3 Evidence | Adalimumab for treating moderate to severe hidradenitis suppurativa | Guidance

Clinical effectiveness

3.1 The pivotal clinical evidence for treating hidradenitis suppurativa with adalimumab came from 2 randomised double-blind phase III trials (PIONEER I, n=307, and PIONEER II, n=326). The PIONEER trials compared adalimumab with placebo in adults who had been diagnosed with moderate to severe hidradenitis suppurativa at least 1 year earlier and who were intolerant to, or whose disease had not responded to, oral antibiotics. Moderate to severe disease was defined as people with Hurley stage II or III hidradenitis suppurativa in at least 1 affected anatomic region, and a total abscess and inflammatory nodule (AN) count greater than 3. Neither of the trials recruited people from the UK. Treatment with oral or topical antibiotics during the trial was allowed in PIONEER II but not in PIONEER I. Extensive surgical procedures were not allowed, but incision and drainage of lesions or corticosteroid injections directly into lesions were allowed. Supportive care interventions (such as tobacco cessation or weight-control counselling) were not given to anyone in the trials.

3.2 The primary end point in the PIONEER trials was the proportion of people with a Hidradenitis Suppurativa Clinical Response (HiSCR) at week 12. HiSCR is defined as at least a 50% reduction in the total AN count, with no increase in abscesses or draining fistulas. The secondary outcomes were: the proportion of people with Hurley stage II disease who had an AN count of 0, 1 or 2 at week 12; the proportion of people who had a 30% or more reduction, and at least 1 unit reduction, in the Patient's Global Assessment of Skin Pain from baseline to week 12; and change in the Modified Sartorius Score from baseline to week 12. EuroQol (EQ‑5D) data were only collected in PIONEER II. Other health-related quality-of-life instruments used in the PIONEER studies included the Short Form‑36 Health Status Survey (SF‑36; PIONEER I only), Dermatology Life Quality Index (DLQI), and Hidradenitis Suppurativa Quality of Life (HSQOL).

3.3 Both trials included 2 study periods and an open-label extension study:

Period A (12 weeks 'induction'): people were randomised to adalimumab 40 mg every week or placebo.
Period B (24 weeks 'maintenance'): people who had adalimumab 40 mg every week in period A were re-randomised to have either adalimumab 40 mg every week, adalimumab 40 mg every other week or placebo. In PIONEER I, people who had placebo in period A were re-randomised to adalimumab 40 mg every week, whereas in PIONEER II people who had placebo in period A stayed on placebo for period B.

Eligibility for period B depended on clinical response at the end of period A. People who had a clinical response (HiSCR) at week 12 were enrolled in period B until the end of week 36, but were excluded from the study if their condition stopped responding to treatment. People who did not have an HiSCR response at week 12 were enrolled in period B until week 16; if the severity of their hidradenitis suppurativa worsened or did not improve after week 16 they were excluded from the study. The open-label extension study included people who had completed PIONEER I or II and who:
had an HiSCR response at the end of period B
had an HiSCR response at the start of period B then experienced loss of response or
did not have an HiSCR response at the start of period B, then experienced worsening or absence of improvement on or after week 16.

3.4 The company indicated that baseline characteristics were generally similar in the different arms of the trials. But, people in PIONEER I had more severe disease than those in PIONEER II. The average duration of hidradenitis suppurativa in the trials was about 11.5 years.

3.5 More people treated with adalimumab had an HiSCR response than those having placebo; these differences were statistically significant in both PIONEER trials (table 1). The differences between adalimumab and placebo were statistically significant for all secondary outcomes at week 12 in PIONEER II (showing a benefit in favour of adalimumab), but none of the differences were significant at week 12 in PIONEER I. Pre-planned analyses showed that a consistent treatment effect was seen across most subgroups, with a few exceptions in subgroups with small sample sizes.

Table 1 Primary outcomes at week 12 for adalimumab 40 mg every week compared with placebo, from phase III randomised controlled trials

Trial	Intervention	People with clinical response, n (%)	Difference (95% CI)	p value
PIONEER I	Adalimumab (n=153)	64 (41.8%)	15.9% (5.3% to 26.5%)	0.003
	Placebo (n=154)	40 (26.0%)
PIONEER II	Adalimumab (n=163)	96 (58.9%)	31.5% (20.7% to 42.2%)	<0.001
	Placebo (n=163)	45 (27.6%)
Abbreviations: CI, confidence interval; n, number.

3.6 The company stated that the benefits seen with adalimumab at 12 weeks continued up to 36 weeks (period B) in the PIONEER studies. The company provided an interim analysis of the primary end point from the open-label extension study, noting that patient numbers were small. A post hoc analysis of pooled data from the PIONEER studies and the open-label extension study showed that the continued benefit of adalimumab was seen in people with a partial HiSCR response (defined as at least a 25% reduction in the total AN count with or without an increase in abscess count or draining fistula count), as well as people with a complete clinical response.

3.7 In PIONEER I and II, adalimumab was associated with significant improvements from baseline in health-related quality of life after 12 weeks of treatment. Adalimumab was associated with larger improvements from baseline than placebo; these differences were statistically significant, as measured by the EQ‑5D, the physical components of SF‑36, DLQI and the HSQOL. The difference between adalimumab and placebo in the mental component of the SF‑36 was not significant.

3.8 The company reported that the most common adverse events with adalimumab were worsening of hidradenitis suppurativa, nasopharyngitis and headache. These were usually mild to moderately severe. The company noted that during the first 12 weeks of both PIONEER studies, adverse events and treatment discontinuation caused by adverse events were less common in people treated with adalimumab than in people treated with placebo. The company reported that the open-label extension study did not identify any new safety risks for adalimumab.

Cost effectiveness

3.9 The company provided a Markov model to assess the cost effectiveness of adalimumab compared with supportive care. The company stated that it was not appropriate to compare adalimumab with any active pharmacological agents, because adalimumab would be used after all conventional systemic treatments (including antibiotics, dapsone, retinoids and immunomodulators). The company based the efficacy data for adalimumab on pooled data from the PIONEER trials (using an integrated arm-based summary). Efficacy data for supportive care were based on the placebo arms in the PIONEER clinical trials.

3.10 The model used a lifetime horizon, with a cycle length of 4 weeks (except for the first 2 cycles, which were each 2 weeks). All patients entered the model in the non-response health state and then transitioned between health states based on the responses of their disease to treatment and the natural mortality rate. Four of the health states were defined according to varying levels of HiSCR response:

high response: 75% or greater reduction in AN count with no increase in abscess count or draining fistula count
response: 50–74% reduction in total AN count with no increase in abscess count or draining fistula count
partial response: 25% or greater reduction in total AN count with or without an increase in abscess count or draining fistula count
non-response: less than 25% reduction in total AN count
death.

The high response and response health states together make up the complete HiSCR response. People in the partial response and non-response health states would have been classified as 'HiSCR non-responders' in the PIONEER trials. The company provided several justifications for splitting the HiSCR into 4 health states:
There was a statistically significant difference in the EQ‑5D utility values (collected in PIONEER II) between the high response and response health states (p=0.036), and between the partial response and non-response health states (p=0.034).
The difference in the response rates between adalimumab and placebo were statistically significant across 3 of the 4 response health states.
Resource use differed across the 4 health states.
A post hoc analysis of the PIONEER studies identified a population in which continued treatment with adalimumab could be beneficial (that is, people with a partial response or higher).

3.11 The level of HiSCR response at 12 weeks determined whether patients continued having adalimumab; people who had at least a partial response continued treatment. For patients who continued having adalimumab, there was an ongoing chance of stopping treatment at any time point:

Weeks 12–36: The company used rates from the PIONEER studies, based on people who had a response at 12 weeks, to estimate 4‑week discontinuation rates for the model. The company applied the same discontinuation rate to everyone having adalimumab, regardless of their response state.
Long‑term discontinuation (beyond 36 weeks): The company used data from the open-label extension study to estimate discontinuation rates specific to each response state. The company's application of discontinuation rates aimed to reflect its assumption that people in the non-response health state at 36 weeks would continue treatment for an additional 12 weeks, not stopping until 48 weeks, based on clinical advice and guidance in the adalimumab summary of product characteristics.

People who stopped adalimumab treatment (at either 12 weeks or later) were assumed to move on to supportive care.

3.12 The company estimated the transition probabilities between health states for the first 36 weeks of treatment using the distribution of people across the 4 response health states in the PIONEER clinical trials. The company imputed missing values using the same method specified in the clinical trial protocol for analysis of the primary end point (non-responder imputation). To extrapolate data beyond what was available from clinical trials (that is, beyond 36 weeks), the company used separate generalised logit models from different sources depending on the treatment:

For people who continued having adalimumab, the company used data from the open-label extension study and imputed missing values using last observation carried forward.
For people who stopped adalimumab treatment, the company used data from period B of the PIONEER I and PIONEER II trials (weeks 12–36) and missing values were imputed using non-responder imputation.
For people having supportive care, the company used data from period B of the PIONEER II trial (weeks 12–36) and missing values were imputed using non-responder imputation.

3.13 The company assigned utility values to each health state in the model using EQ‑5D data collected in the PIONEER II clinical trial (table 2). The model did not incorporate reductions in utility values (disutilities) from treatment-related adverse events. The company stated that this was likely to have a minimal effect on the results because the adverse-event rates were similar between people who had adalimumab and people who had placebo in the PIONEER clinical trials.

Table 2 EQ‑5D derived utility values in the company model (using data from weeks 12 and 36)

Model health state	Utility value	95% confidence interval	p value ^a
High response	0.782	0.746 to 0.816	0.036
Response	0.718	0.667 to 0.766	0.036
Partial response	0.576	0.512 to 0.639	0.034
Non-response	0.482	0.402 to 0.542	0.034
^a p values reflect the significant differences in utility values between the high response and response health states, and the difference between the partial and non-response states.

3.14 The company included the following costs in its model:

treatment costs
adverse-event-related costs, for adverse events with an incidence of 5% or more in the PIONEER trials
resource-use costs, assigned to each health state independent of the treatment, for inpatient stays, outpatient visits, visits to wound-care (each divided into surgery related and non-surgery related) and emergency department visits
one-off set-up costs (£0.70 per patient) and ongoing operational costs (£8.21 per 4‑week cycle) associated with the patient access scheme.

3.15 Adalimumab costs were based on the discounted price agreed by the Department of Health in the patient access scheme for adalimumab in hidradenitis suppurativa (see section 2.3). The company did not include any drug costs for supportive care because it considered that any of the conventional treatments taken by people having supportive care would also be taken, less often, by people having adalimumab. The company estimated resource use based on the results of a survey of 40 physicians who treat people with moderate to severe hidradenitis suppurativa in the UK, and obtained costs associated with each type of resource use from NHS reference costs 2013/14.

3.16 The company's original base-case deterministic cost-effectiveness analysis showed that adalimumab was more costly and more effective than supportive care. The results of the company's one-way deterministic sensitivity analyses showed that the incremental cost-effectiveness ratio (ICER) was sensitive to the assumptions about:

long‑term transition probabilities (after week 36)
number and cost of hospital admissions, specifically the surgery-related hospital admissions, especially in the non-response health state
utility values for partial and non-response health states.

The company stated that the ICER was relatively robust to any other changes in model inputs.

Key issues

3.17 The ERG noted that the benefit with adalimumab was greater in PIONEER II than PIONEER I for the primary and secondary outcomes, possibly because PIONEER II patients had less severe disease than people in PIONEER I. The ERG was concerned that the company had not done a formal meta-analysis of the PIONEER trials, and considered that the company's method of pooling data from trials, to inform the transition probabilities in the model, was inappropriate.

3.18 The ERG noted that although the differences between the improvements associated with adalimumab and the improvements with placebo were statistically significant for some health-related quality-of-life outcomes, they did not always exceed the minimum clinically important difference for the instrument. For example, the difference in change from baseline between adalimumab and placebo on the DLQI was 2.5 in PIONEER I (p<0.001) and 2.8 in PIONEER II (p<0.001); the minimum clinically important difference for the DLQI is 4.

3.19 The ERG had concerns about the company's assertion that adalimumab may delay or reduce the need for surgery, because it was not substantiated by empirical evidence. Based on a post hoc analysis of pooled data from the PIONEER studies, the company stated that a greater proportion of people who had adalimumab, compared with placebo, experienced improvement of both draining fistulas (33% compared with 19%; p<0.001) and non-draining fistulas (15% compared with 9%; p=0.017). The ERG was unclear whether this reduction in minor procedures fully reflected an overall reduction in surgery, particularly surgical-inpatient admissions, which were a key cost driver in the company's model.

3.20 Given that the HiSCR is a dichotomous outcome (that is, either a clinical response or not), the ERG had concerns about the company's decision to model 4 health states according to the different levels of HiSCR response. The ERG questioned whether the company's assumption that people continued treatment if their disease had a partial response or higher reflects what would happen in clinical practice; it suggested that this assumption, and the decision to model 4 response states, was not consistent with the primary end point in the adalimumab trials or the validation study of the HiSCR measure by Kimball (2014). The ERG was also concerned that dividing the efficacy data across 4, rather than 2, health states resulted in small sample sizes for the calculation of some transition probabilities, which could be considered as a structural uncertainty.

3.21 The ERG had concerns about the company using 1 source to model the benefits of treatment (the clinical trials) and another source to model the resource use needed to get these benefits (the physician survey), and was unsure about the appropriateness of specifying resource use according to different levels of HiSCR response.

3.22 The ERG had concerns about the uncertainty in transition probability estimates beyond week 12, attributed to the small sample sizes in the maintenance period of the trials. The ERG also questioned the robustness of long-term transition probabilities in the company model (beyond week 36), because the company calculated them using data from the open-label extension study. The ERG was concerned because these data:

were immature
might have produced optimistic estimates of treatment effect because of the company's method for imputing missing data
included people who did not reflect the modelled population
introduced a risk of bias and confounding in the model.

3.23 The ERG's main concern about costs in the model related to the estimation of surgical-inpatient admissions, because this was a key cost driver in the model. The ERG agreed that the company's modelled estimate of total lifetime surgeries for people having supportive care (33.87 procedures) was reasonable, and that the length of stay associated with a wide excision (5.1 days) was appropriate, but considered that not all procedures would involve wide excisions or inpatient stays. Based on clinical advice, the ERG generated alternative estimates and assumptions, which suggested that the company overestimated the mean cost of surgical-inpatient admissions in the model, for both the supportive care and adalimumab groups. The company applied a cost of £5,488.32 to each inpatient admission. The ERG's alternative assumptions resulted in an estimated mean cost of £1,525.74 per surgical-inpatient procedure, which the ERG used as the unit cost for all surgical-inpatient admissions in the model, based on the following assumptions:

67% of all inpatient surgeries are intermediate procedures done in a day case setting (based on the company's retrospective study using Hospital Episode Statistics data)
6% of surgeries are wide excisions, meaning people have an average of 2 wide excisions over their lifetime
the remaining 27% of surgeries are an equal mix of planned and unplanned intermediate procedures with an average stay of 2 days
a wide excision costs £5,488, a day case intermediate procedure costs £943, and an intermediate procedure needing admission costs £2,103.

The ERG was also concerned that the company had not included costs of other pharmacological therapies taken during the trial.

New evidence submitted by the company after consultation

3.24 The company, in response to consultation, responded to all the committee's requests described in the appraisal consultation document.

3.25 The company submitted revised deterministic and probabilistic cost-effectiveness analyses (table 3). The revised model compared adalimumab with supportive care and incorporated the following changes to the base-case analysis:

clinical estimates were based on the results of a formal meta-analysis of the PIONEER trials instead of the integrated arm-based summary
people stopped treatment if their disease was not responding after 36 weeks, rather than continuing for an additional 12 weeks.

To apply the results of the network meta-analysis to the model, the company used a different approach for the transition probabilities in weeks 12–36. The company considered weeks 12–36 as a single transition instead of dividing weeks 12–36 into 6 cycles of 4 weeks, as it had done in its original submission. The company explained that the patient numbers were too small to estimate reliable transition probabilities for the additional 4‑weekly time points during this period.

Table 3 Company's revised base-case incremental cost-effectiveness analysis results (using adalimumab PAS price)

Scenario	Total cost	Total QALY	Incremental cost	Incremental QALY	ICER
Deterministic analysis (random effects model)
Supportive care	£128,647	11.63	–	–	–
Adalimumab	£140,342	12.58	£11,695	0.95	£12,336
Probabilistic analysis (random effects model)
Supportive care	£129,062	11.64	–	–	–
Adalimumab	£142,407	12.61	£13,345	0.98	£13,676
Abbreviations: ICER, incremental cost-effectiveness ratio; PAS, patient access scheme; QALY, quality-adjusted life year.

3.26 The company also provided the results of 3 scenario analyses (table 4), applied to the revised base case as requested in the appraisal consultation document:

Scenario 1: Partial response was defined as 25% to 50% reduction in the total AN count and no increase in abscesses and draining fistulas.
Scenario 2: Transition probabilities beyond week 36 were based on the PIONEER trials instead of the open-label extension study, and missing data were handled consistently.
Scenario 3: Assumptions for scenarios 1 and 2 were combined.

The definitions of health states in the company's revised base-case analysis were unchanged from its original model (see section 3.10). The definitions of partial response and non-response were amended in line with the committee's preferred assumptions in the scenario analyses only. The company did not include the committee's preferred assumptions about the cost of surgical-inpatient procedures in either the revised base case or the scenario analyses; the assumptions were unchanged from the company's original model (see section 3.23 and section 4.12).

Table 4 Company's scenario analyses, applied to the revised base case (deterministic analysis, random effects model, using adalimumab PAS price)

Scenario	Total cost	Total QALY	Incremental cost	Incremental QALY	ICER
Scenario 1 (new definition of partial responders)
Supportive care	£125,243	11.86	–	–	–
Adalimumab	£130,225	12.51	£4,982	0.65	£7,646
Scenario 2 (transition probabilities beyond week 36 from PIONEER instead of OLE)
Supportive care	£128,647	11.63	–	–	–
Adalimumab	£130,247	12.39	£1,599	0.76	£2,098
Scenario 3 (scenarios 1 and 2 combined)
Supportive care	£125,243	11.86	–	–	–
Adalimumab	£126,373	12.43	£1,131	0.57	£2,002
Abbreviations: ICER, incremental cost-effectiveness ratio; OLE, open-label extension; PAS, patient access scheme; QALY, quality-adjusted life year.

3.27 The company also provided the results of a formal network meta-analysis of the primary and secondary outcomes from the 2 PIONEER trials, for the overall population as well as subgroups. The company used a different method for this meta-analysis, compared with the meta-analysis of outcomes used in the model, because of the difference in the way the HiSCR outcome was reported:

the primary end point of the trials was dichotomous: response or no response
the model divided the HiSCR into 4 categories according to the level of response: high response, response, partial response or no response.

The results of the meta-analysis suggested that the likelihood of an HiSCR response was about 3 times greater for people having adalimumab compared with people having placebo.

Key issues

3.28 The ERG reviewed the new evidence submitted by the company in response to consultation on the appraisal consultation document. The ERG was broadly satisfied with the company's methods for the network meta-analysis of outcomes used in the model, and although the ERG highlighted a few issues with the company's methods, it considered the network meta-analysis was fit for purpose. The ERG was able to replicate the results from the company's revised base-case and scenario analyses.

3.29 The ERG suggested that the results of the company's revised base-case model were unreliable for the following reasons:

The company applied the 4‑week discontinuation rate (1.75%) to the 24‑week fifth cycle.
The company did not use the committee's preferred assumptions about the cost of surgical-inpatient procedures.
The company applied the transition probabilities from the network meta-analysis to weeks 0–36 of the model, but used data from its original arm-based summary for transition probabilities beyond week 36. The ERG noted that this issue applies only to people who have stopped adalimumab treatment, because transition probabilities for people continuing adalimumab were based on the open-label extension study. In reviewing the company's revised base case, the ERG considered that data from the network meta-analysis may not be preferable beyond week 36, because the relevant patient group is not randomised. The ERG also identified another issue, which was that the company's approach to long-term transition probabilities resulted in a better prognosis for people who had adalimumab, but stopped treatment, compared with people who had never had the drug. The ERG did not consider this to be clinically realistic.

3.30 The ERG addressed the issues with the company's revised base case in its exploratory analyses (table 5); all exploratory analyses used the corrected discontinuation rate for the fifth cycle (10.04%). In scenarios 2–5, the company's assumption that adalimumab reduces the number of surgical-inpatient admissions compared with supportive care was maintained, but the mean cost of surgical-inpatient admissions was varied based on the following responses to the appraisal consultation document:

The company suggested that the ERG overestimated the proportion of inpatient surgeries done in a day case setting, because the ERG misinterpreted the Hospital Episode Statistics data. The company suggested that only 49% of inpatient surgeries, rather than 67%, are done in a day-case setting.
A professional group suggested that a patient might have 3–4 wide excisions in their lifetime.

In scenario 6, the ERG assumed that surgical procedures were the same regardless of treatment and set the costs of surgical-inpatient admissions to zero. In scenarios 7–9, the ERG assumed that there was no difference in prognosis beyond week 36 between people who previously had adalimumab and those who had never had the drug, and applied different costs for surgical-inpatient admissions.

Table 5 ERG's exploratory cost-effectiveness analyses, to address issues with the company's revised base case (using adalimumab PAS price)

ERG scenario ^a		ICER for adalimumab compared with supportive care, £/QALY
ERG scenario ^a		Deterministic	Probabilistic
1	Corrected discontinuation rate for cycle 5.	£10,770	NR
2	ERG's original mean surgery cost estimate (£1,525.74); 2 wide excisions, 67% inpatient procedures done in a day setting.	£19,816	NR
3	Increased mean cost of surgery (£1,738.73); 2 wide excisions, 49% inpatient procedures in a day setting.	£19,330	NR
4	Increased mean cost of surgery (£1,838.69); 3 wide excisions, 49% inpatient procedures in a day setting.	£19,101	NR
5	Increased mean cost of surgery (£1,938.65); 4 wide excisions, 49% inpatient procedures in a day setting.	£18,873	£20,196
6	No difference in surgical procedures between people having adalimumab and people having supportive care; costs set to zero.	£23,299	£24,769
7	Same transition probabilities beyond week 36 for people stopping adalimumab and people having supportive care; with ERG's original mean surgery cost estimate (£1,525.74).	£27,701	NR
8	Same transition probabilities beyond week 36 for people stopping adalimumab and those having supportive care, with most favourable ERG estimate for mean surgery cost (£1,938.65).	£26,763	£28,525
9	Same transition probabilities beyond week 36 for people stopping adalimumab and those having supportive care, with least favourable ERG estimate for mean surgery cost (costs set to zero).	£31,167	£33,231
^aAll ERG scenarios were based on the company's random-effects network meta-analysis and included the corrected discontinuation rate for cycle 5. Abbreviations: ERG, evidence review group; ICER, incremental cost-effectiveness ratio; NR, not reported; QALY, quality-adjusted life year.

3.31 The ERG was concerned about the company's scenario analyses. In the first scenario analysis, the company used different transition probabilities taken from separate network meta-analyses based on the new definition of partial response. But, the ERG noted that the company did not change other model parameters that would be affected by the new definition, such as utility values, adalimumab discontinuation caused by adverse events, and costs for the partial response and non-response health states. The ERG highlighted that in the company's second scenario analysis, in which long-term transition probabilities were based on weeks 12–36 of the PIONEER trials instead of the open-label extension study, the company did not use the new transition probabilities estimated from the network meta-analysis done at the request of the committee. Instead, the company applied the transition probabilities for weeks 12–36 from its original model to all patients in the model, which the committee had expressed concerns about in the first appraisal meeting (see section 4.9).