 mg dominate dupilumab. In the adult first-line systemic treatment population, compared with CsA with TCS, abrocitinib 100 and 200 mg in combination with TCS are associated with probabilistic ICERS of £94,520 and £79,270. In the adult second-line systemic treatment population, abrocitinib 100 mg in combination with TCS is less costly and less effective than dupilumab in combination with TCS (south-west quadrant probabilistic ICER of £182,263). However, as a monotherapy for both doses and as combination therapy with the 200 mg dose, abrocitinib dominates dupilumab in the adult second-line systemic treatment population.

The proposed position in the treatment pathway by the company for tralokinumab was for the adult second-line systemic population. Compared with dupilumab as a monotherapy and combination therapy, tralokinumab was less expensive and less effective, resulting in south-west quadrant probabilistic ICERs of £337,165 for monotherapy and £220,640 for combination therapy.

For all treatments in all populations, probabilistic ICERs were consistent with deterministic ICERs. Variation in the probabilistic ICERs compared to the deterministic ICERs were seen for tralokinumab monotherapy, abrocitinib 100 mg, upadacitinib 15 and 30 mg in the adult second-line systemic combination treatment population. The EAG notes that in all analyses, incremental QALYs were relatively small for each treatment resulting in sensitive ICERs. Furthermore, the sensitivity in the ICERs was seen in the OWSA and scenarios (discussed below) with changes in magnitude but rarely direction of results.

The EAG cautions the interpretation of the cost-effectiveness results presented in the MTA report as they are based on list prices for abrocitinib, baricitinib, dupilumab, tralokinumab and upadacitinib as all have confidential PAS discounts in place.

In the OWSA for all populations, the key drivers of cost effectiveness were week-16 response probabilities and conditional discontinuation probabilities (used to inform the week-52 response and annual discontinuation), which are as expected as these are the key effectiveness estimates in the model. In particular, the NMA for week-16 response was associated with substantial uncertainty, in particular, for abrocitinib due to small numbers informing the network.

The EAG conducted a range of scenarios to test the impact on the ICER of alternative assumptions and data inputs for key parameters. In the adult second-line systemic treatment population, baricitinib is also a comparator, but it could not be included in the EAG base-case analyses as response data using the composite outcome were not made available to the EAG. However, EASI 75 response data were available for baricitinib combination therapy. Thus, the EAG conducted a scenario comparing the three new drugs against baricitinib as combination therapy using the EASI 75 response outcome. When compared with baricitinib in combination with TCS, upadacitinib 15 mg in combination with TCS is less costly and less effective (south-west quadrant deterministic ICER of £19,830) and in the 30 mg dose is more costly and more effective (deterministic ICER of £321,819). Both tralokinumab and abrocitinib 100 mg are dominated by baricitinib. However, abrocitinib 200 mg is more costly and more effective than baricitinib (deterministic ICER of £331,473). It should be noted that while abrocitinib and upadacitinib (both doses) are more effective than baricitinib, long-term discontinuations are more favourable for baricitinib compared with the low doses of abrocitinib and upadacitinib and are similar for the high doses. Furthermore, baricitinib is less expensive than all three new drugs. However, as mentioned previously, the results are based on list prices for interventions and results based on inclusion of confidential PAS discounts are more appropriate for decision-making.

The EAG explored the use of the EASI 75 response outcome for comparisons against dupilumab in the adult second-line systemic treatment population. For both the monotherapy and combination therapy analyses, ICERs were consistent with the base case, although total QALYs were higher for all treatments (except dupilumab combination therapy).

Focus on the composite outcome for the EAG base-case analyses was informed by the recommendations of TA534 and TA681, where the committee considered that EASI 50 and an improvement in the DLQI of at least four are sensitive to changes in treatment outcomes and more clinically relevant than an EASI 75. The EAG’s clinical experts fed back that EASI 50 + ΔDLQI ≥ 4 does inform their assessment of response to treatment, but they went on to caution that the subjective nature of the DLQI, as a patient-assessed tool that is open to recall bias, is also borne in mind and, consequently, their preference to assess clinical effectiveness is changed in EASI by 75%. Furthermore, the EAG’s clinical experts considered that EASI 75 is much harder to achieve compared to the composite outcome.

The majority of the QALYs generated for each treatment are derived from occupation in the BSC health state and over a lifetime most patients end up on BSC. Patients on BSC will experience periods of response and relapse and data on the disease course is limited. As such, the EAG explored a scenario reducing the time horizon of the model to 5 years for the adult analyses, which had a substantial impact on the results in the second-line population analyses. For both monotherapy and combination therapy for all interventions, ICERs become substantially more cost-effective. In particular for abrocitinib 100 mg (both monotherapy and combination therapy) and abrocitinib 200 mg monotherapy, ICERs were within the range to be considered cost-effective using a £30,000 per QALY threshold. Both tralokinumab and upadacitinib 15 mg dominated dupilumab for both monotherapy and combination therapy analyses. However, the ICERs for upadacitinib 30 mg (monotherapy and combination therapy) and abrocitinib 200 mg monotherapy were still above the £30,000 threshold, but substantially lower than the base case. The ERG notes that for the 5-year time horizon scenario, incremental QALYs increased for all treatments compared with dupilumab, suggesting that longer time horizon gives less-effective treatments time to ‘catch up’ in terms of overall QALYs accrued.

For the adolescent analyses, reducing the time horizon to 18 years of age resulted in the ICERs for all treatments remaining dominant.

Across all populations, using parameter utility values, conditional discontinuation and flare treatment estimates from TA534 had a substantial impact on the ICERs, in particular, using utility values from TA534 was a key driver of cost effectiveness. However, the TA534 scenario did not change the direction of the results, except for the abrocitinib 100 mg comparison in the adult second-line systemic treatment population (changed from south-west quadrant ICER to dominated by dupilumab) and all treatments in the adolescent population (changed from dominant compared with dupilumab to south-west quadrant ICERs). However, the TA534 scenario should be considered as illustrative as data from the key trials for each of the three new drugs are available and can be considered as more appropriate for the base case compared to values in TA534 that just reflect dupilumab only.

Most of the other scenarios across all populations resulted in a change in the magnitude of the ICER, but not in the direction of the results. The scenario using the NMA where censoring for rescue therapy was included resulted in a change in the ICER for upadacitinib 15 mg from dominant to south-west quadrant in the adult second-line systemic treatment population. However, the EAG considers that in clinical practice, patients would receive rescue therapy while on treatment and this was also noted in the FAD for baricitinib. As such, the scenario using the NMA where patients were censored for rescue therapy can be considered only illustrative. Furthermore, rescue therapy was not allowed in abrocitinib trials and very few patients received rescue therapy in the upadacitinib and tralokinumab studies.

Generalisability of results

The perspective of the cost-effectiveness analysis reflects the NHS in England and thus results are generalisable to the patients in England with moderate-to-severe AD. To ensure consistency with current clinical practice, the EAG has used relevant NICE guidance (TA534 and TA680) to inform key assumptions and parameters within the MTA model. Furthermore, the EAG consulted with its clinical experts to determine the trials which are most representative of the patient population in England to inform the baseline characteristics and effectiveness used in the model.

Strengths and limitations of analysis

The primary strength of the EAG’s analysis of the three new drugs compared with current practice for each of the subpopulations is that the results have been produced using a consistent approach to the CEA. Specifically, common assumptions have been used across all comparators, such that results facilitate a consistent basis for decision-making. The EAG has utilised available trial data for each of the interventions to ensure results are as robust as possible and has defined the populations under consideration to reflect the NICE final scope and the treatment pathway which closely reflect patients who would be seen in clinical practice. Furthermore, the EAG has specified a single NMA for each population for all relevant treatments to produce consistent effectiveness estimates to use in the MTA model.

The EAG’s approach to the CEA can be considered conservative, in particular, the implementation of the drug class approach to fill gaps in the data. The strength of the drug class approach can be seen when considering utility values as it makes the inherent assumption that HRQoL is unlikely to be affected by different treatments within a drug class but there could be differences across different drug classes due to the mode of action and administration of treatment.

Despite the strengths of the EAG’s approach, there were several limitations with the analyses that required assumptions to be made where possible and where not possible, omissions within the analysis. The most significant limitation with the analysis is that the composite outcome of EASI 50 + DLQI ≥ 4 could not be obtained for the adolescent and adult first-line systemic treatment populations due to a paucity of data for dupilumab informing the adolescent NMA and CsA data informing the adult first-line systemic treatment NMA. Furthermore, even though combination therapy data were available for abrocitinib and upadacitinib, only monotherapy could be assessed for the adolescent population as combination data for dupilumab were unavailable to inform the NMA. Conversely, only combination therapy could be assessed for the adult first-line systemic treatment population as monotherapy data were unavailable.

The EAG considered the feasibility of estimating an ‘adjustment factor’ to estimate combination therapy outcomes based on monotherapy outcomes (and vice versa) to fill the data gaps. Treatment response data presented in the section Treatment effectiveness suggest that combination therapy is more effective than monotherapy for adults in the second-line systemic treatment subgroup. However, there is not a consistent trend in terms of inflation of benefit when comparing combination therapy and monotherapy across the treatments. Additionally, the populations considered in the MTA are heterogeneous and an adjustment factor based on one population for one treatment may not reflect the true outcomes for population where data are unavailable. Therefore, the EAG found that there was not a robust method to estimate the missing data.

As the combination therapy analyses are more relevant for clinical practice, the EAG considered missing monotherapy data are unlikely to be critical for decision-making for the adult first-line systemic treatment subgroup. However, for the adolescent population, monotherapy analyses may potentially underestimate the effectiveness of the treatments when used in combination with TCS in clinical practice. Therefore, adolescent monotherapy analyses may reflect a conservative view of clinical effectiveness.

Another significant limitation with the analysis is that comparisons with baricitinib could only be included in scenario analysis. In TA681, baricitinib was assessed in the adult second-line systemic population using the composite outcome but data were redacted, and the company did not provide these data for inclusion in the MTA NMA analyses. As such, only EASI 75 data for baricitinib combination therapy could be included in the NMA for the adult second-line systemic treatment population. The downstream implication of a lack of base-case result for the three new drugs compared with baricitinib is that incremental analysis of dupilumab and baricitinib against each of the three new drugs could not be presented and any recommendations for the adult second-line systemic population based on the primary composite outcome are limited to comparisons with dupilumab.

Analyses exploring increasing or decreasing dose for abrocitinib and upadacitinib were not possible as efficacy data based on titrating dose are unavailable. However, the SmPC guidance for both abrocitinib and upadacitinib does take into consideration circumstances where moving to the lower or higher dose of each drug may be beneficial and this is likely to happen in clinical practice.

Another consideration for clinical practice that could not be explored in the current analyses was treatment sequencing. Currently, there is a lack of clinical data on the effectiveness of sequences of AD treatments, especially changing drug class (e.g. starting on a JAK inhibitor and then moving to a monoclonal antibody). As such, more clinical data are needed to understand how patients respond to subsequent lines of systemic treatment and the resulting cost effectiveness of treatment sequences.