A systematic review and economic evaluation of adalimumab and dexamethasone for treating non-infectious intermediate uveitis, posterior uveitis or panuveitis in adults

Patient characteristics

Patients included in the HURON trial48 (mean age 44.8 years) were slightly older than those included in the VISUAL I46 and VISUAL II47 trials (mean age 42.7 and 42.5 years respectively). The proportion of women varied from 57%46 to 63%. 48

Inclusion criteria for patients with active uveitis in the VISUAL I trial46 were based on the manifestation of one or more of the following: VH score of ≥ 2; AC cell grade of ≥ 2 and/or active inflammatory chorioretinal or retinal vascular lesions while on high-dose oral corticosteroids (10–60 mg/day) for ≥ 2 weeks. Inactive uveitis in patients included in the VISUAL II trial47 was characterised by a VH score of ≤ 0.5 and an AC cell grade of ≤ 0.5, with no active inflammatory chorioretinal or retinal vascular lesions (i.e. uveitis inactivity), while receiving 10–35 mg/day of oral prednisone or its equivalent to maintain an inactive state of inflammation ≥ 28 days before study entry. Patients were considered for inclusion if control of their disease was corticosteroid dependent, that is, they had had more than one uveitic flare in the past 18 months occurring within 1 month of tapering steroids. In the HURON trial,48 active intraocular inflammation was based on the presence of a VH score of ≥ 1.5+ and patients unresponsive to previous treatment with corticosteroids were excluded. 48

The mean duration of uveitis was shorter in the active treatment arms than in the comparator arms across all three studies (40.2 vs. 51.0 months for VISUAL I,46 59.5 vs. 62.9 months for VISUAL II46 and 50.5 vs. 61.2 months for HURON48). Intermediate uveitis was the most common site of inflammation in patients (81% of patients) in the HURON trial,48 whereas panuveitis was seen more frequently in patients in the VISUAL trials46,47 (approximately 46% panuveitis vs. 22% intermediate uveitis vs. 33% posterior uveitis). Uveitis-related systemic conditions reported for patients in the VISUAL trials46,47 included Behçet’s disease, sarcoidosis and VKH. More patients with active uveitis had no diagnosed systemic condition (73%) than those with inactive uveitis (56%) in the VISUAL trials. 46,47 Limited information about relevant coexisting systemic conditions was provided for the HURON trial in the journal article,48 company submission43 or clinical study report,71 only that no patients had uncontrolled systemic conditions. Over 90% of patients in the VISUAL trials46,47 presented with bilateral uveitis; outcomes in the left and right eyes were considered separately and were then averaged across eyes in the analysis of the study findings. Conversely, in the HURON trial,48 the proportion of patients with bilateral uveitis was not reported (the AG queried this and was informed by the company that these data were not collected). In patients with bilateral uveitis, the right eye was selected for treatment; only the study eye was analysed for relevant outcomes. Overall, 84% of patients received treatment in the worse-seeing eye.

Study treatment and follow-up

The active treatment in the HURON trial48 was a single DEX intravitreal implant. The study compared the licensed dose of 0.7 mg (DEX 700, n = 77 patients, reported here) and a dose of 0.35 mg (DEX 350, n = 76 patients, not reported here) with a sham procedure (n = 76 patients). One implant was received per patient; no repeat implants were given during the 26-week follow-up period and patients had an implant in only one eye.

The active treatment evaluated in the VISUAL trials46,47 was ADA. Patients randomised to the study arms (n = 11146 and 11547 patients) received a loading dose of 80 mg by subcutaneous injection and then a 40-mg dose, repeated every other week. 46 A corresponding placebo was administered to patients in the comparator arms (n = 11246 and 11447 patients). For patients with active uveitis,46 visits during the study were scheduled at baseline and weeks 1, 4, 6 and 8. Subsequently, further visits occurred every 4 weeks until the primary end point (treatment failure) was achieved or until completion of 80 weeks of treatment. The treatment and follow-up duration was therefore up to 80 weeks (18 months) or until treatment failure. The median duration of treatment and follow-up in the VISUAL I trial46 was 19 weeks for ADA and 13 weeks for placebo. In the VISUAL II trial,47 the median duration of treatment and follow-up was 35 weeks for ADA and 22 weeks for placebo. There was a longer duration of ADA treatment in both studies because patients in the placebo groups met the treatment failure end points earlier than patients in the ADA groups and were taken off treatment.

Previous treatments and concomitant treatments

All patients in the VISUAL trials46,47 had previously received high-dose oral corticosteroids (> 10 mg/day of prednisone or its equivalent) prior to study entry. Within the VISUAL I trial,46 all patients received standardised oral prednisone (60 mg/day; hereafter referred to as a steroid burst) from randomisation, which was gradually tapered to 0 mg by week 15 of the study. Furthermore, topical corticosteroids were permitted but were tapered and discontinued by week 9. In the VISUAL II trial,47 all patients were already receiving oral prednisone (10–35 mg/day); this was tapered to 0 mg by week 19 or earlier depending on the steroid dose at baseline. During the study, patients were eligible to receive at least one immunosuppressant including azathioprine, ciclosporin, mycophenolate mofetil or methotrexate, at the discretion of the study investigator(s).

Limited information on prior and concomitant treatments for uveitis was reported in the HURON trial,48 although one-quarter of patients in the relevant population (DEX 700 and sham groups) for this review had received or were using systemic immunosuppressants or anti-inflammatory treatment at baseline (n = 38/153, 25%). The company did, however, provide patient-level data, which showed that treatment received was generally similar across arms but that more patients received immunosuppressant rescue therapy in the sham arm (10.5%) than in the DEX 700 arm (1.3%).

In the HURON trial,48 patients were permitted to receive different treatments at the discretion of the investigator(s) if they were indicated. Permitted treatments before the study and at baseline as well as during the study included the following:71

• perioperative prophylactic antibiotics (at the visit prior to implantation and 3 days postoperatively)

• IOP-lowering treatments (if IOP was > 30 mmHg in the study eye)

• topical corticosteroids or non-steroidal anti-inflammatory drugs (NSAIDs) in the study eye (if doses remained stable for ≥ 2 weeks before screening, were stable throughout the study visits and were anticipated to remain stable up to week 8)

• intravitreal, topical or periocular corticosteroids in the non-study eye (if inflammation occurred in the non-study eye)

• cycloplegics (indication not specified)

• cataract surgery (if reduced visual acuity had a limiting impact on the patient, the cataract interfered with uveitis management and/or the cataract resulted in local inflammation or glaucoma; the decision to operate was at the discretion of the investigator and patient and a delay to surgery until after week 26 was encouraged)

• systemic immunosuppressants, for example methotrexate or ciclosporin (if doses remained stable for ≥ 3 months before screening, were unchanged throughout study visits and were anticipated to remain stable up to week 8)

• systemic corticosteroids, for example oral prednisone or equivalent (if doses remained stable and were ≤ 20 mg/day at ≥ 1 month before screening, were stable throughout study visits and were anticipated to remain stable up to week 8)

• oral NSAIDs (indication not specified).

Within the HURON trial,48 new treatment or previous management requiring dose escalation with systemic corticosteroids or immunosuppressants or local (intravitreal, periocular and topical) corticosteroids was permitted only if any of these interventions was administered as rescue treatment. In general, rescue anti-inflammatory treatments were permissible if the VH score increased by ≥ 1 unit from week 3 to the start of week 8 and if VH = 1.5+ was recorded from week 8 to week 26. 48 Other rescue medications included anticoagulants and surgical procedures on the study eye. 48,71

Study outcomes

The primary study end point varied across the studies:

• In the VISUAL I46 trial of ADA for active uveitis, the primary end point was time to treatment failure, a composite outcome including worsening of at least one of the following in one or more eyes (from best state achieved following the steroid burst, on or after week 6): AC cell grade, VH grade, BCVA or new active inflammatory retinal or chorioretinal vascular lesions.

• In the VISUAL II47 trial of ADA for inactive uveitis, the primary end point was time to treatment failure, a composite outcome including worsening of at least one of the following in one or more eyes (from baseline, on or after week 2): AC cell grade, VH grade, BCVA or new active inflammatory retinal or chorioretinal vascular lesions.

• In the HURON48 trial of DEX, the primary outcome was the proportion of patients with a VH score of zero at week 8 in the study eye (outcomes were also measured up to week 26).

Outcomes reported in the included studies and grading criteria for intraocular inflammation are presented in Tables 5 and 6 respectively.

Secondary outcomes for the VISUAL I trial46 (see Table 5) were measured from the best state prior to week 6 (following the steroid burst), whereas secondary outcomes for the VISUAL II trial47 were measured from baseline. Secondary outcomes in the VISUAL trials were measured only up to treatment failure or the study end and, as treatment failure occurred in more patients in the placebo arms than in the ADA arms, the results may have been worse in the placebo arms at the point of outcome measurement. The last observation carried forward (LOCF) method was used for dealing with missing data.

Feasibility of meta-analysis

It was not considered appropriate to meta-analyse the findings of the VISUAL I46 and VISUAL II47 trials because the VISUAL I trial46 enrolled patients with active uveitis and the VISUAL II trial47 enrolled patients with inactive uveitis. Active uveitis refers to current inflammation in the eye whereas patients with inactive uveitis have limited inflammation, usually because of treatment with corticosteroids or immunosuppressants. In addition, the magnitude of the treatment effect is likely to be associated with the degree of disease activity and inflammation at baseline, with patients with little inflammation or vision loss at baseline less likely to show an improvement in outcomes. NMA was also not considered feasible or appropriate for the reasons discussed in Indirect comparison of treatments: rationale for not undertaking.

Treatment failure

Best corrected visual acuity

Patient-reported outcome measures

Data on patient-reported outcome measures (PROMs) derived from the publications and submissions related to the VISUAL and HURON studies are reported here. These data are presented in this report before additional clinical outcomes because of their importance for the cost-effectiveness modelling.

Vitreous haze grade

Vitreous haze was measured by dilated indirect ophthalmoscopy in the VISUAL46,47 and HURON48 studies. In both cases grading was based on the original scale proposed by Nussenblatt et al. 37 and later adopted by SUN8 (with the minor modification of ‘trace’ being replaced by 0.5+ in the ordinal scale). An important difference between the VISUAL trials and the HURON trial, however, was that the HURON trial48 proposed an additional 1.5+ grade for cases that were deemed to lie between the 1+ and 2+ grades (see Table 6). Vitreous grade scores are presented in Study outcomes.

Anterior chamber cell grade

Macular oedema

Measures of macular oedema were reported in terms of change in central macular thickness for patients with macular oedema at baseline and time to optical coherence tomography evidence of macular oedema for patients who developed the condition during the studies.

Dexamethasone studies

A summary of effectiveness data from 11 non-randomised, non-comparative studies of the DEX 700 implant is shown in Appendix 5. 18,50,51,79–86 This is based on data from the company submission for DEX;43 the original study publications have not been examined because of time constraints. These data are included here as they provide some data on repeat implants, implants in both eyes and corticosteroid reduction, which were not assessed in the HURON trial. 48 Non-randomised study data for ADA are not included here as they were not provided in the company submission and it was beyond the scope of this assessment to undertake a de novo review of such data.

Following a single implant, two studies reported significant improvements in BCVA at 2–3 months, with a return to baseline values by 6 months,18,82 and significant improvements in VH up to 6 months,18,82 with a return to baseline by 12 months in the study with a longer follow-up. 18 Significant improvements in central retinal thickness were reported in one study up to 6 months after a single implant82 and in another study up to 3 months, with a return to baseline by 6 months. 83

Studies in which patients received between one and four implants reported improvements in BCVA at 12 months,50,83,86 with the improvements reported to be significant in one study. 50 Among studies with patients having a mix of single or multiple implants and macular oedema, significant improvements in central retinal thickness were reported up to 12 months in one study,50 with another study reporting significant improvements at 3 months but not at 6 months. 85

In terms of repeat implants, one study reported that after the second implant BCVA improved significantly by 1 month but then decreased, with a similar trend following up to six implants (not significant but small patient numbers). 18 Central retinal thickness and VH also showed a significant temporary improvement after the second implant, with similar (non-significant) improvements after the third and fourth implants. 18 Another study reported that the improvements in BCVA and central retinal thickness at 1 month were similar (not significantly different) following the first and second implants. 84

In a study of uveitis patients the median time from first to second implant was 10 months50 whereas in four studies of uveitic macular oedema the mean/median time from first to second implant was 4.7,81 5.0,51 7.183 and 1086 months. The mean time from second to third implant was 3.4 months in one study of uveitic macular oedema. 81

Implants in both eyes were assessed in one study, with 3 out of 11 patients (27%) receiving implants in both eyes having a response (reduced central retinal thickness and improved BCVA) in the second eye. 18

In terms of reduction in other therapies following a single implant, one study reported that 21 out of 27 patients (78%) reduced or stopped systemic or local treatment,18 whereas in another study 3 out of 12 patients (25%) reduced their corticosteroid dose79 and in another study systemic corticosteroids were reduced or discontinued in 14 out of 32 patients (44%) and discontinued in 8 out of 32 patients (25%) at 6 months. 82 Among studies using a mix of single or multiple implants, in one study systemic corticosteroids or immunosuppressants were reduced in 62% of patients and steroids were discontinued in 36% of patients at 12 months50 and in another study systemic corticosteroids were reduced or discontinued in 78% of patients and discontinued in 32% of patients at 12 months. 51

Adalimumab studies

Non-RCT data were presented in the company submission72 and these were based on a retrospective audit presented in the Clinical Commissioning Policy for anti-TNF treatment options for adult patients with severe refractory uveitis. 87 The study evaluated data for patients aged > 18 years with different clinical forms of uveitis receiving ADA (40 mg every other week) or infliximab (3–5 mg/kg every other week). The main findings of the audit were as follows:

• Clinical remission of uveitis was observed in all patients (n = 41) on biologics [mean ± standard deviation (SD) follow-up of 1.36 ± 0.88 person-years].

• In total, 48.78% of patients experienced visual acuity improvement (mean ± SD follow-up of 2.51 ± 2.01 person-years).

• Fewer patients (17.07%) had worsening of visual acuity (mean ± SD follow-up of 4.38 ± 3.50 person-years).

• Patients receiving biologics, in due course, required fewer or reduced concomitant treatments:

  • 88.89% of patients showed a reduction in steroid dose to ≤ 10 mg (mean ± SD follow-up of 3.06 ± 2.32 person-years)

  • 75.85% of patients showed a reduction in steroid dose to ≤ 5 mg (mean ± SD follow-up of 3.15 ± 1.76 person-years)

  • 45.16% of patients discontinued steroid treatment (mean ± SD follow-up of 3.49 ± 1.59 person-years)

  • 83.33% of patients showed a reduction in the number and/or frequency of immunomodulatory therapy (mean ± SD follow-up of 1.54 ± 0.99 person-years).

• Patient-reported outcomes reported in the audit87 are summarised as follows:

  • A significant decrease in vision-related quality of life [Vision Core Measure (VCM) scale88] was directly associated with a decrease in visual acuity in the worse eye within 1 year of starting biologics (p = 0.0064).

  • Median vision-related VCM scores decreased with increasing follow-up time from the time of starting treatment with biologics.

  • Mean SF-36 physical component summary scores (< 47) were lower than estimates for the general population. However, the SF-36 mental component summary scores (> 47) were higher than estimates for the general population, with the exception of scores obtained at year 3 (duration of audit period not reported).

Safety information from Summaries of Product Characteristics

The SmPC for the DEX implant45 states that the most commonly reported AEs are those frequently observed with ophthalmic steroid treatment or intravitreal injections, including elevated IOP, cataract and conjunctival or vitreal haemorrhage. Less frequently reported but more serious adverse reactions include endophthalmitis (severe eye infection), necrotising retinitis (viral infection of the retina), retinal detachment and retinal tear. 45

The SmPC for ADA44 summarises AEs from studies of 9506 patients across a range of conditions. The SmPC states that the most commonly reported AEs are infections (such as nasopharyngitis, upper respiratory tract infection and sinusitis), injection site reactions (erythema, itching, haemorrhage, pain or swelling), headache and musculoskeletal pain. TNF antagonists such as ADA affect the immune system and their use may affect the body’s defence against infection and cancer. Fatal and life-threatening infections (including sepsis, opportunistic infections and tuberculosis), hepatitis B virus reactivation and various malignancies (including leukaemia, lymphoma and hepatosplenic T-cell lymphoma) have also been reported with use of ADA. Serious haematological, neurological and autoimmune reactions have also been reported, including rare reports of pancytopenia, aplastic anaemia, central and peripheral demyelinating events and lupus, lupus-related conditions and Stevens–Johnson syndrome. 44

Safety data from pivotal randomised controlled trials

Safety data from the RCTs are based on the published journal articles for the HURON trial48 and VISUAL I46 and II47 trials, the company submissions43,55 and the clinical study reports. 52,71,72 In the case of the HURON trial the safety data are based on all patients who were randomised to a group and received treatment [n = 76/77 (99%) for the DEX 700 group and n = 75/76 (99%) for the sham group]. Within the 26-week trial, the mean exposure to the intervention was 25.9 weeks in both groups. For the two RCTs of ADA compared with placebo, safety data are based on all randomised patients in both trials [n = 111 (100%, ADA) and 112 (100%, placebo) in VISUAL I and n = 115 (100%, ADA) and 114 (100%, placebo) in VISUAL II]. It should be noted that, in these trials, exposure to ADA was longer than exposure to placebo because treatment failure (and cessation of study treatment) occurred earlier in the placebo groups (median exposure: VISUAL I – 19 weeks ADA vs. 13 weeks placebo; VISUAL II – 35 weeks ADA vs. 22 weeks placebo). Therefore, one might expect more events in the ADA groups than in the placebo groups.

A summary of AEs is provided in Table 19. An AE of any type occurred in 80% of patients in the DEX 700 group compared with 68% in the sham group in the HURON trial48 and in 85–91% of patients in the ADA group compared with 79–84% of patients in the placebo group in the two VISUAL trials. 46,47 Serious AEs (SAEs) occurred in 9% of patients in the DEX 700 group compared with 6.7% of the sham group in the HURON trial48 and in 6–14% of patients in the ADA group compared with 5–8% of patients in the placebo group in the VISUAL trials. 46,70 There were no deaths in the HURON trial48 and one death in the ADA arm in each of the VISUAL trials;46,70 neither death was considered to be treatment related.

Systemic adverse events

Serious systemic AEs are shown in Table 20. Table 21 lists other systemic AEs that (1) occurred in ≥ 5% of patients in any treatment group in the HURON trial,48 (2) occurred in ≥ 5% of patients in the ADA groups in the VISUAL trials46,89 and (3) were noted as potentially important within uveitis treatments by clinical advisors to the AG. No reported systemic AEs (serious or non-serious) had a substantially higher rate in the DEX 700 arm than in the sham arm. The rate of serious infections was higher in the ADA group than the placebo group in the VISUAL I trial46 (4.5% vs. 1.8%) but not the VISUAL II trial47 (1.7% vs. 1.8%). Malignancies and chronic renal failure each occurred in a total of three patients across the ADA arms of both VISUAL trials but in no patients in the placebo arms. The majority of the listed systemic AEs had a somewhat higher rate in the ADA arms than the placebo arms.

Immunogenicity

In the VISUAL I trial,46 anti-ADA antibodies were detected in 3 out of 110 patients (2.7%) in the ADA group. These three patients had treatment failure at 16, 44 and 48 weeks (compared with a median time to treatment failure of 24 weeks among the remaining 107 patients). In the VISUAL II trial,47 anti-ADA antibodies were detected in 6 out of 115 patients (5.2%) in the ADA group. Five of these six patients had treatment failure at weeks 13, 16, 16, 24 and 31 (not estimable for the remaining patient).

Ocular adverse events

Ocular AEs are shown in Table 22. In terms of serious ocular AEs, endophthalmitis (severe eye infection) and severe uveitis worsening occurred in one patient each in the DEX 700 group and in no patients in the sham group in the HURON trial. 48 Conjunctival haemorrhage occurred in 30% of patients in the DEX 700 group and 21% of patients in the sham group in the HURON trial,48 whereas rates for this AE were low in the VISUAL trials. 46,47

Raised IOP occurred in 25% of the DEX 700 group compared with 7% of the sham group in the HURON trial,48 whereas there was little difference between the ADA group and the placebo group in the VISUAL trials. 46,47 In the DEX 700 group, IOP of ≥ 25 mmHg peaked at week 3 (7.1% vs. 1.4% in the sham group), whereas IOP of ≥ 35 mmHg peaked at week 12 (4.1% vs. 0% in the sham group). By week 26, no patients in the DEX 700 group had an IOP of ≥ 25 mmHg whereas 4.2% of patients in the sham group had an IOP of ≥ 25 mmHg.

Glaucoma rates showed little difference between the DEX 700 group (0%) and the sham group (2.7%) in the HURON trial48 or between the ADA group (0.9%) and the placebo group (0%) in the VISUAL I trial. 46 In the HURON trial48 no patients required incisional surgery for glaucoma, whereas two patients (2.6%) in the DEX 700 group required laser iridotomy in the study eye for iris bombe and raised IOP. At any single time point across the 26 weeks, up to 23% of patients in the DEX 700 group required IOP-lowering medication (the percentage requiring this at any point in the study is not reported).

In the HURON trial,48 9 out of 62 eyes (15%) in the DEX 700 group and 4 out of 55 eyes (7%) in the sham group had cataracts in eyes that were phakic (had a natural lens) at baseline. Among phakic eyes with no cataract at baseline, 9 out of 42 eyes (21%) in the DEX 700 group and 4 out of 28 eyes (14%) in the sham group had cataracts. For the ADA group, no data were reported on whether eyes were phakic or had cataract at baseline; among all patients there were more cataracts in the ADA group than in the placebo group in the VISUAL I trial46 (3.6% vs. 1.8%) but more in the placebo group than in the ADA group in the VISUAL II trial47 (1.7% vs. 5.3%). Cataract surgery among phakic eyes occurred in 1 out of 62 patients (1.6%) in the DEX 700 group and 2 out of 55 patients (3.6%) in the sham group; in the VISUAL II trial47 cataract surgery occurred in one patient in the ADA group and two patients in the placebo group.

Safety data from non-randomised studies of dexamethasone

A summary of safety data from 11 non-randomised, non-comparative studies of the DEX implant is shown in Appendix 6. 18,50,51,79–86 This is based on data presented within the company submission for DEX. 43 The proportion of patients with an increased IOP is typically higher in real-world studies than in a RCT, which may reflect the inclusion of patients with a prior need for IOP-lowering medications, who were excluded from the HURON trial. 43 Implant migration to the AC has been reported in a few patients and occurred in eyes that were aphakic (no lens) or pseudophakic (artificial lens). 43 A few cases of endophthalmitis or retinal detachment were reported after administration of DEX 700. 43 Non-randomised studies of ADA are not included here as they were not included in the company submission and it was beyond the scope of this assessment to undertake a de novo review of these data.

Ongoing studies of dexamethasone

Two ongoing RCTs of DEX 700 were identified, both in patients with macular oedema as a result of uveitis. Both compared DEX 700 against other local treatments. The POINT trial [PeriOcular and INTravitreal Corticosteroids for Uveitic Macular Edema Trial (NCT02374060), due to complete in 2018] compares DEX 700 with intravitreal triamcinolone or periocular triamcinolone, whereas the MERIT trial [Macular Edema Ranibizumab v. Intravitreal Anti-inflammatory Therapy Trial (NCT02623426), due to complete in 2019] compares DEX 700 with intravitreal methotrexate or intravitreal ranibizumab. In addition, a long-term safety cohort study of DEX 700 (NCT01539577) in 875 patients with posterior segment-involving uveitis or central or branch retinal vein occlusion was due to complete in March 2016, but no published results were identified.

Ongoing studies of adalimumab

Three ongoing RCTs of ADA were identified. One small RCT [Adalimumab in Uveitis Refractory to Conventional Therapy (ADUR) (NCT00348153)]90 compared ADA plus corticosteroids and immunosuppressants with corticosteroids in combination with immunosuppressants and was due to be completed in March 2013. This study is potentially of interest because of its active comparator arm. However, no published results were identified other than an abstract reporting intermediate results for 20 of 25 patients;90 this was not included in the clinical effectiveness review because of the limited results presented. The two other RCTs of ADA are due to complete in 2019. The RUBI trial [Randomized Trial Comparing Efficacy of Adalimumab, Anakinra and Tocilizumab in Non-infectious Refractory Uveitis (NCT02929251)] aims to compare ADA against two further biological therapies: anakinra (an interleukin-1 receptor antagonist) and tocilizumab (an antibody against the interleukin-6 receptor). The IVAS trial [Intravitreal Adalimumab Versus Subcutaneous Adalimumab in Non-infectious Uveitis (NCT02706704)] aims to compare subcutaneous ADA against intravitreal ADA.

In addition, a non-randomised extension study of ADA [A Study of the Long-term Safety and Efficacy of Adalimumab in Subjects with Intermediate-, Posterior-, or Panuveitis (VISUAL III) (M11-327, NCT01148225)] enrolled patients from the VISUAL I and VISUAL II studies (ADA or placebo arms) who either completed these trials or experienced treatment failure. Patients who discontinued the VISUAL I or II study because of treatment failure were defined as having active disease on entry to the VISUAL III study, whereas patients who completed the VISUAL I or II study were defined as having inactive disease. Patients received open-label ADA (40 mg every other week) and were followed up for 78 weeks (active uveitis patients) or 54 weeks (inactive uveitis patients). The study is due to be completed in 2018. Preliminary data are available from a conference abstract. 91 This states that, of 243 patients with active uveitis after 78 weeks, 96.3% had no new inflammatory lesions relative to week 8, 91.0% had an AC cell grade of ≤ 0.5+ and 87.8% had a VH grade of ≤ 0.5+. Of 128 patients with inactive uveitis after 54 weeks, 98.5% had no new inflammatory lesions relative to baseline, 98.5% had an AC cell grade of ≤ 0.5+ and 92.6% had a VH grade of ≤ 0.5+. The mean systemic corticosteroid daily dose decreased from 12.7 to 3.68 prednisone equivalents by year 1 for patients with active uveitis and remained stable from 1.48 to 1.21 prednisone equivalents for inactive patients. AE rates were stated to be comparable to those in the VISUAL I and VISUAL II trials but no data were presented in terms of numbers of patients with events. No data were presented for visual acuity or the VFQ-25.

Padula et al.:100 a cost-effectiveness analysis of off-label biologics to treat sarcoid posterior uveitis versus standard of care: comparing infliximab to methotrexate and systemic steroids

The study by Padula et al. 100 was reported only as a conference abstract. The authors presented the methods and results of a cost-effectiveness analysis of infliximab compared with methotrexate and systemic steroids over a lifetime horizon. The economic evaluation used a semi-Markov approach to estimate health outcomes and costs. Patients entered the model following the onset of sarcoid posterior uveitis. No further information was provided about the population reflected in the model. Cost-effectiveness was evaluated in terms of the incremental cost per quality-adjusted life-year (QALY) gained from a societal perspective.

Probabilities, health utilities and costs used in the model were reported to be taken from the literature, although parameter values were not reported in the abstract. It was not specified whether or not a systematic review was conducted. Costs and health outcomes were discounted at a rate of 3% per annum. Costs were expressed in 2010 US dollars. The authors conducted univariate sensitivity analyses, threshold analyses and probabilistic sensitivity analysis (PSA) using 10,000 simulations.

The incremental cost-effectiveness ratio (ICER) for methotrexate compared with systemic steroids was estimated to be US$10,053 per QALY gained. Methotrexate dominated infliximab in the base case. However, if a patient’s health utility after successful recovery was < 0.750 (base-case value of 0.84), then infliximab produced a greater net benefit than methotrexate, assuming a willingness-to-pay (WTP) threshold of US$50,000 per QALY gained. The PSA suggested that the probability of methotrexate dominating infliximab was 0.60.

It is not possible to assess the validity of the model as only limited information is provided within the conference abstract. The AG notes that this analysis does not include either of the interventions being assessed within this appraisal (DEX and ADA) and the model does not appear to differentiate between unilateral and bilateral uveitis, which may be associated with different cost-effectiveness results. There is insufficient information provided within the abstract for this analysis to be useful in the current appraisal.

Sugar et al.:101 cost-effectiveness of fluocinolone acetonide implant versus systemic therapy for noninfectious intermediate, posterior, and panuveitis

Sugar et al. 101 presented a cost-effectiveness analysis of fluocinolone acetonide intraocular implant compared with oral corticosteroids and immunosuppressive agents. Costs and health benefits were estimated from data collected during the MUST trial. 57 The economic analysis used a time horizon of 3 years and costs and benefits were discounted at a rate of 3% per annum. Costs were expressed in US dollars (year unclear). The authors used a payer’s perspective for costs and the patient’s perspective for outcomes. The authors estimated the cost to a payer to maximise health benefits by using the more effective, but more expensive, treatment.

The within-trial data (differences in costs and utilities), reported at 2 years’ follow-up, were extrapolated by a further year for a 3-year time horizon. The difference in the mean total costs of the treatments was determined with a linear regression with a saturated means model. The history of the disease was modelled through a sequence of utility values measured during the trial at different points in time. No health states were used. Uncertainty was assessed using bootstrapping and was represented using cost-effectiveness planes and cost-effectiveness acceptability curves (CEACs).

For bilateral uveitis, the fluocinolone acetonide implant for both eyes was estimated to generate 0.057 additional QALYs at an additional cost of US$16,900; the ICER was reported to be US$297,800 per QALY gained. The probabilities of the fluocinolone acetonide implant being cost-effective compared with systemic therapy at WTP thresholds of US$50,000 and US$100,000 per QALY gained were 0.003 and 0.04 respectively. For unilateral uveitis, the implant resulted in 0.130 additional QALYs at an additional cost of US$5300; the ICER was reported to be US$41,200 per QALY. The probabilities of the implant being cost-effective compared with systemic therapy at WTP thresholds of US$50,000 and US$100,000 per QALY gained were 0.53 and 0.74 respectively.

The study highlights the importance of considering unilateral and bilateral uveitis separately within future economic evaluations in terms of (1) the cost difference between types of treatments, (2) quality-of-life impacts and (3) the greater risk to vision of an operative procedure on both eyes compared with one eye. However, this study does not consider the cost-effectiveness of the implant in one eye for patients with bilateral uveitis as all patients with bilateral uveitis within the MUST trial were given an implant in both eyes. 57 The model has several additional key limitations:

• All relevant comparators were not included in the model. Systemic steroids and immunosuppressants were assumed to be the gold standard as they were the only included comparators of the fluocinolone acetonide implant, with no discussion about whether or not this was appropriate.

• AEs were not taken into consideration.

• It is not clear how the 2 years of data from the MUST trial were extrapolated to the 3-year time horizon.

• It is not clear whether or not the implant would have benefits after this 3-year period.

• No model validation was reported.

• The analysis of uncertainty was not well described.

Patient population

The model population consists of people with non-infectious intermediate uveitis, posterior uveitis or panuveitis. Patients receiving DEX were assumed to have active disease, whereas the model assessed the cost-effectiveness of ADA separately for patients with active and inactive disease. An analysis was undertaken to explore the cost-effectiveness of DEX use in one eye in patients with unilateral disease as a separate subgroup; the trial did not provide data separately for this group and hence this analysis is considered to be exploratory. Because of a lack of evidence, it was not possible to explore additional subgroups. A cohort of uveitis patients was assumed to enter the model with a mean age of 44.8 years, based on the mean age within the HURON trial,48 and was followed over a lifetime. The model population was limited to adults aged ≥ 18 years because the marketing authorisations for the technologies being considered relate only to this group.

Interventions

The two technologies considered were ADA (40 mg every 2 weeks until treatment failure) and the DEX implant (0.7 mg, once only in the base case).

Within the clinical trials of ADA (VISUAL I46 and II47), patients were already receiving high-dose corticosteroids at randomisation, with a corticosteroid burst given to all patients at the start of the VISUAL I trial; corticosteroids were tapered to zero by week 15 (VISUAL I) or week 19 (VISUAL II). Clinical advisors to the AG suggested that this is also likely to reflect clinical practice, although the SmPC suggests that ADA may be given alongside corticosteroids or alone. 44 Given the evidence available, for patients with active disease, the model considers the cost-effectiveness of ADA plus an initial oral corticosteroid burst, rather than ADA alone.

The DEX implant can be administered in the affected eye to unilateral patients, in one eye for patients with bilateral disease or in both eyes at staggered intervals for patients with bilateral disease. Patients could also receive more than one consecutive implant. Clinical advisors to the AG suggested that the DEX implant would most probably be used when disease affects only one eye (or is more severe in one eye in the case of asymmetric disease), or to treat a temporary flare-up in one or both eyes, when systemic disease is not present or is well controlled. The base-case model assumed that patients would receive one DEX implant in one affected eye, as in the HURON trial. 48 There are no RCTs that assess the use of more than one consecutive implant or the use of implants in two affected eyes. However, there are several non-randomised trials, with 12–24 months of follow-up, that allow the use of repeat implants. 18,50,51 These studies consistently report that, after around 6 months, patients’ outcomes return to those at baseline and that up to three repeat implants are each likely to have a similar treatment effect. Given the limited evidence around repeat implants, this was explored within sensitivity analysis. In one of the studies that assessed implants in both eyes, 3 out of 11 patients (27%) receiving implants had a response (reduced central retinal thickness and improved BCVA) in the second eye. 18 Clinical advisors to the AG suggested that it is more likely that systemic treatment would be used if both eyes required treatment; however, the direction of the ICER for treatment in both eyes compared with one eye is considered in Chapter 6.

Comparators

The two technologies were compared independently with current practice, which includes a range of immunosuppressants (such as methotrexate, mycophenolate mofetil, ciclosporin and azathioprine) and corticosteroids. Given the concerns regarding the robustness of undertaking a NMA (see Chapter 3, Indirect comparison of treatments: rationale for not undertaking), within the base-case analysis, current practice was assumed to be equivalent to practice in the control arm (sham or placebo) of the clinical trials of the interventions. In the VISUAL trials of ADA,46,47 patients received initial corticosteroids, which were tapered by 15 weeks (VISUAL I46) and 19 weeks (VISUAL II47), and 32% in the VISUAL I trial46 and 48% in the VISUAL II47 study were receiving one immunosuppressant at baseline (across arms), which they were able to maintain according to the study protocol. Given that a greater proportion of patients in practice are likely to receive systemic corticosteroids, these comparators are denoted throughout as limited current practice based on the VISUAL I trial [LCP(VI)] or the VISUAL II trial [LCP(VII)]. In the HURON trial of DEX,48 patients were allowed to receive rescue therapy with corticosteroids or immunosuppressants and 25% were using systemic immunosuppressants or anti-inflammatory treatment at baseline, which they were able to maintain according to the study protocol. This comparator is denoted throughout as limited current practice based on the HURON trial [LCP(H)]. Apart from rescue therapy with immunosuppressants within the HURON trial, the proportions receiving immunosuppressants and corticosteroids were similar across the arms of the HURON48 and VISUAL I46 and II47 trials. In current practice, a greater proportion of patients would receive systemic immunosuppressants or anti-inflammatory treatment than in the control arms of the pivotal studies; consequently, the base-case analysis is likely to underestimate both the effectiveness and the AE profile of current practice, as well as the costs associated with treatment. Within exploratory analyses, the AG assessed the impact on the results of increasing the value of these parameters within the model. However, it should also be noted that in clinical practice a greater proportion of patients being treated with ADA and DEX are also likely to receive concomitant treatment.

Outcomes

The model was used to estimate the incremental cost per QALY gained for each intervention compared with current practice.

The VISUAL trials and the HURON trial48 each reported VFQ-25 HRQoL data at baseline and at each follow-up visit. The VISUAL trials46,47 also reported EQ-5D data at baseline and at each follow-up visit. The model used the EQ-5D data directly for modelling the effectiveness of ADA. The HURON trial48 reported EQ-5D data at baseline but not at subsequent time points. It was therefore not possible to use the EQ-5D data directly; however, Allergan shared patient-level data from the HURON trial with the AG, which allowed an analysis of the relationship between the VFQ-25 and the EQ-5D using the baseline data (see Model structure). It was necessary to convert VFQ-25 data to EQ-5D utilities to estimate QALYs associated with treatment with DEX and LCP(H). 104

The use of the outcomes from the HURON trial48 representing vision and inflammation (visual acuity, VH) was considered by the AG as an alternative to the use of the VFQ-25 for estimating QALYs; however, the VFQ-25 outcome was preferred because of the difficulties associated with using vision as an outcome in uveitis and capturing all impacts of the interventions (see Chapter 1, Description of the health problem). Clinical advisors to the AG suggested that clinicians measure ocular outcomes based on multiple factors, including visual acuity, VH and macular oedema. The VFQ-25 captures multiple components of vision, as well as broader considerations such as general health and the vision-related impact on the ability to drive and undertake normal activities. It is also essential to capture the AEs associated with the treatments and it is difficult to determine the utility decrements associated with the multiple interacting AEs associated with these treatments. The AG considered that the VFQ-25 should largely capture the impact of AEs, as well as treatment effects, on HRQoL.

The presence of unilateral or bilateral uveitis is important in terms of estimating outcomes for several reasons. The BCVA in the better-seeing eye is more representative of quality of life than the BCVA in the worst-seeing eye. 105 In addition, a patient with bilateral disease is expected to have a lower quality of life on average than a patient with unilateral disease. Thus, a person with bilateral disease has more scope to benefit from treatment. However, in patients with bilateral disease receiving local treatment, the choice of study eye is important in determining the extent to which quality of life can increase.

In the VISUAL I46 and VISUAL II47 trials, 91% and 96% of patients had bilateral disease respectively. Clinical advice received by the AG suggests that this is representative of patients who would be given ADA in practice because it is a systemic treatment. Within the HURON trial48 it was not recorded whether patients had unilateral or bilateral disease. Based on the patient-level data provided by Allergan, the proportion of patients with VH that was greater than zero in the non-study eye was 51%; clinical advisors to the AG stated that this suggests that ≥ 51% of patients had bilateral disease. Within the HURON trial,48 when patients had bilateral uveitis, the right eye was chosen for treatment. This resulted in the better-seeing eye being treated in 10.7% and 17.1% of cases for DEX 700 and the sham procedure respectively.

Given that the presence of unilateral or bilateral uveitis was not reported in the HURON trial,48 it was not possible for the AG to undertake robust subgroup analysis around this factor. The base-case model is therefore dependent on the assumption that the patients included within the HURON trial and the way in which DEX was used within the trial would be representative of current practice. It was not possible to draw robust conclusions about the subgroups separately in terms of cost-effectiveness; however, an exploratory subgroup analysis was undertaken (see Model evaluation methods). As described within Chapter 1 (see Description of the health problem), it is expected that around 70–80% of this patient population would have bilateral disease. However, it may be that, because DEX is a local treatment, patients with unilateral disease would be more likely to be selected for DEX treatment, both within the trial and in practice. Given that patients with bilateral disease have a greater capacity to benefit from treatment because of the BCVA of the better-seeing eye being the best predictor of quality of life, and that treatment of one eye would cost the same whether given to a person with unilateral or bilateral disease, if the trial had a lower proportion of bilateral cases than in practice, the effectiveness of DEX may be underestimated. Conversely, if the trial had a higher proportion of bilateral cases than in practice, the effectiveness of DEX may be overestimated.

Time horizon

The time horizon of the model was the lifetime of patients (up to age 100 years) and a starting age of 44.8 years was used, representing the average age of patients with non-infectious posterior segment-involving uveitis within the HURON trial. 48 A cycle length of 2 weeks was chosen as this is the time between administration of ADA doses and assessment of patients for disease progression. This is also a sufficiently short cycle length to capture all relevant clinical events associated with DEX and current practice.

Discounting

All costs and QALYs were discounted at a rate of 3.5% per year.

Treatment discontinuation

In the base-case analysis, the DEX implant was assumed to be administered only once to one eye and to have an efficacy of 30 weeks, based on data from the HURON trial. 48

Patients could discontinue ADA as a consequence of any of the four criteria for treatment failure used within the VISUAL trials,46,47 including (1) the development of new inflammatory lesions, (2) worsening of AC cell grade, (3) worsening of VH grade or (4) worsening of visual acuity. Treatment discontinuation was modelled using parametric curves fitted to Kaplan–Meier curves for time to treatment failure from the trials. The Kaplan–Meier curves for time to treatment failure included in the VISUAL I52 and II72 clinical study reports were digitised and individual patient data were reconstructed using the methods described by Guyot et al. 106 A number of parametric curves were fitted to the data using the flexsurv R package (R version 3.3.2, flexsurv version 1.0.0; The R Foundation for Statistical Computing, Vienna, Austria). Tables 30 and 31 present the Akaike information criterion (AIC) and Bayesian information criterion (BIC) scores for statistical goodness of fit.

It should be noted that these are relative measures of goodness of fit and it is possible that other models not tested here could provide a better fit to the data. Figures 9–12 show the Kaplan–Meier data and the fitted parametric distributions for the VISUAL I46 and II47 trials for the ADA and comparator groups.

FIGURE 9.

Observed and fitted curves for time to treatment discontinuation in the ADA arm in the VISUAL I trial.

FIGURE 10.

Observed and fitted curves for time to treatment discontinuation in the placebo arm in the VISUAL I trial.

FIGURE 11.

Observed and fitted curves for time to treatment discontinuation in the ADA arm in the VISUAL II trial.

FIGURE 12.

Observed and fitted curves for time to treatment discontinuation in the placebo arm in the VISUAL II trial.

The statistical analysis suggested that, of those tested, the parametric distributions with the best fit to the data were the Gompertz and the log-normal distributions for both the ADA group and the placebo group in the VISUAL I46 and II47 trials. Clinical advisors to the AG suggested that it is clinically plausible that some patients would remain on ADA for years; hence, the plateauing of these curves seems potentially reasonable. However, the Gompertz curve seems clinically implausible as observational studies of ADA in similar patient populations have suggested that patients are likely to continue to fail treatment in the longer term. 107,108 The log-normal distribution appears to be the most plausible for the placebo arm so that patients do fail on treatment relatively quickly. The log-normal distribution also appears clinically reasonable for the ADA group. It should be noted that, although based on these predictions alone some patients would continue to receive treatment for an implausibly long period of time, within the model patients may die of other causes, which negates the need for a cure model to be employed.

It was assumed that after patients fail and discontinue treatment with ADA, or 6 months after the DEX implant is injected, patients receive limited current practice, which includes a range of immunosuppressants (such as methotrexate, mycophenolate mofetil, ciclosporin and azathioprine) and corticosteroids for a proportion of patients. It was also assumed that ADA and DEX are effective only while they are being given. Therefore, patients who are no longer being treated with ADA, and patients who received the DEX implant > 6 months ago, accrue no additional health gains.

Permanent legal blindness

The VISUAL46,47 and HURON48 trials did not report any occurrence of blindness, which is likely to be because of the short duration of these trials. However, it may be that the use of ADA or DEX could prevent damage to the eye, which may in turn prevent future blindness. Conversely, it is possible that the AEs associated with treatment (such as raised IOP) could lead to an increased risk of blindness via glaucoma. There is no evidence about any longer-term (positive or negative) impacts of the interventions on vision loss beyond the treatment duration. As such, it was not possible to include a complex model of long-term outcomes associated with the interventions. However, as these interventions ultimately aim to reduce permanent damage to the eye, a state for becoming permanently legally blind was included as this has the largest impact on quality of life and costs.

We defined blindness as a BCVA of ≤ 20/200 in the better-seeing eye, according to the UK definition of legal blindness. 109 The AG considered two approaches for modelling permanent blindness based on the evidence from the key RCTs. The first was to extrapolate the decrease in BCVA over time using the mean change and distribution from the trials and estimate the proportion of patients who would go below the legal blindness threshold in each group. The AG considered that this approach had three weaknesses: (1) the follow-up period of the clinical trials was not long enough to capture the total impact on visual acuity because damage to the eye does not always immediately impact on visual acuity; (2) there are different trajectories according to the cause of the damage to the eye, which could not be appropriately captured by a single parametric distribution; and (3) for patients with unilateral disease, additional assumptions about the probability of blindness in both eyes would need to be made.

The second approach considered by the AG was to use outcomes from the trials such as glaucoma and macular oedema as surrogate outcomes for blindness in the future. In principle, this would allow a more accurate estimate of blindness over time to be made, and could exclude outcomes such as cataract, for which blindness is reversible through surgery. However, the AG did not identify any evidence that could provide a link between these shorter-term outcomes and blindness. The only evidence of blindness caused by uveitis identified by the AG was cross-sectional and did not specify time to blindness. 18,27 This means that populating the model would have required elicitation or an assumed distribution for how long it would take patients to become blind and for this to be extended beyond the period of the cross-sectional study data. In addition, the key long-term outcome to include in the model is blindness in both eyes, given that the BCVA in the better eye is the best predictor of quality of life and blindness in both eyes would incur the greatest costs. The cross-sectional studies do not provide sufficient information to estimate the probability of blindness in both eyes; hence, numerous assumptions would be required. The identified studies also include patients with anterior uveitis. The AG requested the patient-level data from the authors of one of the cross-sectional studies18 to be able to predict blindness over time from the outcomes reported within the clinical trials; however, these data were not provided. Given the number of assumptions that would be required to undertake this analysis in the absence of patient-level data, and the low proportions of patients reported to have glaucoma (< 3% in any arm) and new cases of macular oedema (< 8% in any arm) in the clinical trials,46–48 the AG decided that adopting such a complex analysis within the model may produce potentially misleading results.

Therefore, a simpler approach was taken and the assumptions were tested within exploratory analyses. For the base-case analysis, clinical experts to the AG helped to identify sources that could be used to estimate a constant blindness rate associated with (limited) current practice. All studies identified were cross-sectional rather than longitudinal. The best source of evidence was considered to be a study by Dick et al. ,19 a retrospective analysis of insurance claims data, because all patients (n = 1769) had posterior segment, non-infectious uveitis. A constant rate of blindness and uncertainty around this parameter was estimated by the AG based on the proportion of patients going blind within the study by Dick et al. 19 and the mean follow-up time. By 10 years, this study predicted that 6.6% of patients would go legally blind, in the absence of death from other causes. The proportions of patients who had unilateral and bilateral disease are not reported within the study. Two alternative similar sources were also identified as being potentially relevant, studies by Tomkins-Netzer et al. 18 and Durrani et al. 11 The constant blindness rate derived from the study by Tomkins-Netzer et al. 18 was deemed to be an underestimate by one of the clinicians consulted by the AG (Alastair Denniston, personal communication) and the study included a wider population than the target population of the current appraisal (including patients with infectious and anterior uveitis). The rate derived from the study by Durrani et al. 11 was substantially higher than the rate derived from Dick et al. 19 but this study also included a wider population. As the authors warned, ‘being a tertiary referral centre, more patients are likely to suffer from severe, often bilateral uveitis’ and they acknowledged that the results of their study ‘could not be applied to the general population because of the tertiary nature of the patient population’. 11 The AG explored the impact of using blindness rates based on these other two sources in exploratory analyses (see Exploratory sensitivity analyses and Results).

As discussed earlier, there is no evidence on the treatment effect of ADA or DEX on legal blindness. To model the impact of treatment with ADA on the rate of blindness, given the strict criteria for treatment failure within the VISUAL trials,46,47 it was assumed that patients could not go blind before treatment failure. This was assumed for both the intervention and the comparator. The rate of blindness following treatment failure was then approximated so that the rate of blindness at each cycle in the placebo group was equivalent to the estimate from Dick et al. 19 It was not considered clinically reasonable that a DEX implant would prevent all cases of blindness during treatment, but it was deemed equally unreasonable to assume that it would prevent no cases of blindness. In light of the absence of evidence around this parameter, the AG sampled from a uniform distribution between 0 and 1 within the PSA and used the mean of this distribution (0.5) for the deterministic analysis. Therefore, the AG assumed that half of the cases of blindness in this group would be avoided for the period in which the treatment effect was applied (30 weeks in the base case). It was assumed that patients in the comparator group would have the same blindness rate as in the general population.

The AG heard from clinicians that around 20–30% of patients with uveitis remain unilateral and that patients treated with DEX are more likely to be unilateral. The AG assumed that patients who remained unilateral would not go blind and therefore the rate of blindness in the DEX target population would be lower than that in the general population and this is turn would be lower than that in the ADA target population. For the base case, the AG assumed that, in the general population, 25% of patients would remain unilateral whereas in the DEX target population 30% of patients would remain unilateral. For ADA, the proportions of patients with unilateral uveitis as reported in the VISUAL I46 (9.2%) and VISUAL II47 (4.4%) trials were used for active and inactive patients respectively. The blindness rate for bilateral patients was determined by dividing the blindness rate estimated from Dick et al. 19 by the proportion of bilateral patients in the general population. The incidence of blindness in each analysis was adjusted by multiplying the rate of blindness for bilateral patients by the proportion of bilateral patients in each population.

Adverse events

One of the key drivers for new treatment options is the substantial AE profile of existing treatments, which reduce HRQoL and incur treatment costs. In addition, treatment with ADA and DEX is associated with AEs. Given that the main outcome measures being used from the clinical trials were the VFQ-25 and EQ-5D, it was assumed that these will capture the quality-of-life impacts associated with AEs during the period in which the treatments are provided. The incidence of AEs from the trials was therefore used only to calculate the additional costs associated with their management. As such, AEs included within the model are limited to those for which the cost of treatment is substantial. Based on advice from the clinical experts to the AG, the AEs associated with substantial costs of treatment are cataract, raised IOP, glaucoma, serious infections, hypertension, fractures and diabetes.

There are no longer-term safety data for ADA and DEX for uveitis beyond the data reported within the key clinical trials. 46–48 However, Burmester et al. 110 reported the results of a study on the long-term safety of ADA in 23,458 patients with other indications compared with the general population in terms of the incidence of disease. The study found that overall malignancy and lymphoma rates were not increased as a result of ADA use. Although the incidence of non-melanoma skin cancer was greater for patients receiving ADA for some of the indications included, for all indications death rates were equivalent to or lower than those expected in the general population. Given the findings of this study, the model does not include any longer-term cumulative impacts of ADA on patient outcomes. For DEX, the model assumed that patients will receive only one implant. The application of more than one implant is considered later in the discussion section.

There is no clinical rationale for the AEs associated with corticosteroid use to differ between study arms because usage is similar between the arms within the trials. Therefore, although diabetes and osteoporosis are associated with substantial costs, there are no real differences in incidence between the arms of the trials. Within the exploratory analysis assessing the impact of greater corticosteroid use in the comparator groups, the proportion of patients with these AEs was increased according to their incidence in the MUST trial. 57

The probabilities of AEs per cycle used in the model (Table 32) were calculated based on their incidence in the trials and the mean follow-up time of each trial.

Quality of life

Estimating the relationship between the Visual Function Questionnaire and the EuroQol-5 Dimensions

The AG considered the published studies for mapping the VFQ-25 to the EQ-5D included in the database of mapping studies by Dakin. 111 However, none of the published mapping studies was based on a population with uveitis and, considering that the AG had access to the VFQ-25 and EQ-5D patient-level data at baseline from the HURON trial,48 the AG decided to fit a new mapping model. The AG used the approach that produced the best fit according to Browne et al. 112 (ordinary least squares) and noticed that the mapping resulted in similar coefficient values to those presented by Payakachat et al. ,113 who used an alternative modelling method (censored least absolute deviation). The mapping was used for all of the analyses involving DEX, within the exploratory analyses comparing the interventions with current practice, as provided in the MUST trial,57 and within a sensitivity analysis for ADA.

The patient-level data from the HURON trial48 were used to test for a correlation between the VFQ-25 and the EQ-5D at baseline. The scatterplot is presented in Figure 13. A linear regression model was fitted to the data to predict EQ-5D utilities from VFQ-25 scores. One regression model was fitted to all three arms of the HURON trial (sham, DEX 350 implant and DEX 700 implant) to maximise the sample size for the regression analysis. The underlying assumption was that the relationship between VFQ-25 scores and EQ-5D utilities would be independent of treatment. The fitted regression used in the economic model was:

⁽¹⁾ EQ-5D utility = 0.4454059 + VFQ-25 score × 0.0051322 .

FIGURE 13.

The relationship between the VFQ-25 and the EQ-5D based on patient-level data from the HURON trial. 48

It is recognised that a linear model is not bounded and is likely to have poor performance for utility values at the extremes. However, given that the mapping was only used for means, no extreme values were used. Alternative non-linear models (e.g. quadratic regression) were also tested but did not significantly improve the fit to the data. The variance–covariance matrix of the slope and the intercept of the regression model is presented in Table 33. To represent the uncertainty in the regression model, the matrix was used to sample the two coefficients of the regression model in the PSA.

TABLE 33 - Variance–covariance matrix of the intercept and the covariate of the regression model

	Intercept	VFQ-25
Intercept	1.75E-03
VFQ-25	–2.42E-05	3.63E-07

The baseline utilities, that is, the utilities for patients at week 0, were estimated based on the patient-level data from each trial: the HURON trial48 for DEX and its comparator [LCP(H)], the VISUAL I trial46 for ADA and its comparator in active patients [LCP(VI)] and the VISUAL II trial for ADA and its comparator in inactive patients [LCP(VII)]. In the HURON trial,48 the baseline utility and visual acuity were substantially different between the sham arm and the DEX arm (visual acuity was 71.3 for the sham arm and 63.7 for the DEX 700 arm). Clinical advisors to the AG were asked to consider whether or not the baseline difference in both utility and visual acuity were reasonably due to random variation. All three experts agreed that a difference in visual acuity of ≥ 10 letters is considered to be clinically significant and a difference below this level could be owing to random variation and therefore it is plausible that the difference between the arms at baseline in the HURON trial was due to random variation. The baseline utilities were not varied to represent any population subgroups because these data were not available from the trials. The impact of changing the baseline utility has been assessed within the univariate sensitivity analysis; however, this analysis assumes that the relative treatment effect remains the same. This is unlikely to be the case for subgroups with differing baseline utilities such as patients with unilateral or bilateral uveitis. However, there is no evidence from the trials around outcomes for these subgroups that would enable a robust subgroup analysis.

Estimating utility over time

The VFQ-25 data from each follow-up point within the HURON trial48 (weeks 0, 8, 16 and 26) and the EQ-5D data from each follow-up point of the VISUAL trials46,47 (weeks 0, 1, 4, 6, 8, 12, 16, 20, 24, 27 and 32 and then every 4 weeks until week 80) were used to estimate the change in utility for each treatment group over the time period of the trials. These were adjusted according to the average baseline utilities but maintaining the change from baseline in each arm.

When comparing DEX with its comparator, the AG assumed that the utility of patients who received DEX would drop to that of patients in the comparator arm after the duration of the treatment effect. Within the base-case analysis, the treatment effect was assumed to last for 30 weeks (4 weeks longer than the trial period). Within the sensitivity analyses the utility was assumed to decrease to the baseline utility over varying time periods. When comparing ADA with its comparator, for patients who fail and hence discontinue treatment it was assumed that utility returns to the baseline utility score, adjusted for any reduction in utility associated with age. For patients who receive ADA beyond the duration of the trial (80 weeks), it was assumed that their utility remains constant after the last follow-up point until treatment discontinuation. This utility is based on the mean of the last 6 months of data. Figures 14–16 present the predicted mean utility values over time, excluding any adjustments for blindness, for DEX compared with LCP(H) for active patients, ADA compared with LCP(VI) for active patients and ADA compared with LCP(VII) for inactive patients respectively.

FIGURE 14.

Mean utilities for DEX compared with LCP(H) for active patients over time.

FIGURE 15.

Mean utilities for ADA compared with LCP(VI) for active patients over time.

FIGURE 16.

Mean utilities for ADA compared with LCP(VII) for inactive patients over time.

Age adjustments to utility were based on the regression equation reported by Ara and Brazier. 114 Age-related utility was calculated using the following formula:

⁽²⁾ Utility = A × ( Male ) + B × ( Age ) + C × ( Age × Age ) + D ,

where A = 0.0212126, B = –0.0002587, C = –0.0000332 and D = 0.9508566.

The ratio between the utility for the general population at the starting age and that of the mean cohort age at each cycle was applied within the model.

Resource use and costs

Mortality

Mortality rates within the model were assumed to reflect those of the general population, based on the most recent Office for National Statistics life tables for England. 129

The model assumed that AEs have no impact on mortality, although it is recognised that in practice diabetes, osteoporosis and blindness would have some impact on mortality.

Probabilistic sensitivity analysis

To assess the uncertainty around the parameters used in the model, the AG defined probability distributions for most parameters using the available evidence and undertook PSA. Gamma distributions were generally used for costs and beta distributions for utility values and probabilities. The RR of blindness for DEX was based on a uniform distribution because of a lack of evidence. Table 37 summarises the input parameters and their base-case mean values and distributions used in the PSA. In addition to the parameters listed in Table 37, beta distributions were defined for utility scores at each time point in each arm, as well as the prevalence of concomitant therapy and the incidence of AEs and rescue therapy. Multivariate normal distributions were used for the parameters of the survival curves used to determine time to treatment failure. A Dirichlet distribution was used for the weight distribution of the cohort, which determined the mean dose cost of azathioprine and ciclosporin.

Exploratory sensitivity analyses

A number of exploratory analyses were undertaken to explore the uncertainties within the model. Although there was a lack of evidence to fully inform these exploratory analyses, the aim was to provide an indication of the impact of alternative assumptions on the results.

1. A greater proportion of patients are treated with immunosuppressants and corticosteroids in the comparator groups. In clinical practice it would be expected that a higher proportion of patients would receive systemic therapy. This would result in greater efficacy associated with the comparator, with a higher AE rate and higher costs.

   As discussed in Chapter 3 (see Indirect comparison of treatments: rationale for not undertaking), it was not possible to undertake a NMA to compare DEX or ADA with an alternative comparator, which might be more representative of current practice. However, the comparator arm of the MUST trial57 (identified within the systematic review) was made up of patients who received systemic corticosteroids, supplemented in 86% of cases with immunosuppressants, and was thought by the clinical experts to the AG to be reasonably representative of clinical practice. Hence, this study was used to inform an exploratory analysis. This exploratory analysis was not undertaken for patients with inactive uveitis because the MUST trial included only patients with active uveitis. For active patients, data from the comparator arm of the MUST trial were used relating to (1) an estimate of the total proportion of patients receiving (i) corticosteroids and (ii) immunosuppressants, to estimate costs; (2) an estimate of the HRQoL of patients; and (3) the rates of any AEs associated with substantial resource use. With respect to the total proportion of patients receiving corticosteroids and immunosuppressants, it was unclear from the MUST trial publication exactly which immunosuppressants were used57 and hence the composition was assumed to be the same as that for the VISUAL I trial. 46 It should be noted that using the data from the MUST trial without performing any formal mixed treatment comparison assumes that the trial population was comparable with the populations within the VISUAL and HURON trials and does not include any measure of uncertainty around the comparison.

   Within the base-case analysis, HRQoL was assumed to return to baseline levels following treatment failure with ADA or after 6 months following DEX implantation. Given that the comparator arm patients were able to receive immunosuppressants and corticosteroids, it was assumed within this exploratory analysis that patients treated with ADA or DEX were also able to receive immunosuppressants and corticosteroids, as in the comparator arm of the MUST trial following the end of treatment with the intervention. Therefore, the overall effectiveness of DEX and ADA was expected to increase as well as the effectiveness of the comparator.

   The analysis assumed that treatment with prednisolone includes concomitant therapy with Adcal D3 (£47.58) and 20 mg of omeprazole once daily (£15.25).

2. Incidence and HRQoL impact of blindness. As there is limited evidence around the rate of legal blindness for this patient group, and there is no evidence around the impact of treatment on this rate, the AG performed exploratory analyses around these parameters. This was done by varying the rate of legal blindness in patients with uveitis who are treated with (limited) current practice (from 0 to 0.0374) based on alternative sources11,18 (see Permanent legal blindness) and varying the RR of legal blindness cases avoided as a result of treatments (from 0 to 1).

   These analyses were also undertaken using (1) alternative utilities from Brown115 and (2) a higher cost of blindness based on the upper bound of the 95% CI for this parameter.

3. Patients who go into remission after ADA treatment. A proportion of patients who continue treatment with ADA may achieve remission. The base-case analysis assumed that these patients would continue to receive ADA until treatment failure; however, the clinical advisors to the AG suggested that, after around 2 years of stable disease, patients may no longer require treatment but because they are in remission they may maintain the same level of HRQoL as that while on treatment. This sensitivity analysis therefore assessed the impact of assuming that, after 2 years on treatment, a varying proportion of patients (0–1) would no longer receive ADA but their HRQoL would decrease only with age, until the treatment failure curve predicts failure or they die from other causes.

4. Using the VFQ-25 data from the VISUAL trials of ADA46,47 to map to EQ-5D utility data. This sensitivity analysis assessed the impact of using the regression analysis of the HURON trial data48 to map the VFQ-25 data from the VISUAL trials46,47 to EQ-5D utilities.

5. Extrapolation of time to treatment discontinuation for ADA. The impact of using alternative plausible parametric distributions (Weibull, Gompertz) for time to treatment discontinuation was explored.

6. Varying the time period over which the utility decreases to that of baseline after treatment. The treatment effect beyond 6 months for DEX and beyond treatment discontinuation for ADA is unknown. Within the base case, patients receiving DEX were assumed to take 4 weeks to return to baseline utility beyond the trial follow-up of 6 months. HRQoL for patients receiving ADA was assumed to return to baseline levels immediately on treatment discontinuation. Within this exploratory analysis the treatment effect beyond 6 months for DEX was varied from 0 to 8 weeks and the treatment effect beyond discontinuation for ADA was increased to 4 weeks.

Univariate sensitivity analyses

Each parameter within the base case was varied to assess its impact on the model results, as shown in Table 38.

Base case

The base-case results are presented in Table 39. Based on the probabilistic version of the model, a single DEX implant combined with limited current practice as provided in the HURON trial48 [DEX 700 + LCP(H)] was estimated to produce 0.029 incremental QALYs compared with LCP(H) alone at an additional cost of £573, resulting in an ICER of £19,509 per QALY gained. Figure 17 presents the CEAC. Assuming WTP thresholds of £20,000 and £30,000 per QALY gained, the probability that a single DEX implant produces more net benefit than LCP(H) is estimated to be 0.47 and 0.72 respectively. The deterministic results were similar to those generated using the probabilistic model (Table 40), with an estimated ICER of £20,058 per QALY gained for DEX 700 + LCP(H) compared with LCP(H). A breakdown of the results of the deterministic analysis is provided in Appendix 7.

FIGURE 17.

Cost-effectiveness acceptability curve for DEX 700 + LCP(H) vs. LCP(H).

The small differences in both costs and QALYs between the two groups mean that the ICER is very sensitive to alternative model parameters and assumptions, as shown within subsequent sensitivity analyses.

Figure 18 shows the cost-effectiveness scatterplot for DEX 700 + LCP(H) compared with LCP(H). The scatterplot shows that there is a negative correlation between incremental costs and incremental QALYs. The AG believes that this is because blindness has a strong impact both on QALYs gained and on costs, and the impact of DEX on blindness is very uncertain. A low RR of blindness would lead to increased QALY gains and important cost savings.

FIGURE 18.

Cost-effectiveness plane scatterplot of DEX 700 + LCP(H) vs. LCP(H).

Exploratory analyses

Univariate sensitivity analyses

The AG explored the impact of different parameters on the results of the model, as shown in Table 46.

The model results were generally robust to changes in the values of these parameters. The model was most sensitive to assumptions around the comparator, assumptions around permanent blindness and the duration of the treatment effect.

Base case

The base-case results are presented in Table 47. In the base case, ADA + LCP(VI), as provided in the VISUAL I trial,46 was estimated to produce 0.194 incremental QALYs compared with LCP(VI) alone in patients with active uveitis at an additional cost of £18,321, resulting in an ICER of £94,523 per QALY gained. The ICER generated using the deterministic version of the model (£95,506) (Table 48) was similar to that in the probabilistic model. A breakdown of the results of the deterministic analysis is provided in Appendix 7. Figure 19 shows the CEAC for ADA + LCP(VI) compared with LCP(VI) in patients with active uveitis. It should be noted that within the VISUAL I trial46 both treatment groups included an initial systemic steroid burst, which was tapered by week 15, and around 30% of patients in both arms received systemic immunosuppressants.

FIGURE 19.

Cost-effectiveness acceptability curve for ADA + LCP(VI) vs. LCP(VI) in patients with active uveitis.

Figure 20 shows the cost-effectiveness plane for ADA + LCP(VI) compared with LCP(VI). The scatterplot shows that there is a positive correlation between the incremental costs and the incremental QALYs. The AG believes that this is because longer ADA treatments lead to more QALYs but also incur important additional costs.

FIGURE 20.

Cost-effectiveness plane scatterplot of ADA + LCP(VI) vs. LCP(VI) in patients with active uveitis.

Exploratory analyses