Adalimumab, etanercept and ustekinumab for treating plaque psoriasis in children and young people: systematic review and economic evaluation

Physician Global Assessment

The PGA is an instrument that provides a subjective overall evaluation of plaque psoriasis severity using a scale of seven categories (‘clear’, ‘almost clear’, ‘mild’, ‘mild to moderate’, ‘moderate’, ‘moderate to severe’, ‘severe’). 21 There are two primary forms: a static form [Static Physician Global Assessment (sPGA)], which measures the physician’s impression of the disease at a single point, and a dynamic form [Dynamic Physician Global Assessment (dPGA)], in which the physician assesses the global improvement from baseline. 17

The sPGA uses seven scaled scores to describe the severity of disease: 0 = ‘clear’, 1 = ‘almost clear’, 2 = ‘mild’, 3 = ‘mild to moderate’, 4 = ‘moderate’, 5 = ‘moderate to severe’ and 6 = ‘severe’. 17,22 The dPGA, on the other hand, uses six scaled scores to describe either improvement or deterioration of disease. For disease improvement the scores are +1 = ‘mild’, +2 = ‘moderate’, +3 = ‘moderate to large’, +4 = ‘large’ and +5 = ‘very large’. For disease deterioration the scores are –1 = ‘mild’, –2 = ‘moderate’, –3 = ‘moderate to large’, –4 = ‘large’ and –5 = ‘very large’. A score of zero indicates no or minimal change.

As the sPGA scoring system is simpler to use than the dPGA scoring system, because, with the dPGA, physicians have to record the severity of psoriasis at baseline to evaluate the change in disease status after a follow-up period, the sPGA has become a widely used treatment response assessment tool in practice. 17 However, the sPGA does not discriminate small changes and the score ranges are not robust. 17

Psoriasis Area and Severity Index

In clinical trials of patients with psoriasis, assessment of the response to treatment is usually based on the PASI. 17 Although it is widely used, the PASI also has a number of deficiencies: its constituent parameters have never been properly defined; it is insensitive to change in mild-to-moderate psoriasis; estimation of disease extent is notoriously inaccurate; and the complexity of the formula required to calculate the final score further increases the risk of errors. It combines an extent and a severity score for each of the four body areas (head, trunk, upper extremities and lower extremities). The extent score of 0–6 is allocated according to the percentage of skin involvement (e.g. 0 and 6 represent no psoriasis and 90–100% involvement respectively). The severity score of 0–12 is derived by adding scores of 0–4 for each of the qualities of erythema (redness), induration and desquamation, representative of the psoriasis within the affected area. It is probable, but usually not specified in trial reports, that most investigators take induration to mean plaque thickness without adherent scale and desquamation to mean thickness of scale rather than severity of scale shedding. The severity score for each area is multiplied by the extent score and the resultant body area scores, weighted according to the percentage of total body surface area (BSA) that the body area represents (10% for head, 30% for trunk, 20% for upper extremities and 40% for lower extremities), are added together to give the PASI score. Although the PASI score can theoretically reach 72, scores in the upper half of the range (> 36) are not common, even in severe psoriasis. Furthermore, the PASI score fails to capture the disability that commonly arises from involvement of functionally or psychosocially important areas (hands, feet, face, scalp and genitalia), which together represent only a small proportion of total BSA. 23 However, PASI-based measures have discriminatory capability and are generally accepted for the assessment of treatment effects. However, clinical expert opinion is that the PASI is not widely used in clinical practice.

Despite the fact that it has not been validated in children and young people as a measure of disease severity, the PASI was chosen as the primary outcome variable for psoriasis in the economic evaluation because it is used in the majority of randomised controlled trials (RCTs). Typically, the PASI is reported as a dichotomous measure indicating a 50%, 75% or 90% reduction in PASI score from baseline (PASI 50, PASI 75 and PASI 90 respectively).

Children’s Dermatology Life Quality Index

The CDLQI is a 10-item questionnaire that aims to measure the quality of life of children (aged 4–16 years) based on how much they have been affected by a skin problem over the week preceding the date of questioning. 18 The 10 items cover six areas of daily activities: symptoms and feelings, leisure, school or holidays, personal relationships, sleep and treatment. 24,25 Usually, children, either alone or with the help of their parents, choose one of the four possible replies (scored from 0 to 3), with a maximum overall score of 30 and with a high score corresponding to low quality of life and vice versa. 25

Children’s Dermatology Life Quality Index scores can be divided into scoring bands – band 0 (score of 0–1), band 1 (score of 2–6), band 2 (score of 7–12), band 3 (score of 13–18) and band 4 (score of 19–30) – that respectively correspond to no, small, moderate, very large or extremely large effects on the child’s quality of life. 25 However, the CDLQI is not considered appropriate for use as a health-related quality-of-life (HRQoL) assessment tool beyond the age of 16 years.

Pediatric Quality of Life Inventory

The PedsQL is a modular instrument for measuring HRQoL in children and adolescents (age 2–18 years). It consists of 23 items in four domains: physical functioning (8 items), emotional functioning (5 items), social functioning (5 items) and school functioning (5 items). Each item receives a score of 0–4 (0 = ‘never a problem’, 1 = ‘almost never a problem’, 2 = ‘sometimes a problem’, 3 = ‘often a problem’, 4 = ‘almost always a problem’) and are reverse scored and linearly transformed to a 0–100 scale (0 = 100, 1 = 75, 2 = 50, 3 = 25, 4 = 0), so that higher scores indicate better HRQoL. 26 Paediatric self-report is measured in children and adolescents aged 5–18 years and parent proxy report of child HRQoL is measured for children and adolescents aged 2–18 years.

Teenager’s Quality of Life Index

Built on qualitative data from patients, the Teenager’s Quality of Life Index (T-QoL) is a validated tool to quantify the impact of skin disease on adolescents’ quality of life. 20 The index consists of 18 items categorised into three domains: self-image, physical well-being and the future, and psychological impact and relationships. The authors have proposed the T-QoL as an outcome measure in both clinical practice and clinical research. 20

General issues with quality-of-life measurement in childhood psoriasis

Quality-of-life measurements may not be particularly meaningful in younger children with psoriasis, who are less good at articulating how much the disease is bothering them. In the case of younger children, proxy measurements may more accurately reflect parental perception or concern. There is only moderate correlation between PASI/PGA response measures and the CDLQI;27 some children with relatively mild disease can have very poor HRQoL scores, whereas others with more severe disease can have acceptable HRQoL. As well as disease symptoms and consequences, the frequency of injections can be an important quality-of-life consideration in children.

Alternative treatments in children and young people

To inform the network meta-analysis (NMA), searches were undertaken to identify relevant RCTs of systemic non-biological (acitretin, methotrexate and ciclosporin) and other biological (infliximab, secukinumab) therapies used in children and young people with plaque psoriasis. No language, date, geographical or study design limits were applied to the searches.

The searches were carried out on 31 May 2016 in the following databases: MEDLINE (including MEDLINE Epub Ahead of Print, MEDLIINE In-Process & Other Non-Indexed Citations, Ovid MEDLINE Daily and Ovid MEDLINE), CENTRAL, CDSR, CINAHL Plus, DARE, EMBASE, HTA database, PubMed and Science Citation Index.

In addition, the following resources were searched for ongoing, unpublished or grey literature: ClinicalTrials.gov, Conference Proceedings Citation Index – Science, EU Clinical Trials Register, PROSPERO and WHO International Clinical Trials Registry Platform portal.

The search results were imported into EndNote X7 and deduplicated. The search was updated in September 2016 to capture more recent studies. Full search strategies can be found in Appendix 1.

Registry data

To identify longer-term follow-up evidence, a literature search was conducted within the MEDLINE database for the search terms ‘psoriasis AND regist*’. The results of this search were screened for publications from psoriasis registries, secondary analyses of registry data and systematic reviews of broader dermatological and psoriasis registry data. The list of registries generated through these searches was compared against those in three relevant systematic reviews30–32 to verify the studies included and to identify any that had been overlooked. Twenty patient registries for psoriasis treatment were identified in this way; 14 were located in European countries, three were international in scope, two were based in the USA and one was based in Malaysia. Each registry name was then separately used as a search term in MEDLINE and any publications referencing these that had not been found in the initial searches were retrieved.

In addition, representatives of the 14 psoriasis registries from European countries (Austria, Australia, Czech Republic, Denmark, France, Germany, Italy, Netherlands, Portugal, Slovenia, Spain, Sweden, Switzerland and the UK) were contacted and asked to provide any relevant information on the use of the biologics adalimumab, etanercept and ustekinumab for the treatment of psoriasis in children and young people.

Study design

Randomised controlled trials (including any open-label extensions of RCTs) were eligible for the review of clinical efficacy.

Information on adverse events (AEs) was also sought from regulatory sources when appropriate. Registries and observational studies were included when relevant outcome data were available.

To address longer-term measures of efficacy and drug survival, published analyses based on large and long-term data sets (including studies of registry data) were also considered.

Participants

Studies of children and/or young people with moderate to severe plaque psoriasis were included. Severity could be defined using the PASI, PGA, BSA or other measures, alone or in combination, although there is no universal definition of severity for this population. Studies of guttate, erythrodermic and pustular psoriasis were excluded, as were studies of psoriatic arthritis.

Studies in children or young people with psoriasis in whom topical therapies, systemic therapies or phototherapies were inadequate, inappropriate or not tolerated were eligible for inclusion. Participants aged < 12 years were considered to be children whereas those aged 12–17 years were considered to be young people.

Interventions

The relevant interventions were adalimumab, etanercept and ustekinumab.

Comparators

The relevant comparators were:

• alternative biological therapies with relevant marketing authorisation (adalimumab, etanercept or ustekinumab)

• non-biological systemic therapy (including, but not limited to, ciclosporin and methotrexate)

• topical therapy (for people in whom non-biological systemic therapy is not suitable), that is, best supportive care (BSC)

• biological treatments used outside their marketing authorisation (such as infliximab, adalimumab, etanercept or ustekinumab if used outside the constraints of the relevant marketing authorisation in children and young people)

• biosimilars of etanercept, adalimumab or ustekinumab

• placebo.

Outcomes

Data on effectiveness, adverse effects, patient-centred outcome measures, costs to the health service and cost-effectiveness were eligible for inclusion, including the following outcomes:

• severity of psoriasis (e.g. BSA, PGA score)

• response and remission rates (e.g. PASI 50/75/90 response)

• relapse rate

• rates of treatment discontinuation and withdrawal

• short- and long-term adverse effects of treatment (e.g. injection site and allergic reactions, serious infections, reactivation of infections including tuberculosis, malignancy)

• HRQoL [e.g. CDLQI, PedsQL and EuroQol-5 Dimensions (EQ-5D)33 scores].

Baseline characteristics of participants

The baseline characteristics of the participants in the included RCTs are presented in Table 4. Although only older children and adolescents (aged 12–17 years) were included in the ustekinumab trial, the median age of children across the three trials did not differ greatly, as it appears that relatively few younger children were included in the adalimumab and etanercept trials.

All three trials used a composite measure of disease severity incorporating baseline PASI, PGA and BSA measurements. When used in isolation, a PASI score between 10 and 20 is considered to indicate moderate to severe psoriasis, whereas severe psoriasis has a score of > 20. Across the included studies, the average PASI score ranged from 18.3 to 21.2, with 93–100% of participants having a PGA score of > 3 (mild/moderate disease). Although adalimumab and etanercept are licensed for severe chronic plaque psoriasis and ustekinumab is licensed for moderate to severe plaque psoriasis, on average, measures of disease duration and the component measures of severity did not appear to differ markedly between the three trials. The degree of psoriasis affecting high-impact and difficult-to-treat sites (e.g. face, scalp, palms, soles, flexures and genitals) across the three studies was less clear.

A key difference between the licences for the three agents is the availability of adalimumab for patients for whom topical therapy and phototherapy are inadequate or inappropriate. Unlike the licences for etanercept and ustekinumab, there is no mention in the licence for adalimumab of previous non-biological systemic treatment. However, a substantial minority of participants in the adalimumab trial (29.8%) had received prior systemic therapy, compared with 42.7% of participants in the ustekinumab trial; 56.8% of participants in the etanercept trial had received either prior systemic therapy or prior phototherapy (separate data were not reported).

A similar proportion of participants in the adalimumab and ustekinumab trials had received some form of biological treatment prior to enrolment (9.6% and 10.8% respectively). As etanercept was the first TNF-α inhibitor to be approved for psoriasis, none of the participants recruited to the etanercept trial had previously been treated with biological therapy.

Although there were noticeable differences in participant characteristics between the trials, these were not as clear as the respective licences for the three treatments might suggest. Notwithstanding methodological differences, there appears to be sufficient overlap in the trial populations to discuss these three trials together.

Length of follow-up and early escape

The initial randomised treatment period was 12 weeks in the etanercept and ustekinumab trials and 16 weeks in the adalimumab trial. Twelve-week outcome data were not available for the adalimumab trial, although clinical advice (Dr Ruth Murphy, Consultant Dermatologist, Sheffield Teaching Hospitals NHS Foundation Trust, Sheffield, 25 October 2016, personal communication) suggested that the difference in length of follow-up between treatments was acceptable.

All three trials allowed participants to ‘escape’ from the randomised treatment period before the 12-/16-week follow-up. The criteria for early escape and statistical handling of early escape data are discussed separately for each trial in the relevant sections.

Post-randomised treatment periods are briefly summarised in Table 3.

Outcomes

The adalimumab and etanercept trials considered the PASI 75 response to be the primary outcome measure, whereas the ustekinumab trial used a primary outcome measure of a PGA score of 0 or 1 (‘clear’ or ‘almost clear’). However, all three trials reported PASI and PGA scores and some measure of HRQoL (CDLQI and/or PedsQL), which are presented in the following sections.

Psoriasis Area and Severity Index response

(Confidential information has been removed) and PASI 75 response rates at 16 weeks were significantly greater for standard-dose adalimumab (0.8 mg/kg) than for methotrexate [(confidential information has been removed) 58% vs. 32% respectively)]. Low-dose adalimumab (0.4 mg/kg) did not show a statistically significant improvement over methotrexate for these outcomes [(confidential information has been removed) and 44% vs. 32% respectively]. PASI 90 response rates did not differ significantly between the three treatment arms (Table 8).

(Confidential information has been removed.)

Physician Global Assessment

The proportion of participants achieving a sPGA score of 0 or 1 (‘clear’ or ‘minimal’) at 16 weeks was greater for standard-dose adalimumab than for low-dose adalimumab or methotrexate (61% vs. 41% vs. 41% respectively), although this difference was not statistically significant.

(Confidential information has been removed.)

Quality of life

Two HRQoL measures, CDLQI and PedsQL, were reported at 16 weeks. All three treatment groups showed improvements from baseline in the dermatology-specific quality-of-life measure (CDLQI), exceeding the published minimal clinically important difference (MCID) of a 2.5-point change from baseline. 48 However, these improvements were similar across the three treatment groups, with no significant difference between either dose of adalimumab or methotrexate (6.6 for adalimumab 0.8 mg/kg vs. 4.9 for adalimumab 0.4 mg/kg vs. 5.0 for methotrexate respectively).

Unlike the CDLQI, improvements on the generic HRQoL measure (PedsQL) significantly favoured both doses of adalimumab over methotrexate (mean changes of 10.8 and 9.5 for standard- and low-dose adalimumab, respectively, vs. 1.9 for methotrexate). The mean changes in the adalimumab groups both exceeded the published MCID of 4.4 for the PedsQL. 26

It is unclear why PedsQL scores would increase in the absence of dermatology-related quality-of-life benefits as measured by the CDLQI. However, both mean and median PedsQL scores at baseline were noticeably higher in the methotrexate arm than in the adalimumab treatment arms (see Table 4) and so the observed PedsQL change scores in the adalimumab arms may be overestimates because of regression to the mean. 89

Period B: withdrawal

(Confidential information has been removed.)

Period C: retreatment

(Confidential information has been removed.)

(Confidential information has been removed.)

Period D: long-term follow-up

(Confidential information has been removed.)

Adverse events at 16 weeks

Adverse event rates were comparable among the three treatment groups (Table 11). Three serious adverse events (SAEs) considered unrelated to treatment (hand fracture, gastrointestinal infection from food poisoning and agitation as a result of alcohol consumption) were reported, all of which occurred in participants receiving 0.4 mg/kg of adalimumab. One participant in the same treatment arm withdrew because of an AE (moderate psoriasis flare).

Longer-term safety of adalimumab

Table 12 shows that the overall numbers of AEs during patient follow-up across all four study periods were similar across treatment arms. A total of nine SAEs were reported in six participants. In terms of episodes per 100 patient-years, the total rate of SAEs was 5.9 for all participants ever treated with 0.8 mg/kg of adalimumab from the first dose of 0.8 mg/kg adalimumab and 7.4 for all participants treated with adalimumab (0.4 mg/kg and 0.8 mg/kg) from the first dose of 0.8 mg/kg adalimumab.

One SAE (haemorrhagic ovarian cyst) occurred in period B in a participant who had been initially randomised to 0.8 mg/kg of adalimumab.

Five SAEs occurred during period D, including one death from an accidental fall, one tendon injury in a participant receiving 0.4 mg/kg of adalimumab, one maculopapular rash in a participant receiving 0.8 mg/kg of adalimumab, one case of chest pain in a participant randomised to methotrexate but receiving 0.8 mg/kg of adalimumab and one case of eye naevus in a participant receiving 0.8 mg/kg of adalimumab. All SAEs were considered by investigators to be unrelated or probably unrelated to the study drug with the exception of the case of eye naevus, which was assessed as being possibly related.

In addition to the participant who discontinued treatment because of a moderate psoriasis flare in period A, one participant initially randomised to methotrexate but receiving 0.8 mg/kg of adalimumab during period D discontinued treatment because of severe urticaria.

The rate of all infections reported by participants receiving 0.8 mg/kg of adalimumab was 170.4 episodes per 100 patient-years. Only two events of tuberculosis occurred, both during period D.

Psoriasis Area and Severity Index response

Psoriasis Area and Severity Index 50, 75 and 90 responses in the etanercept group were 74.5%, 56.6% and 27.4% respectively. Response rates for the placebo group were 22.9%, 11.4% and 6.7%. When translated into relative risk (RR) values, the etanercept group had a significantly higher probability of achieving PASI 50, 75 and 90 responses, with RRs of 3.26 [95% confidence interval (CI) 2.26 to 4.71], 4.95 (95% CI 2.84 to 8.65) and 4.10 (95% CI 1.88 to 8.95) respectively.

Physician Global Assessment

The proportion of participants achieving a PGA score of 0 or 1 (‘clear’ or ‘minimal’) at 12 weeks was significantly greater in the etanercept group than in the placebo group (52.8% vs. 13.3%), equating to a RR of 3.96 (95% CI 2.36 to 6.66).

Quality of life

Data for two HRQoL measures, CDLQI and PedsQL, were available at 12 weeks (see Table 14). Both the etanercept group and the placebo group showed improvements from baseline in CDLQI scores, exceeding the published MCID of a 2.5-point change from baseline,48 although the improvement in the etanercept group was statistically significantly greater than that in the placebo group (mean difference 2.3, 95% CI 0.85 to 3.74).

Both treatment groups also showed improvements on the PedsQL, although for the placebo group this fell below the published MCID of 4.4. The mean change in PedsQL score from baseline, although favouring etanercept, was not statistically significantly different between the treatment groups (mean difference 3.0, 95% CI –0.87 to 6.87).

Subgroup outcomes

Age-based subgroup analysis results of PASI responses for the etanercept trial (20030211) were available (Table 16). A higher proportion of the etanercept treatment group than the placebo group achieved PASI 50, 75 and 90 responses in all age categories. Imputation of treatment failure for participants entering the early escape arm reduced the magnitude of the difference between treatments, although these differences remained formally statistically significant for all comparisons, with the exception of PASI 90, which was of borderline statistical significance (p = 0.054).

Weeks 12–36: open-label etanercept treatment

At the end of the 12-week double-blind period a total of 208 participants (105 and 103 of the original etanercept and placebo groups respectively) entered an open-label treatment phase (i.e. all were treated with etanercept) and were followed up until week 36.

Patients who did not achieve a PASI 50 response at week 24 were given the option to discontinue the study or enter the incomplete responder arm. Participants in the incomplete responder arm had the option to receive topical psoriasis therapy according to the standard of care in addition to receiving open-label etanercept (Figure 2).

FIGURE 2.

Short-term and long-term participant follow-up flow chart for the etanercept trial.

By weeks 24 and 36 (i.e. after 12 and 24 weeks of open-label etanercept respectively), participants who were originally randomised to placebo during the double-blind period achieved similar PASI and PGA responses as participants receiving etanercept throughout (Table 17).

Weeks 36–48: re-randomised ‘withdrawal–retreatment’ period

At week 36, 138 patients who had achieved a PASI 50 response at week 24 or a PASI 75 response at week 36 were randomised in a 1 : 1 ratio to receive either etanercept or placebo in a double-blinded fashion. Patients were followed up for a further 12 weeks until week 48.

During the follow-up, 42 participants from the ITT population (29/69 and 13/69 from the placebo and etanercept arms respectively) lost their PASI 75 response and so were allocated to receive etanercept in an open-label fashion until week 48.

Overall, 52 out of 65 participants (80%) who received etanercept throughout the withdrawal–retreatment period maintained a PASI 75 response and 85% of those re-randomised to placebo and who did not lose a PASI 75 response during follow-up retained their response at week 48. Only 36% of those who were retreated with open-label etanercept after losing a PASI 75 response on placebo had regained a response by week 48 (Table 18). The use of PASI 75 response as both a retreatment rule and as an outcome makes these results difficult to interpret; however, a relatively high rate of late crossover from placebo to etanercept could partly explain a lack of response on PASI and PGA measures among these participants.

Weeks 48–312 (20050111)

In total, 194 participants completed 48 weeks of follow-up in the etanercept trial (20030211) (57 participants who received etanercept and topical therapy starting from the open-label treatment phase, 95 participants who were randomised to the etanercept and placebo arms and who continued to receive the blinded drug and 42 participants who were randomised to either etanercept or placebo but who did not achieve a PASI 75 response and who were retreated with etanercept until the end of the study).

Of the 194 participants who completed the etanercept trial (20030211), 182 were enrolled in an open-label extension study (20050111)66 to establish the long-term safety of etanercept. Participants received 0.8 mg/kg of etanercept (up to a maximum dose of 50 mg) subcutaneously once weekly for a further 264 weeks. In total, 63 participants (34.6%) completed 264 weeks of follow-up.

During the 264 weeks of further follow-up, the probability of achieving a PASI 50, 75 and 90 response was similar across all of the outcome recording points (Table 19). However, it should be noted that, by week 264, 63.6% of the participants (115/181) had withdrawn from the study and the reasons for withdrawal were unavailable.

Adverse events at 12 weeks

The number of AEs reported during the 12-week randomised phase was similar for etanercept and placebo (68 vs. 62). There were 50 infections and seven injection site reactions in the etanercept group, compared with 33 infections and five injection site reactions in the placebo group. Although the difference in rate of infections fell short of formal statistical significance, there were noticeably more infections in the etanercept group (47.2% vs. 31.4%; p = 0.0683). One participant in the etanercept group withdrew because of an adverse effect (no further details available). No SAEs were observed during the 12-week randomised phase (Table 20).

Adverse events at weeks 12–36

Up to week 36, 282 infections were reported during treatment with etanercept (238.18 events per 100 person-years). The most common infections were upper respiratory tract infection, nasopharyngitis, influenza, pharyngitis, streptococcal pharyngitis and viral upper respiratory tract infection (59.97, 33.79, 16.05, 15.20, 8.45 and 8.45 events per 100 participant-years respectively). During the same period, a single serious non-infectious AE (benign ovarian mass) and one serious case of gastroenteritis with dehydration were observed (Table 21).

Over the same time period, 2.4% of participants (5/208) withdrew because of AEs: three participants withdrew because of non-infectious AEs (psoriasis, atopic dermatitis and muscle cramps) and two participants withdrew because of infections (pneumonia and skin infection).

Adverse events at weeks 48–312

A total of 161 participants (89.0%) reported at least one AE up to week 264 of the follow-up study (20050111). Seven participants (3.9%) reported a SAE, with each participant reporting a single event: anxiety, cellulitis, infectious mononucleosis, postoperative intestinal obstruction, osteonecrosis and a thyroid cyst, with the seventh participant undergoing an elective abortion (Table 22). Of the seven SAEs, only the cellulitis infection was considered by the investigator to be related to etanercept treatment.

Six participants (3.3%) withdrew from the study because of either an infectious or a non-infectious AE. Two participants withdrew because of Crohn’s disease and one participant each withdrew because of glomerulonephritis (secondary to infection), psoriasis, sinusitis and nerve paralysis. The case of glomerulonephritis and one of the cases of Crohn’s disease were considered to be related to treatment.

No SAE led to study withdrawal. No opportunistic infections or deaths occurred during the study and no malignancies were reported.

Efficacy at week 12

Data on treatment response (PASI and sPGA) and HRQoL (CDLQI and PedsQL) outcomes for the CADMUS RCT are presented in Tables 24 and 25.

Psoriasis Area and Severity Index response

Among participants continuing ustekinumab treatment, PASI responses observed in week 12 appeared to be sustained at week 52, with few participants lost to follow-up (one participant was lost from the standard-dose arm and two from the half-standard-dose arm) (Table 27).

Participants who were randomised to placebo and who crossed over to the standard ustekinumab dosage (0.75 mg/kg) achieved better PASI responses than those who were randomised to placebo and crossed over to the half-standard dosage of ustekinumab (0.375 mg/kg) (see Table 27).

Physician Global Assessment

A similar pattern of responses was seen for the sPGA, with similar response rates at week 52 as at week 12 among participants continuing ustekinumab treatment and a large improvement between weeks 12 and 52 for participants crossing over to active treatment from placebo (see Table 27).

Adverse events at week 12

The percentage of participants reporting AEs did not differ significantly between the ustekinumab groups (44.4% standard-dosage group and 51.4% half-standard-dosage group) and the placebo group (56.8%) (Table 28). No SAEs were reported in the standard-dosage group or the placebo group, whereas one participant in the half-standard-dosage group was hospitalised for worsening of psoriasis (see Table 28).

One participant in the standard dosage group had a mild injection site reaction. There were no incidences of serious infection, tuberculosis, malignancy or withdrawals because of AEs during the initial 12-week treatment period (see Table 28).

Discontinuation up to week 40

Up to week 40, 8.2% of the participants (9/110) discontinued the trial. The most common reasons for discontinuation were a lack of efficacy and AEs (Table 29). Five participants (13.5%) who were originally randomised to the half-standard-dosage group discontinued the trial compared with two participants (5.6%) in the standard-dosage group. Two patients who crossed over from placebo to the half-standard dosage of ustekinumab withdrew because of AEs.

Adverse events up to week 60

Up to week 60, 81.8% of participants (90/110) in the ustekinumab combined group reported one or more AEs (Table 30). Of the 74 participant-recorded infections, 18 (24%) were considered reasonably related to ustekinumab treatment.

In total, 5.5% of participants (6/110) in the ustekinumab combined group (five participants in the half-standard-dosage group and one participant in the standard-dosage group) reported SAEs (see Table 30). Four (one because of worsening of psoriasis) of the 110 participants in the ustekinumab combined group discontinued the trial because of an AE by week 60.

Registry data for children

Further screening was carried out to find registry publications making explicit reference to children with psoriasis and specifically providing information on the survival of biological treatment. Two registries (Child-CAPTURE and DERMBIO) were found to include children with psoriasis who were treated with biologics. Child-CAPTURE (the Netherlands) contained seven children treated with etanercept,90 but did not differentiate between biological and non-biological therapies in drug survival analyses. We identified from the 2014 annual report91 by the DERMBIO (Denmark) registry that there are 37 children enrolled who are undergoing treatment with adalimumab, etanercept or ustekinumab, although data for this group were not reported separately. Cox regression modelling of covariates in two studies found no significant predictive relationship between patient age and drug survival, suggesting that treatment withdrawal rates among children were similar to those in adults. 92,93

Wider registry data

In one 2015 DERMBIO study following 1277 (predominantly adult) psoriasis patients for up to 10 years,92 median drug survival for etanercept was 30 months (95% CI 25.1 to 34.9 months), which was significantly lower than for adalimumab (59 months, 95% CI 45.6 to 72.4 months) and ustekinumab (median not reached). Year-on-year drug survival for etanercept (estimated from the Kaplan–Meier curve) was 0.70 at 1 year, 0.53 at year 2 and 0.30 at year 5, whereas year-on-year drug survival for ustekinumab was 0.85 after 1 year, 0.78 at year 2 and 0.65 at year 5 (Table 32). Loss of efficacy was the most likely reason for drug discontinuation, but this was of greater significance proportionally for etanercept than for the other biologics analysed.

Findings from the British Association of Dermatologists Biologic Interventions Register (BADBIR) were broadly similar to those seen in the Danish cohort. A study by Warren et al. 94 on drug survival over 3 years in 3523 biologic-naive patients found that 77% of patients remained on biologic treatment over the first year, falling to 53% by the third year. Again, there were significant differences in the treatment withdrawal rates between biologics. Ustekinumab exhibited the highest first-course survival rate at 0.89 at year 1 and 0.75 at year 3. Adalimumab showed the highest survival of the anti-TNF-α drugs, at 0.79 at year 1 and 0.59 at year 3. Disregarding the very small population of patients on infliximab, etanercept was consistently the worst-performing drug in terms of treatment withdrawal, with a 1-year survival rate of 0.70, dropping to 0.40 at 3 years (Table 33). Etanercept was also found to be a significant predictor of discontinuation of therapy because of loss of efficacy. Other significant predictors of treatment withdrawal were female sex, smoking status and a higher baseline Dermatology Life Quality Index (DLQI) score.

Analysis 1: results using minimum evidence from the adult population

Table 43 summarises the results of the NMA in terms of absolute PASI response rates for the unconstrained (no explicit adjustment for differences in placebo response rates across the trials) and constrained (placebo response rates predicted based on the placebo arm trials in children and young people only) models. The results of both sets of analyses show that all active treatments are more effective than placebo. In terms of mean response rates (analysis 1b results), ustekinumab is estimated to have the highest probability of achieving a PASI 50 (90%, 95% CrI 81% to 96%), PASI 75 (79%, 95% CrI 64% to 90%) and PASI 90 (57%, 95% CrI 39% to 74%) response compared with any of the other treatments, suggesting that it is the most effective intervention. This is followed by adalimumab, etanercept and methotrexate in both sets of analyses, that is, the ranking of treatments based on mean response rates is unchanged in the different models.

The unconstrained baseline model (analysis 1a), however, predicts a placebo effect for PASI 75 of 20.3% (95% CrI 14% to 27%) whereas the constrained baseline model (analysis 1b) predicts a placebo effect of 13.1% (95% CrI 8% to 19%). This difference is driven by the CHAMPION trial,106 which had a substantially higher placebo response rate of approximately 19% for PASI 75 compared with the placebo response rates observed in the trials of children and young people (approximately 11% in 20030211 and CADMUS). The constrained baseline model (analysis 1b) adjusts the baseline predictions to consider only placebo effect evidence from trials in the younger population. In this analysis, the mean PASI 75 response rate for placebo is reduced and closer to the observed response in the children and young people trials.

As shown by the CrIs around the mean response rates, which are wide and overlap, there is uncertainty around these response rates. This is also shown in terms of the RRs of each treatment compared with placebo and their CrIs for the best-fitting model 1b (Table 44).

Analysis 2: results using all relevant evidence from the adult population

Table 45 summarises the absolute PASI response rates from the NMA that uses the full network of evidence in both populations for the unadjusted random-effects model (analysis 2). Relative treatment effects for analysis 2 for PASI 75 response are presented in Table 46. The random-effects approach outperformed the fixed-effect approach in terms of model fit, suggesting that accounting for between-study heterogeneity is an important factor (τ² = 0.02).

The results of this analysis suggest that ustekinumab is the most effective intervention, with the highest mean probability of PASI response (PASI 75: 73%, 95% CrI 67% to 79%), followed by adalimumab (PASI 75: 63%, 95% CrI 55% to 70%), etanercept (PASI 75: 40%, 95% CrI 34% to 47%) and methotrexate (PASI 75: 34%, 95% CrI 25% to 42%). Ustekinumab was statistically significantly more effective than any other agent based on relative effect estimates for PASI 75 (vs. etanercept: RR 1.78, 95% CrI 1.50 to 2.12; vs. adalimumab: RR 1.15, 95% CrI 1.01 to 1.35) and adalimumab was statistically significantly more effective than etanercept (RR 1.54, 95% CrI 1.25 to 1.88). The estimated pooled placebo absolute effect is in line with that observed, on average, across all studies in all populations.

These unadjusted results, however, do not consider an explicit adjustment for differences in placebo response rates across trials or differences across the populations (i.e. children and young people compared with adults). In the following sections, the results from the adjusted analyses are presented.

Adjustment for differences in placebo response rates across the trials

The NMA in the full population compares treatment outcomes across a large number of separate clinical trials. The reliability of these comparisons depends on the cross-trial similarity of the patient populations included in the network. An important difference between the included trials is the observed PASI response rates in the placebo arms of the trials, which is a common reference treatment across the majority of the trials. Table 41 showed that the PASI response rates in the placebo arms of the trials ranged from 0%136 to 20%. 108 All of the trials varied by design, eligibility criteria, previous medication use, average age and other characteristics. All of these variations could contribute to differences in placebo response rates and, therefore, to differences in the relative efficacy of the intervention compared with placebo. However, there is no systematic way to identify the reasons for these differences. A ‘placebo creep’ phenomenon has been discussed in the literature, which identifies a relationship between placebo response rates and time since publication of the trial results. However, such a phenomenon has not been identified in the trials considered in the NMA (Figure 6). The average PASI 75 response rate in the placebo arm across all trials is 6.2%, whereas the average rate in studies of adult populations is 5.9% and in studies of children and young people is 11.1%. Three adult studies106,108,132 have substantially higher placebo response rates (approximately 18–20%) than the other studies. Four studies, including the two trials in children and young people72 and two in adults,114,116 have approximately double the average placebo rate.

FIGURE 6.

Probability of a PASI 75 response in the placebo arms of trials in the NMA by publication year.

It is not clear exactly how these varying placebo rates affect treatment effects; however, it is clear that any differences will affect the relative efficacy of the interventions compared with placebo. Therefore, a potential relationship between baseline risk and relative treatment effect was explored103 in analysis 2a.

Tables 47 and 48 present the results of the model that adjusts for differences in placebo response rates. As for the unadjusted analysis (i.e. analysis 2), the baseline adjusted random-effects model was found to fit the data considerably better than the fixed-effect counterpart (DIC: 1303.7 fixed effects vs. 1177.6 random effects; total residual deviance: 473.5 fixed effects vs. 380.9 random effects). Furthermore, the 95% CRIs for the estimated mean baseline effect derived in the baseline-adjusted model do not include zero (–0.93, 95% CrI –0.97 to –0.88). This suggests that adjusting for baseline risk heterogeneity is important to explain existing between-study variation.

The results of analysis 2a suggest that ustekinumab is the most effective intervention, with the highest mean probability of PASI response (PASI 75: 73%, 95% CrI 66% to 79%), followed by adalimumab (PASI 75: 66%, 95% CrI 58% to 74%), and etanercept (PASI 75: 41%, 95% CrI 35% to 49%) and methotrexate (PASI 75: 34%, 95% CrI 25% to 44%). Ustekinumab is statistically significantly more effective than etanercept based on relative effect estimates for PASI 75 (RR 1.77, 95% CrI 1.48 to 2.11), but not statistically significantly more effective than adalimumab (RR 1.10, 95% CrI 0.96 to 1.28). Adalimumab is also statistically significantly more effective than etanercept (RR 1.60, 95% CrI 1.31 to 1.95).

Adjusting for differences in population and placebo response rates

Although evidence from trials in both children and young people and adults contributed to the full network of evidence (effectively assuming independence between age and treatment effectiveness), it is important to recognise that the age of the population could contribute to differences in treatment efficacy. Therefore, in analysis 2b we adjusted for differences in the population and differences in placebo response rates (as the placebo response rates were considerably different in the trials of children and young people and the trials of adults). Table 49 summarises the results of this analysis in terms of PASI response outcomes for both populations. Table 50 presents the corresponding RRs for PASI 75 for children and young people.

The model from analysis 2b fits the data as well as model 2a, as both present similar average total residual deviance [380.8 (2b) vs. 381.7 (2a)]. However, the DIC is substantially higher for model 2b. This suggests that this model is being penalised because of issues of parsimony. The children and young people subgroup effect is estimated not to be statistically significantly different from the adult subgroup effect, implying that the PASI absolute effect distributions of these populations overlap. This is not unexpected because of the limited number of existing studies in the population of children and young people.

The adjustment for population resulted in similar treatment rankings for children and young people when compared with the whole population results (see Table 47). The pooled placebo response rate for children and young people was estimated to be higher than that for adults (PASI 75: 12%, 95% CrI 5% to 20% in children and young people vs. 5%, 95% CrI 4% to 6% in adults), reflecting the higher placebo response rates observed in the trials in children and young people. This affects the efficacy of treatments by substantially increasing the estimated absolute PASI response rates across all treatments, but affecting the relative effects to a smaller extent. On average, PASI 75 response rates were estimated to be 10–15% higher in children and young people than in adults. The treatment rankings, however, remained unchanged. This is consistent with clinical opinion, with efficacy rates expected to be generally higher in children and young people than in adults as the biological interventions tend to work better in individuals with a lower body weight. Also, children and young people tend to have fewer comorbidities and generally have a greater exposure to ultraviolet (UV) light from participating in outside activities. The CrIs for PASI 75 response for children and young people and adults overlap, as shown in Figure 7.

FIGURE 7.

Absolute PASI 75 probability of response for children and young people and adults from NMA model 2b. ADA, adalimumab 0.8 mg/kg, maximum 40 mg/week; ETA, etanercept 0.8 mg/kg, maximum 50 mg/week; MTX, methotrexate 0.1–0.4 mg/kg/week; PLB, placebo; UST 45, ustekinumab 0.75 mg/kg or 45 mg/week.

The results of analysis 2b in children and young people suggest that ustekinumab is the most effective intervention with the highest mean probability of PASI response (PASI 75: 82%, 95% CrI 71% to 90%), followed by adalimumab (PASI 75: 79%, 95% CrI 64% to 90%), etanercept (PASI 75: 54%, 95% CrI 39% to 69%) and methotrexate (PASI 75: 49%, 95% CrI 31% to 68%). The relative efficacy of ustekinumab and adalimumab is similar based on relative effectiveness estimates for PASI 75 response (ADA vs. UST 45: RR 0.96, 95% CrI 0.85 to 1.05). In children and young people, ustekinumab (RR 1.54, 95% CrI 1.28 to 1.92) and adalimumab (RR 1.47, 95% CrI 1.23 to 1.79) are statistically significantly more effective than etanercept.

A consistency assessment was undertaken that involved excluding the trials of children and young people from the evidence network. This assessment indicated that the results were consistent across populations (see Appendix 7 for further details).

Overview

Given that the literature review identified only one unpublished model assessing the cost-effectiveness of etanercept in children and young people and that this model was adapted from TA10397 in adults and largely populated with adult data, additional hand-searching of published documents associated with the previous NICE TAs of plaque psoriasis in adults was carried out. The aim was to examine existing decision-analytic models to identify important structural assumptions, highlight key areas of uncertainty and outline the potential issues associated with generalising evidence from the adult population to a population of children and young people.

The first NICE TA on biological therapies for the treatment of psoriasis was a MTA examining the cost-effectiveness of etanercept and efalizumab (Raptiva^®, Serono Europe Ltd, London, UK) within their licensed indications in adults (TA103,97 published in July 2006). As part of this appraisal, the York Assessment Group developed a de novo cost-effectiveness model, which was subsequently referred to as ‘the York model’. Five subsequent STAs followed TA103:

1. TA134 – Infliximab for the Treatment of Adults with Psoriasis (published in January 2008)98

2. TA146 – Adalimumab for the Treatment of Adults with Psoriasis (published in June 2008)99

3. TA180 – Ustekinumab for the Treatment of Adults with Moderate to Severe Psoriasis (published in September 2009)100

4. TA350 – Secukinumab for Treating Moderate to Severe Plaque Psoriasis (published in July 2015)101

5. TA368 – Apremilast for Treating Moderate to Severe Plaque Psoriasis (published in November 2015). 102

All of these STAs employed a similar modelling approach to that used in the original York model in TA103. 97 The only study identified that deviated from the original model was the most recent STA of apremilast (TA368102), which included the modelling of sequences of treatment. Therefore, the main differences between the TAs lie in the evidence base, intervention and comparators rather than there being any major structural differences in the modelling approach used. A summary of the York model and the key differences between the assumptions and evidence base used in subsequent adaptations of the model are described in the following section. Table 51 provides an overview of the NICE TAs in adults.

Summary of the York model (TA103) and subsequent adaptations (TA134, TA146, TA180, TA350 and TA368)

The York model was a cohort Markov model that was developed to estimate the costs and QALYs of etanercept and efalizumab compared with BSC over a time horizon of 10 years (primary analysis). 97 A secondary analysis was also conducted to compare the interventions with additional systemic therapies of ciclosporin, Fumaderm^® (combination of fumaric acid esters), methotrexate and infliximab. The model adopted the perspective of the UK NHS. The price year for costs was 2004–5 and an annual discount rate of 6% for costs and 1.5% for outcomes was applied (in line with NICE guidance142 at the time of the appraisal).

The model consisted of a two-part structure: a ‘trial’ period for initial response and a ‘treatment’ period for long-term response to treatment (Figure 8). The initial response period was used to determine initial response rates and the decision to continue treatment. The duration of the trial period was based on the period over which response was assessed in the efficacy trials for each treatment – this was 12 weeks for etanercept and efalizumab and between 10 and 16 weeks for the other systemic therapies. Individuals with a PASI 75 response were considered ‘responders’ and continued treatment after the trial period (i.e. they entered the treatment period), whereas individuals who were ‘non-responders’ discontinued treatment and received BSC. The treatment duration for responding individuals was based on an annual withdrawal rate of 20%. On withdrawal, individuals were assumed to receive BSC.

FIGURE 8.

Structure of the York model. Source: TA103. 97 © NICE 2016 Etanercept and Efalizumab for the Treatment of Adults with Psoriasis. Available from www.nice.org.uk/guidance/ta103. All rights reserved. Subject to Notice of rights. NICE guidance is prepared for the National Health Service in England. All NICE guidance is subject to regular review and may be updated or withdrawn. NICE accepts no responsibility for the use of its content in this product/publication.

© NICE 2016 Etanercept and Efalizumab for the Treatment of Adults with Psoriasis. Available from www.nice.org.uk/guidance/ta103. All rights reserved. Subject to Notice of rights. NICE guidance is prepared for the National Health Service in England. All NICE guidance is subject to regular review and may be updated or withdrawn. NICE accepts no responsibility for the use of its content in this product/publication.

The base-case analysis considered a single line of therapy consisting of a biological treatment (etanercept or efalizumab) followed by BSC. Although specific sequences of treatments were not considered in the York model, an analysis showing the expected costs and QALYs associated with each treatment option compared with BSC was used to determine the most ‘cost-effective order’ in which to give the treatments, which varied according to the cost-effectiveness threshold.

The same model structure based on a single line of treatment was used in the subsequent TAs – TA134,98 TA146,99 TA180100 and TA350101 – and to a lesser extent in TA368,102 in which it was used only for the subpopulation of patients with a DLQI score of ≤ 10. The only difference in the modelling approach across the appraisals was a variation in the cycle length of the Markov model (12 months in TA134,98 TA14699 and TA350;101 3 months in TA180;100 and 28 days in TA368102), which was adapted to reflect the different length of the trial periods when treatment response was assessed. All appraisals used a time horizon of 10 years in their base-case analysis and used additional scenarios to show the implications of a change in the time horizon.

In TA368,102 the company adapted the model structure to allow a comparison of treatment sequences, with up to five sequential lines of treatment. The model structure followed the same approach as in the York model but, if the treatment response was considered inadequate at the end of the trial period, individuals moved into the trial period of the next line of treatment (or to BSC if the end of the treatment sequence had been reached). The company’s original economic model considered only apremilast as the first treatment in the sequence and compared different treatment sequences with apremilast as an additional line of therapy, rather than replacing an existing biological therapy in the sequence. However, following the ERG critique additional analyses were presented that compared the use of apremilast at different positions within a sequence.

Clinical effectiveness evidence in the York model and subsequent appraisals

The response rates used in the York model were based on a Bayesian NMA comparing the interventions with a broad range of comparators, including systemic therapies. An ordered probit model was used to predict PASI 50, PASI 75 and PASI 90 response rates, with PASI 75 used as the primary measure of response at the end of the trial period. If a trial reported only PGA 0/1 response (clear or almost clear) as the end point, it was assumed to be equivalent to the PASI 75 response. A similar Bayesian NMA was used in subsequent appraisals but was updated with additional evidence as more interventions and comparators became available. There were some differences across the appraisals in terms of how heterogeneity was accounted for in the meta-analysis and whether or not any adjustment had been made for differences in placebo response rates across the trials.

The effectiveness data were considered to be an area of uncertainty in the previous appraisals,97–102 mainly because of the lack of direct head-to-head comparisons between the biological treatments and the paucity of longer-term data. Although the evidence base expanded over time with many more RCTs included in the NMA, other concerns were raised relating to differences between the trial populations in the network (e.g. exposure to previous therapies and severity of disease). The definition of placebo or BSC across the different trials included in the NMA was also a contentious issue.

Health-related quality of life in the York model and subsequent appraisals

The utility values associated with treatment in the York model were based on the proportion of patients in the different PASI response categories (< 50, 50–74, 75–89, ≥ 90) and the change in utility from baseline associated with the PASI response category. Utility values were estimated based on a two-stage process:

1. Mean change in DLQI score between baseline and week 12 in the etanercept trials was estimated for patients with different levels of PASI response and different baseline DLQI scores. This analysis was facilitated by access to patient-level data from the trials and the placebo and treatment groups were pooled.

2. The DLQI data collected in the etanercept trials were then mapped onto EQ-5D values. This was achieved through access to data from the Health Outcomes Data Repository (HODaR), which included patients who had completed both the DLQI and EQ-5D. These data were used to map the change in DLQI score associated with PASI responses to changes in EQ-5D utility values.

The two-stage process was used to estimate average EQ-5D gains in utility from baseline for the different PASI response categories: 0.05 for PASI < 50, 0.17 for PASI 50–74, 0.19 for PASI 75–89 and 0.21 for PASI ≥ 90. Estimated gains in utility were also presented for individuals in the fourth quartile of the baseline DLQI score, that is, for patients with the worst baseline quality of life. The utility values from the York model were applied directly in TA368102 for the population with a DLQI score of > 10 and in TA13498 (values for the fourth quartile of the baseline DLQI score). For TA146,99 TA350101 and TA368102 (scenario analysis), the company had access to EQ-5D data collected in the trials, which were pooled across treatment groups and reported by PASI response category. For TA180,100 a similar modelling approach was used but the mapping algorithm used in the York model was applied to ustekinumab trial data to generate utility gains by PASI response category based on DLQI scores in the trial. The utility values applied in the TAs are summarised in Table 52.

None of the TAs included a disutility associated with AEs from treatment. Only one TA considered a disutility from flare-ups associated with time off treatment (intermittent etanercept) (TA14699); however, the disutility value applied was not reported in the published documentation.

The modelling approach used in the previous TAs assumed that:

• The PASI response is a perfect proxy for the change in utility arising from treatment. In other words, by conditioning on PASI response, utility is independent of treatment.

• Similarly, if utility is conditioned on DLQI score change, then utility is independent of PASI response.

• The relationship between DLQI score and utility is linear.

• The impact of AEs on HRQoL is unimportant.

The main critique of the approach used in the York model relates to the uncertainty introduced by mapping from the DLQI to the EQ-5D, based on a small sample of 86 patients. The NICE Appraisal Committees favoured the use of EQ-5D data collected directly from the trials when available. Scenario analysis in TA10397 showed that the cost-effectiveness results were very sensitive to the selection of utility values, with greater QALY gains from treatment in the fourth quartile of the baseline DLQI score (subgroup with the worst baseline HRQoL) than in the overall trial population. Furthermore, as the utility values were conditioned on the PASI response at the end of the trial period, these values were extrapolated over the time horizon of the model, which is a key driver of differences between the treatments.

Resource use and costs in the York model and subsequent appraisals

Resource use and costs included in the York model and subsequent appraisals related to drug acquisition, administration and monitoring, outpatient visits and inpatient hospitalisation stays. The cost of tests to assess eligibility for biological treatment was excluded. With the exception of infliximab, all treatments were assumed to be self-administered. Drug costs were sourced from the most recent information in the British National Formulary (BNF). A range of monitoring and laboratory costs were considered, including for a full blood count, liver function tests and regular physician visits. No AE costs associated with treatment were included in the York model. TA350101 was the only appraisal in which the costs of AEs were considered. The rates of AEs applied were sourced from the secukinumab trials and published literature and included non-melanoma skin cancer (NMSC), other malignancies and severe infections.

Resource use and costs included in the earliest TAs (TA134,98 TA14699 and TA180100) mostly followed the assumptions of the York model and sourced their resource use and unit costs from TA103. 97 There were small differences in resource use for drug monitoring and administration between the TAs but these differences had only a minor impact on the cost-effectiveness results. Later TAs (TA350,101 TA368102) based their resource use and cost estimates on the NICE clinical guideline on psoriasis (CG15311) for the cost-effectiveness of second-line biologics and on the accompanying costing report. CG153 included the same categories of costs as the York model but expanded on them to better characterise the costs of BSC. The costs associated with BSC were identified as a key driver of the cost-effectiveness results in TA10397 and were considered to be an area of substantial uncertainty in subsequent TAs.

In all of the appraisals, non-responders to treatment were assumed to receive BSC (with the exception of TA368,102 in which the sequential use of treatments was considered and non-responders moved to BSC only when all other treatment options were exhausted). The costs associated with BSC differ across the appraisals as there was no clear guidance on what BSC consists of. Table 53 provides a summary of the resource use and costs relating to BSC included in each of the TAs, as well as those reported in CG153.

The cost of BSC for non-responders in the York model was limited to two annual outpatient visits in the base-case analysis, but included one annual hospitalisation with a 21-day length of stay (LOS) in a scenario analysis. The rate of hospitalisations for patients on BSC was based on expert opinion, with LOS sourced from Hospital Episode Statistics (HES) data and two local surveys. Subsequent TAs (TA134,98 TA14699 and TA18098) used the 21-day inpatient stay for BSC in their base-case analysis. In each of these appraisals it was assumed that no treatments were given as part of BSC.

National Institute for Health and Care Excellence CG15311 departed from this definition of BSC (see Appendix P of the guideline) because it was believed that it does not reflect what currently happens in clinical practice for patients who require a second-line biologic. In CG153 it was assumed that 45% of patients would receive treatment with methotrexate, 45% would receive continuous ciclosporin for a maximum of 2 years and 16% would undergo 24 sessions of NBUVB therapy per year while on BSC. Furthermore, it was reported in CG153 that patients meeting the eligibility criteria for biological therapy are generally high-need patients who use a sizeable number of health-care resources through inpatient admissions, lengthy hospital stays and frequent visits to day clinics for specialist-applied topical treatments and ultraviolet B (UVB) and who require monitoring for toxicity because of the use of systemic treatments. The NICE Guideline Development Group (GDG) sourced the resource use estimates for BSC from two published cohort studies of patients with a high level of need (i.e. those with severe psoriasis), which were conducted in tertiary dermatology units in the UK (n = 76)143 and the Netherlands (n = 67). 144 Both of these studies estimated the mean number of inpatient days in the year preceding initial treatment with biological therapy. In addition, estimates of LOS from a multicentre prospective service review based on four specialist dermatology centres in the UK145 were used in a scenario analysis. The GDG for CG153 emphasised that there is substantial variability in the long-term costs reported for patients with psoriasis. As a result, CG153 included extensive sensitivity analyses for the elements of cost associated with BSC. These included variations in the number of hospitalisations per year and the average LOS by level of need. The cost-effectiveness of second-line biological therapy compared with BSC in CG153 was highly sensitive to the assumptions about BSC.

The more recent TAs (TA368102 and TA350101) largely followed the resource use reported in CG153 for BSC (Table 54). The NICE Appraisal Committees for TA350 and TA368 noted that in both cases the costs of BSC were likely to have been overestimated. The committees considered that the patient population in CG153 and Fonia et al. 144 did not match that in the appraisals and reflected a sicker group of patients. In particular, Fonia et al. 144 described care in a tertiary care centre, which is known to be associated with treating the most severely affected cases of psoriasis. Furthermore, during the consultation process for TA368,102 the company provided NHS HES data that showed that the average LOS associated with BSC was 3.5 days. However, this was argued by the company to be an underestimate as it included patients with different disease severities and patients receiving concomitant medication. The duration of hospital stay for BSC in adults with moderate to severe psoriasis remains highly uncertain.

Cost-effectiveness results from the York model and subsequent appraisals

The cost-effectiveness results for the base-case analyses in the previous NICE TAs in adults are summarised in Table 55, alongside the drivers of cost-effectiveness stated in the TA documentation. The results reported for TA134,98 TA146,99 TA180,100 TA350101 and TA368102 correspond to those in the company submissions.

In the base-case full incremental analysis for TA103,97 which compared etanercept in three dosing regimens (25 mg intermittent, 25 mg continuous and 50 mg intermittent), efalizumab and BSC, and assuming no hospitalisations for non-responders to biological treatment, BSC was the most cost-effective strategy at cost-effectiveness thresholds of < £66,703 per QALY gained. At a threshold of ≥ £66,703 per QALY gained, intermittent etanercept at 25 mg would be the cost-effective intervention, dominating (i.e. being less costly and more effective than) continuous etanercept at 25 mg and efalizumab. Intermittent etanercept at 50 mg had an incremental cost-effectiveness ratio (ICER) exceeding £1M per QALY gained compared with etanercept at a lower dose (25 mg intermittent).

Inclusion of 21 days of hospitalisation for non-responders to the biological drugs reduced the ICER for intermittent etanercept at 25 mg compared with BSC to £29,420 per additional QALY. When, in addition to the 21 days of hospitalisation, the estimates of utility gains per PASI response were sourced from the subgroup of patients in the highest (worst HRQoL) quartile of the baseline DLQI score (the group with the highest gain in utility from improvement in PASI score), the ICER for intermittent etanercept at 25 mg compared with BSC further reduced to £15,297 per QALY. In both of these scenario analyses, continuous etanercept at 25 mg and efalizumab remained dominated by intermittent etanercept at 25 mg, whereas the ICER for intermittent etanercept at 50 mg compared with intermittent etanercept at 25 mg decreased but not enough to make it cost-effective at commonly accepted cost-effectiveness threshold ranges. In the secondary analysis that compared the full range of systemic therapies (namely infliximab, methotrexate, ciclosporin and Fumaderm) and assumed 21 days of hospitalisation for non-responders, methotrexate dominated all interventions with the exception of infliximab, including BSC. Infliximab was more costly and more effective than methotrexate but the resulting ICER for this comparison exceeded £1M per QALY gained, with methotrexate emerging as the cost-effective intervention for this analysis.

The appraisals subsequent to TA103 (TA134,98 TA146,99 TA180,100 TA350101 and TA368102) all included a cost associated with hospitalisation for non-responders (LOS ranging from 10.7 to 26.6 days per annum). This generally resulted in more favourable cost-effectiveness estimates when biological therapies were compared with BSC. Consistent with the findings of the York model, the duration of hospitalisation for non-responders was identified as a key driver of cost-effectiveness for biological therapies across the appraisals. The base-case analysis in the majority of the appraisals98–100 used estimates of utility gains by PASI response in a subgroup with lower baseline HRQoL (TA134, TA146 and TA180), leading to higher QALY gains for the most effective drugs in terms of PASI response. This parameter can be considered the second driver of cost-effectiveness in the adult models.

The base-case cost-effectiveness results in the company submissions for infliximab, adalimumab and ustekinumab (TA134,98 TA14699 and TA180100 respectively) place the ICERs for these drugs at the upper end of the currently accepted NICE cost-effectiveness threshold range, as long as the assumptions about HRQoL and the costs of hospitalisation for non-responders referred to above hold. The estimates of cost-effectiveness for secukinumab and apremilast presented by the manufacturers in TA350101 and TA368102 were considered overly optimistic by the NICE Appraisal Committees and were largely driven by the costs of BSC in non-responders to biological therapy. These costs were considerably higher than in previous appraisals because of the assumption of a higher consumption of health-care resources by non-responders, in line with CG15311 (see Table 54).

The National Institute for Health and Care Excellence recommended the following biological treatments in adults with psoriasis: efalizumab,97 etanercept,97 infliximab,98 adalimumab,99 ustekinumab100 and secukinumab. 101 With the exception of infliximab the biological treatments were recommended for severe psoriasis, defined as a baseline PASI score of ≥ 10 and a DLQI score of > 10 in patients who had previously failed or who had a contraindication/intolerance to non-biological systemic therapy. The recommendation for efalizumab further required that patients had failed on etanercept or had a contraindication/intolerance to the drug. Efalizumab is no longer marketed in the UK.

Infliximab was recommended only for very severe psoriasis, defined as a baseline PASI score of ≥ 20 and a DLQI score of > 18 in patients who had previously failed or who had a contraindication/intolerance to non-biological systemic therapy. The recommendation for ustekinumab and secukinumab was conditional on the availability of Patient Access Schemes. The Patient Access Scheme for ustekinumab guarantees a flat price for 45 mg and 90 mg of ustekinumab so that the 90-mg dose is provided at the same price as the 45-mg dose for patients weighing > 100 kg, whereas the Patient Access Scheme for secukinumab consists of a confidential discount over the drug list price.

The NICE recommendations for all of these biological treatments require treatment termination if a response is not produced at the end of the ‘trial’ period (12 weeks for etanercept and secukinumab, 10 weeks for infliximab and 16 weeks for adalimumab and ustekinumab). Treatment response is defined as achieving a PASI 75 or PASI 50 response accompanied by a 5-point reduction in DLQI score from baseline.

Review of utility data in children and young people with psoriasis

A systematic literature review was conducted to identify utility values for plaque psoriasis in children and young people. The aim of the search was to identify any studies that reported utility values or other measures of HRQoL that could be converted into utility values specifically for the population of children and young people.

The search strategy was developed in MEDLINE (via Ovid) by an information specialist with input from the project team. The strategy included terms for psoriasis combined, using the Boolean operator AND, with terms for quality of life/utilities or named instruments. No language, geographical or date limits were applied. A search filter to limit retrieval to quality-of-life studies was used when available. The search strategy was adapted for use in the other resources searched. Full search strategies can be found in Appendix 1.

The following databases were searched on 12 July 2016: MEDLINE [including MEDLINE Epub Ahead of Print, MEDLINE In-Process & Other Non-Indexed Citations, Ovid MEDLINE(R) Daily and Ovid MEDLINE(R)], the Cost-effectiveness Analysis (CEA) Registry [see https://research.tufts-nemc.org/cear4/ (accessed 10 August 2017)] EMBASE, PubMed and the School of Health and Related Research Health Utilities Database (ScHARRHUD) [see www.scharrhud.org/ (accessed 10 August 2017)].

The Health Economics Research Centre (HERC) database of mapping studies from the University of Oxford152 was also searched to identify any suitable mapping algorithms that would allow conversion of clinical measures routinely collected in studies of psoriasis in children and young people into utility values.

The results from the searches were imported into an EndNote X7 library and deduplicated. After deduplication, 286 records in total were identified. The titles and abstracts were assessed independently by two reviewers for inclusion and any discrepancies were resolved by consensus. None of the titles identified reported utility values collected in children and young people with psoriasis. The search of the HERC mapping algorithm database identified one study on the development of a mapping algorithm to estimate EQ-5D-Y (EuroQol-5 Dimensions for Youth) utility values from PedsQL general core scales. 153 The PedsQL is a generic instrument for measuring HRQoL in children and adolescents with acute and chronic health conditions. The PedsQL measures core dimensions of health as delineated by the WHO, as well as role (school) functioning. The four multidimensional scales are physical functioning (eight items), emotional functioning (five items), social functioning (five items) and school functioning (five items). 19 The EQ-5D-Y is the youth version of the EQ-5D, which has been specifically adapted in terms of language for children aged 8–11 years and adolescents aged 12–18 years. In the absence of a tariff set specifically for the EQ-5D-Y, the authors153 applied the UK national tariff for the valuation of the EQ-5D-3L (EuroQol-5 Dimensions three-level version) to generate utility values from the EQ-5D-Y instrument. 154

Khan et al. 153 assessed different mapping methods for estimating EQ-5D-Y health utilities from PedsQL response scores. The study used data collected in a cross-sectional survey conducted in four secondary schools in England among children aged 11–15 years. The sample on which the mapping models were estimated included 559 children and the validation sample included 337 children. Children in the full study sample (n = 896) were on average aged 13.3 years (standard deviation 1.3 years), 54% were male and approximately 40% were of non-white ethnicity. The authors explored both direct and response mapping approaches to predict EQ-5D-Y utility values, as well as a number of functional forms, including ordinary least squares (OLS) regression, generalised linear models, two-part logit–OLS regression, censored least absolute deviation and Tobit regression. Model performance was assessed on the validation sample and models were re-estimated on the full study sample. Table 58 presents the two best-fitting models for the mapping algorithm. These correspond to the models estimated using OLS regression with (1) age and sex terms included as regressors and (2) excluding age and sex terms as regressors.

The two models were considered to have similar prediction accuracy for mean EQ-5D-Y values. Model 1, which included age and sex as regressors, had a better fit across a wider range of EQ-5D-Y values than model 2. Model 2 reported a better fit for the EQ-5D-Y utility score range of 0.8–1.0 category. There are a number of potential limitations to the use of this algorithm to predict EQ-5D-Y utilities. The sample on which the models were estimated consisted of healthy children aged 11–15 years, which may limit the predictive accuracy in sicker populations or populations outside this age range. The authors recognise that there is a need for further validation and testing of the algorithm but, in the absence of an alternative source, it remains a useful tool for estimating EQ-5D-Y utility values in situations in which only the PedsQL has been administered.

Utility data reported in company submissions

Health-related quality-of-life assessments were carried out in the etanercept trial (2003021149), the ustekinumab trial (CADMUS72) and the adalimumab trial (M04-71742) using the CDLQI and the PedsQL at selected time points. EQ-5D or EQ-5D-Y values were not collected in any of the trials. Therefore, the only way to include EQ-5D utility values in the model was by mapping from either CDLQI scores or PedsQL scores. The literature review described earlier did not identify any studies that estimated the relationship between CDLQI scores and EQ-5D values, whereas the study by Khan et al. 153 was the only study that estimated the relationship between PedsQL scores and EQ-5D values. The AG requested from the companies access to individual patient data (IPD) for PedsQL domain scores at baseline and follow-up by category of PASI response and PedsQL summary scores at the domain level by response category. The AG did not receive access to IPD; however, Janssen (ustekinumab) submitted aggregated summary data (mean and standard deviation) from the CADMUS trial for the PedsQL subscale and total scale scores by treatment arm (placebo and ustekinumab standard dose) and PASI response category at 12 weeks (< 50, 50–74, 75–89, ≥ 90) for baseline and 12, 28 and 52 weeks.

Utility estimates used in the model

The utility values associated with treatment in previous models in adults were based on the proportion of individuals in the different PASI response categories (< 50, 50–74, 75–89, ≥ 90) and the change in utility from baseline associated with the PASI response categories. Therefore, PASI response rates from the NMA were assumed to be a perfect proxy for change in utility arising from treatment.

The relationship between utility and PASI response was estimated in previous TAs in adults using either DLQI data mapped onto EQ-5D utility values or directly from EQ-5D data collected in the trials. In the population of children and young people, the only possibility of obtaining EQ-5D values was by mapping from PedsQL scores to EQ-5D-Y values, as described earlier. Without access to the IPD, which would allow full uncertainty to be reflected in the values, the mapping algorithm was applied to the summary scores at the domain level from the CADMUS trial.

Validation of the algorithm was performed by examining data reported in a study by Varni et al. ,155 which compared self-reported HRQoL (based on PedsQL scores) between paediatric patients with moderate to severe plaque psoriasis and a healthy population sample. The sample used to represent the psoriasis population corresponded to individuals in the main efficacy trial for etanercept (n = 208, age 4–17 years) and measurements of PedsQL scores at baseline were pooled across the two treatment arms (etanercept and placebo). The healthy population sample was taken from a US children’s health insurance programme evaluation (n = 5079) open to children and young people aged 2–16 years. Table 59 summarises the PedsQL subscale scores reported in Varni et al. 155 for the psoriasis and healthy populations, alongside the estimates obtained by applying the two best-fitting mapping algorithms to obtain EQ-5D-Y utility values (model 1 includes age and sex terms whereas model 2 excludes these variables).

The EQ-5D utility estimates were higher in the healthy population than in the population with psoriasis, irrespective of the model used to map PedsQL scores to EQ-5D-Y values. The distinction between the models was minimal, especially in the psoriasis population: model 1 provided slightly higher utility values than model 2 (0.6% and 2.5% higher in the psoriasis and healthy populations respectively). Model 2 was subsequently used in the base-case analysis as the reference category for the variable sex was unclear in the study by Khan et al. 153

The mapping algorithm (model 2 in Table 58) was used to estimate change in EQ-5D-Y utility values from baseline based on PedsQL data from the ustekinumab trial (CADMUS) at baseline and 12 weeks’ follow-up (the time point at which response to treatment was assessed in the trial and blinding of randomised subjects in the trial was terminated; after this point crossover between treatment arms was possible). The mean change in EQ-5D-Y values between baseline and week 12 was estimated for individuals with different levels of PASI response. Table 60 reports the EQ-5D-Y utility values estimated for the base-case-analysis, with the placebo and treatment arms pooled.

The baseline utility estimate is similar to that derived from the etanercept trial (20030211) (0.864 in Table 59) and is lower than the general healthy population estimate of 0.913 based on the study by Varni et al. 155 The mean changes in EQ-5D utility from baseline by PASI response category are much smaller than the corresponding changes in EQ-5D utility observed in previous TAs in adults. For example, the EQ-5D changes in utility by PASI response category in the York model of adults were 0.05 for PASI < 50, 0.17 for PASI 50–74, 0.19 for PASI 75–89 and 0.21 for PASI ≥ 90.

To examine whether or not the changes in EQ-5D-Y utility values were accompanied by similar changes in other measures of HRQoL in the population of children and young people, the changes in EQ-5D-Y values were compared with reported CDLQI values by PASI response (Table 61). A comparison of EQ-5D and DLQI values by PASI response in adults is also shown in Table 61 (taken from TA180100 for ustekinumab, which was the only TA in adults that reported both outcomes). The mean changes in CDLQI score by PASI response in the CADMUS trial are much smaller than the mean changes in DLQI score in TA180,100 which is consistent with the smaller mean changes estimated for the EQ-5D-Y in the paediatric population than for the EQ-5D in adults. These differences, however, should be interpreted with caution as CDLQI and DLQI scores are not directly comparable and the number of observations was much smaller in the population of children and young people than in the adult population in TA180. 100

The EQ-5D-Y/EQ-5D utility estimates suggest that improvements in HRQoL associated with reductions in PASI response rates are of a much smaller magnitude in children and young people than in adults; however, the evidence is highly uncertain because of the small sample size and the limited data available to validate the findings. In the absence of an alternative source to estimate EQ-5D values for the model, these values were used in the base-case analysis. It is important to highlight a number of limitations of this approach. First, the use of a mapping algorithm to estimate utilities introduces uncertainty compared with direct EQ-5D measurement. Second, the Khan et al. 153 mapping algorithm has not been validated in children aged < 11 years or in a population with psoriasis. Third, the CADMUS trial, from which the PedsQL data that were mapped to EQ-5D utilities were sourced, excluded children aged < 12 years; therefore, it remains uncertain whether or not the mapped utilities are reflective of this population. Fourth, in populations aged < 12 years, there may be issues with lack of agreement or consistency between self-reported and proxy (parent)-reported measurements. 156 Therefore, even if PedsQL data were available for younger children, the mapping algorithm might not consistently perform for self-reported and parent-reported measurements of the instrument. Finally, Khan et al. 153 used the EQ-5D-3L value set as a proxy for EQ-5D-Y in the absence of an alternative tariff set, but this approach is currently not recommended. 157 These limitations reduce the robustness of the utility estimates used in the model.

There might be other potential benefits of treatment that fall outside the QALY estimation. First, children and young people may miss schooldays to attend health-care appointments and may be absent for longer periods from school while experiencing symptoms. This can have a negative impact on their education/academic achievements and, in future, their ability to gain employment. It may also affect their social and psychological health through the reduced ability to participate in social and leisure activities and sport. Second, early treatment of children and young people with biological agents may prevent long-term multisystem morbidity (e.g. hypertension, cardiovascular disease, depression), which has a higher prevalence in adults with psoriasis than in the general population. 158 Finally, there may also be other aspects of HRQoL that are outside the perspective defined by NICE’s reference case,150 namely the potential impact on the HRQoL of carers of children and young people with psoriasis if treatment with biologics reduces the spillover disutility of illness by improving patients’ outcomes. The impact on carers may also extend to a reduced ability to participate in normal activities, both work- and non-work related. Because of an absence of quantitative estimates of the impact on the HRQoL of children and young people with psoriasis receiving any of the interventions and the potential benefits to their carers, it was not possible to incorporate them into the economic analysis. Any attempt to add arbitrary values to the utility estimates, which are already highly uncertain, would introduce further uncertainty.

Given the uncertainty surrounding the utility estimates for children and young people, scenario analyses were conducted using utility estimates from previous TAs in adults for etanercept, adalimumab and ustekinumab. Table 62 summarises the utility estimates considered in the scenario analyses.

Drug acquisition costs

Table 64 details the dose and frequency of administration for each treatment and comparator, including ciclosporin, which forms part of BSC, and the unit costs associated with each treatment.

The dosages of the biological therapies were taken from the Summaries of Product Characteristics. 167–169 For methotrexate and ciclosporin, which are currently not licensed for paediatric use, the dosages were sourced from published literature78,170,171 and confirmed with our clinical advisor to ensure that they reflected UK clinical practice in this population. Methotrexate can be administered orally or injected subcutaneously or intramuscularly. In the model it was assumed that 72% of individuals are given methotrexate in oral solution and 28% in injectable solution, which reflects the distribution of administration identified in the UK psoriasis audit of the use of systemic treatments in children and young people. 148 Therefore, the unit cost per mg for methotrexate is a weighted average of the unit cost per mg of the oral and injectable solutions (i.e. £0.71/mg). Unit costs were sourced from MIMS160 and supplemented with data from the BNF. 161

Figure 11 illustrates the number of doses administered in the first five cycles of the model for each treatment based on the licensed dose. Adalimumab is administered at weeks 0 (baseline) and 1 and then every 2 weeks thereafter until response assessment at the end of week 16. If individuals are responders to treatment they continue to receive adalimumab every 2 weeks until treatment withdrawal (highlighted in grey). Ustekinumab is administered at weeks 0 and 4 and then every 12 weeks thereafter, with response assessment at week 16. Etanercept and methotrexate are administered weekly, with response assessment at weeks 12 and 16 respectively.

FIGURE 11.

Drug dose distribution during the first five cycles in the model. ADA, adalimumab; ETA, etanercept; MTX, methotrexate; UST, ustekinumab; •, ADA administration; ♦, ETA administration; ▪, UST administration; Δ, MTX administration.

The dosages of the biological treatments are dependent on patient weight. The median weight by age and sex in the population of children and young people was extracted from the Royal College of Paediatrics and Child Health’s school-age growth charts. 172 Table 65 shows the weights used in the model by age. These were based on an average of the weight of boys and girls (and when the weight estimate in the growth chart did not correspond to an integer, the next-highest integer was used).

The weight by age was used to estimate the correct dosage of each treatment and the corresponding cost. Table 66 summarises the dosages used in the model for each treatment by age and the corresponding costs per dose. Following clinical advice it was assumed that the vial with the lowest dose available would be used to allow administration of a single dose in the paediatric population. This inevitably results in wastage of the remainder of the vial. For example, for individuals who weigh < 60 kg the full cost of the 45-mg vial of ustekinumab is assumed as the remaining product in the vial cannot be stored. Vial splitting across individuals was considered unlikely because in most cases the majority of the vial is used for a single patient and treating patients together is less likely to occur in this population because of low patient numbers. Therefore, the cost per dose was fixed for adalimumab and ustekinumab (£352.14 and £2147.00 respectively). For etanercept, the 25-mg vial (£89.38) is used for children aged < 10 years whereas the 50-mg vial (£178.75) is used for those aged ≥ 10 years.