4 Evidence

The diagnostics advisory committee (section 8) considered evidence on enzyme-linked immunosorbent assay (ELISA) tests (Promonitor, IDKmonitor, LISA‑TRACKER, RIDASCREEN, MabTrack, and tests used by Sanquin Diagnostic Services) for therapeutic monitoring of tumour necrosis factor (TNF)‑alpha inhibitors in rheumatoid arthritis from several sources. Full details of all the evidence are in the diagnostics advisory committee papers.

Clinical effectiveness

4.1 The external assessment group (EAG) did a systematic review of the evidence on ELISA tests to monitor the levels of TNF‑alpha inhibitors and antibodies to TNF‑alpha inhibitors (adalimumab, etanercept, infliximab, certolizumab pegol and golimumab) in people with rheumatoid arthritis who:

  • have reached their treatment target (remission or low disease activity)

  • have disease that has not responded to TNF‑alpha inhibitors (primary non-response)

  • have disease that has stopped responding to TNF‑alpha inhibitors (secondary non-response).

4.2 Evidence on the following outcomes was of interest in the clinical-effectiveness review:

  • Decision impact – how the test influences decision making in terms of the proportion of people having treatment modifications such as TNF‑alpha inhibitor dose reduction or switching to another treatment.

  • Clinical utility – the ability of the prospective use of the test (through treatment modification) to affect outcomes for people with rheumatoid arthritis such as duration of time in remission, rate of flares, relapse, or health-related quality of life.

4.3 The EAG included only studies in which at least 70% of people had rheumatoid arthritis, but this inclusion criterion was subsequently relaxed because of the low number of studies retrieved. The EAG found no studies that met the initial criterion, and 2 studies (reported in 4 sources) that met the relaxed inclusion criterion, that is, the 2 studies were done in a mixed population of people with rheumatic diseases, rather than specifically in rheumatoid arthritis. Both studies were done in Spain, in people whose rheumatoid arthritis had reached the treatment target (remission or low disease activity). One was a non-randomised controlled study (INGEBIO; Gorostiza et al. 2016, Arango et al. 2017, Ucar et al. 2017) and the other an observational cohort study (Pascual-Salcedo et al. 2013; Pascual-Salcedo et al. 2015).

4.4 The EAG also considered a study by l'Ami et al. (2017), in people with rheumatoid arthritis. It did not meet the inclusion criteria of the systematic review because it did not specify that people were in remission or had low disease activity at study enrolment. But from the description of the inclusion and exclusion criteria, most people had disease that met this criterion.

4.5 The EAG found 1 ongoing Norwegian multicentre randomised controlled trial (NOR‑DRUM) that is evaluating the effect of therapeutic monitoring of infliximab in people with different inflammatory diseases, including rheumatoid arthritis, compared with standard care. Enrolment for NOR‑DRUM started in March 2017, with an expected primary completion date of March 2020 and study completion date of March 2022.

INGEBIO non-randomised controlled study

4.6 INGEBIO was a prospective, non-randomised, non-inferiority, multicentre pragmatic study. It assessed the efficacy and cost of implementing therapeutic monitoring to guide treatment decisions in people with different rheumatic diseases taking adalimumab. The comparator was standard care in Spain, where treatment decisions (including dose reduction) are based on clinical judgement only.

4.7 It recruited a mixed population of 169 people with rheumatoid arthritis (n=63; 37%), psoriatic arthritis (n=54; 32%) and ankylosing spondylitis (n=52; 31%). They had treatment with adalimumab and had remained clinically stable for at least 6 months (Ucar et al. 2017).

4.8 In the study, everyone had therapeutic monitoring using Promonitor adalimumab and anti-adalimumab antibody kits, but test results were only revealed to clinicians in the intervention arm. They were not obliged to follow any therapeutic algorithm based on the test results but could use it to inform their judgement on treatment. In contrast, therapeutic monitoring test results were not revealed to clinicians in the control arm. This reflected standard care in Spain where treatment decisions are based on clinical judgement only, without knowing the drug levels and antidrug antibodies of people with rheumatoid arthritis.

4.9 The frequency of testing was once every 2 to 3 months. People were assessed for up to 18 months for change in disease response and health-related quality-of-life outcomes.

4.10 Results were reported in 3 conference abstracts. During consultation on the draft guidance, Grifols provided the full study report (commercial in confidence). Table 2A and B show the baseline characteristics and 18‑month clinical outcomes reported in INGEBIO. Ucar et al. (2017) reported intention-to-treat analysis, whereas Arango et al. (2017) reported per-protocol analysis, which excluded 19 people who were lost to follow up. The full study report provided further details on the per-protocol analysis. In the intention-to-treat analysis:

  • A total of 35.8% of people in the intervention group and 36.7% in the control arm (standard care) had their adalimumab doses reduced.

  • The mean duration of remission was 344 days in the intervention group and 329 days in the control group.

  • The rate of flares per patient-year was 0.463 in the intervention group and 0.639 for the control group, with a rate difference of −0.176 (95% confidence interval [CI] −0.379 to 0.0289).

  • There was a non-statistically significant reduction in the risk of flares in the intervention group compared with the control group (incidence rate ratio 0.7252, 95% CI 0.4997 to 1.0578).

  • Median time to the first flare was 145 days in the intervention group and 136.5 days in the control group.

  • Quality of life (EQ‑5D‑5L) was statistically significantly better in the intervention group at visits 2 (p=0.001) and 3 (p=0.035) compared with the control group; EQ‑5D‑5L remained higher in the intervention group throughout the 18‑month follow-up period, although the difference was not statistically significant at other visits.

Table 2A Baseline characteristics and 18-month clinical outcomes reported in INGEBIO – Ucar et al. 2017 (intention-to-treat population)

Outcome

Intervention arm (n=109)

Control arm (n=60)

Baseline characteristics

Proportion of people in remission (%)

73.4

83.3

Proportion of people with low disease activity (%)

26.6

16.7

Median trough adalimumab levels (mg/litre)

5.3

5.5

Clinical outcomes

Mean follow up (days)

499

505

Proportion of people on reduced dose % (number)

35.8 (39/109)

36.7 (22/60)

Rate of flares per patient-year

0.463

0.639

Mean duration of remission (days)

344

329

Mean duration of remission or low disease activity (days)

NR

NR

Median time to first flare (days)

145

136.5

Table 2B Baseline characteristics and 18-month clinical outcomes reported in INGEBIO – Arango et al. 2017 (per-protocol population)

Outcome

Intervention arm (n=98)

Control arm (n=52)

Baseline characteristics

Proportion of people in remission (%)

71.4

82.7

Proportion of people with low disease activity (%)

28.6

17.3

Median trough adalimumab levels (mg/litre)

5.04

5.76

Clinical outcomes

Mean follow up (days)

530.8

544.6

Proportion of people on reduced dose % (number)

35.7 (35/98)

34.6 (18/52)

Rate of flares per patient-year

0.463

0.639

Mean duration of remission (days)

362

360

Mean duration of remission or low disease activity (days)

460

475

Median time to first flare (days)

145

136.5

Notes: The rate of flares per patient-year reported in Ucar et al. (2017) is the same as in Arango et al. (2017) even though these sources reported outcomes for different numbers of people and different follow-up periods. This could be because of an error in 1 of the abstracts.

The difference in duration of follow up between the 2 abstracts is most likely because of the exclusion of 19 people who were lost to follow up (and so had a shorter follow-up time) rather than a longer data collection period.

Abbreviations: NR, not reported.

4.11 Using ROBINS-I criteria, INGEBIO was judged to be at serious risk of bias, because of baseline imbalance in disease activity between the intervention and control groups. The full study report highlighted additional important differences in baseline characteristics between the 2 study groups. Also, the findings may not be generalisable to the UK rheumatoid arthritis population because:

  • the study was done in Spain (dose reduction of TNF‑alpha inhibitors is part of standard care in Spain but not in the UK)

  • it enrolled a mixed population of people with rheumatic diseases.

Observational study by Pascual-Salcedo et al.

4.12 The study was initially identified as a conference abstract (Pascual-Salcedo et al. 2013). However, during consultation Sanquin provided a full-text article (Pascual-Salcedo et al. 2015) that was not indexed in any database and therefore not identified in the systematic literature review. Although results were presented for slightly different populations, the 2 sources were generally in line.

4.13 This was a single-centre observational study of daily clinical practice comparing clinical outcomes. The conference abstract included 88 people (43 with rheumatoid arthritis and 45 with spondyloarthritis) and the full-text publication included 77 people (36 with rheumatoid arthritis and 41 with spondyloarthritis). They had treatment with TNF‑alpha inhibitors (infliximab, adalimumab and etanercept) before and after introducing therapeutic monitoring of TNF‑alpha inhibitors (capture ELISA by Sanquin). Inclusion and exclusion criteria were reported only in the full-text publication.

4.14 The study compared outcomes in the same cohort of people in 2 periods: before introducing therapeutic monitoring, from 2006 to 2009 (full-text publication: 2007 to 2009) and after introducing therapeutic monitoring, from 2010 to 2012. Everyone was in sustained remission or had low disease activity for at least 6 months (defined in rheumatoid arthritis as disease activity score [DAS28] less than 3.2), and had treatment with the same TNF inhibitor throughout the entire study.

4.15 There were statistically significant changes after introducing therapeutic monitoring of TNF‑alpha inhibitors for all 3 drugs (reported in the conference abstract and the full-text publication):

  • the mean drug administration interval was longer (p<0.001)

  • the mean weekly dose was lower (approximately 20% reduction; p<0.001).

4.16 Everyone had stable clinical activity in both periods. In people with rheumatoid arthritis, the mean (± standard deviation [SD]) DAS28 was 2.31±0.52 after introducing therapeutic monitoring of TNF‑alpha inhibitors compared with 2.51±0.85 before (p=0.061). In the full-text publication the mean DAS28 was 2.37±0.50 and 2.28±0.47, respectively (p=0.2).

Additional study by l'Ami et al. (2017)

4.17 The study was an open-label, randomised, parallel-group, non-inferiority trial done in the Netherlands. It assessed clinical outcomes in people with rheumatoid arthritis with high serum adalimumab concentrations (above 8 mg/litre) who had dose-interval prolongation, compared with people who continued standard dosing.

4.18 The trial considered people with rheumatoid arthritis who had treatment for at least 28 weeks and had no indication for adjustment of adalimumab treatment, discontinuation or a scheduled surgery in the next 6 months. A total of 55 people were randomised and 54 included for analyses.

4.19 The mean DAS28 after 28 weeks decreased by 0.14 (SD 0.61) in the interval prolongation group and increased by 0.30 (SD 0.52) in the continuation group. The difference in the mean change in DAS28 was 0.44 (95% CI 0.12 to 0.76; p=0.01) in favour of the prolongation group.

4.20 The authors concluded that the frequency of adalimumab dosing can be safely extended without loss of disease control. However, because of the small sample size and comparable median adalimumab doses at week 28 in both groups, the EAG did not include this study in the economic assessment.

Cost effectiveness

Systematic review of cost-effectiveness evidence

4.21 The EAG identified 5 studies investigating the cost effectiveness of ELISA tests used to measure drug levels and antidrug antibodies for monitoring response to TNF‑alpha inhibitors. There were 3 studies that were model-based economic evaluations (cost-effectiveness models) and 2 were observational (Pascual-Salcedo et al. 2013 reported costs and Arango et al. 2017 reported costs and quality-adjusted life years [QALYs]).

4.22 In INGEBIO (see sections 4.6 to 4.11 for study details), the mean QALYs during the 18‑month follow-up period were 1.076 in the control (standard care) group and 1.145 in the intervention group (therapeutic monitoring of TNF‑alpha inhibitors). This was a gain of 0.069 QALYs compared with control (Arango et al. 2017). The average per patient-year costs of adalimumab were €10,665 in the control group and €9,856 in the intervention group. This was a cost saving of €808 (8% of cost). Other healthcare costs were not reported. Data in the full study report (commercial in confidence) were generally in line with the findings in the conference abstract.

4.23 In the observational study by Pascual-Salcedo et al. (2013; see sections 4.12 to 4.16 for study details), introducing therapeutic monitoring of TNF‑alpha inhibitors resulted in lower monthly acquisition costs of TNF‑alpha inhibitors, compared with the monthly costs of the drugs before monitoring. The monthly cost saving was €92 per person on infliximab (assuming a mean weight of 70 kg), €324 per person on adalimumab, and €257 per person on etanercept. In the full-text publication the monthly cost saving was €143, €215, and €136 per person, respectively (Pascual-Salcedo et al. 2015).

4.24 Krieckaert et al. (2015) considered the cost effectiveness of a personalised treatment algorithm in people with rheumatoid arthritis taking adalimumab in the Netherlands. This was based on clinical response and drug levels (in-house ELISA tests, Sanquin) at 6 months of treatment, compared with standard care. The study population included all people who had treatment for 6 months, regardless of disease response. For 272 people starting adalimumab treatment over the period of 3 years, a test-based treatment strategy would:

  • add 3.84 QALYs

  • save €2.5 million in total healthcare costs (including €2.3 million in drug costs) and

  • save approximately €15,000 in productivity costs.

4.25 Laine et al. (2016) assessed the cost effectiveness of routine monitoring of serum drug concentrations and antidrug antibodies in people with rheumatoid arthritis who had TNF‑alpha inhibitors (adalimumab and infliximab), compared with standard care in Finland. Routine monitoring of both drug and antibody levels was estimated to be cost saving, assuming that it would improve treatment decisions for 2.5% to 5% of people who would otherwise have non-optimal treatment for 3 to 6 months in the standard care scenario.

4.26 Gavan (PhD thesis 2017; personal communication with the EAG) assessed the cost effectiveness of using ELISA testing (no test specified) for monitoring people with rheumatoid arthritis taking adalimumab. The analysis considered 10 different testing scenarios and 2 scenarios in which adalimumab doses were halved without previous testing. Routine adalimumab testing (either drug levels alone or drug levels plus antidrug antibodies) was generally cost effective compared with current practice. But it was unlikely to be cost effective relative to dose reduction (without testing) for people in remission.

Economic analysis

4.27 The EAG developed a de novo economic model designed to estimate the health and economic outcomes of adding therapeutic monitoring of TNF‑alpha inhibitors to usual practice to guide treatment decisions in people with rheumatoid arthritis who had reached treatment target (remission or low disease activity). The model was based on the INGEBIO results, so it considered Promonitor tests for measuring adalimumab drug and antibody levels. Several analyses were done:

  • The first primary analysis, based on the intention-to-treat INGEBIO results reported in a conference abstract by Ucar et al. (2017).

  • The second primary analysis, based on the INGEBIO results reported in a conference abstract by Arango et al. (2017), which excluded 19 people who were lost to follow up.

  • There were 2 additional analyses based on the INGEBIO full study report (commercial in confidence; excluding 19 people who were lost to follow up, as in Arango et al. 2017).

    The 2 primary analyses were considered during the first committee meeting and the 2 additional analyses were considered during the second committee meeting.

4.28 Exploratory analyses were done to assess the health and economic outcomes of the Promonitor tests to measure drug levels and antidrug antibodies for TNF‑alpha inhibitors other than adalimumab (that is, infliximab and etanercept). These assumed similar clinical effectiveness across the TNF‑alpha inhibitors and similar performance of the Promonitor tests used for measuring the drug and antibody levels of all TNF‑alpha inhibitors.

4.29 Economic analyses for ELISA tests other than Promonitor were not done because of the lack of evidence to inform the models.

4.30 Economic assessment for the population with primary or secondary non-response was not possible because of the lack of evidence.

Model structure

4.31 The time horizon was 18 months, as defined by the observational period in INGEBIO. Cost and health outcomes were not extrapolated into the future because of the lack of long-term evidence, so external validation of extrapolated outcomes was not feasible. Therefore, no discounting was applied to estimated costs and QALYs, and mortality was not modelled.

4.32 Because of the short time horizon, a simple model was created. It was assumed that people could be in 1 of 2 health states. However, definitions of health states in each analysis differed because different outcomes were reported in each data source:

  • Ucar et al. (2017) reported time in remission; the model assumed that people would be in the remission health state or the active disease health state (low to high disease activity).

  • Arango et al. (2017) reported time in remission or low disease activity; the model assumed that people would be in the remission or low disease activity health state, or the active disease health state (moderate to high disease activity).

  • The INGEBIO full study report reported both time in remission and time in remission or low disease activity. This meant that 2 separate analyses based on these 2 alternative health state descriptions were done.

    The duration of time in each health state was based on the INGEBIO results.

Key assumptions

4.33 At the beginning of the model time horizon, a proportion of people had their doses reduced in both intervention and control groups, as in INGEBIO. This was based either on clinical assessment only (control group) or clinical assessment and therapeutic monitoring (intervention group).

4.34 The dose of adalimumab was reduced by increasing the interval between doses from 2 to 3 weeks (that is, by spacing doses).

4.35 A proportion of people in both the intervention and control groups had flares, as reported in INGEBIO. In INGEBIO, flare rates in the intervention and control arms were not stratified further according to the dose (full or reduced). Therefore, within each arm, the EAG applied the same flare rate for everyone, regardless of their dose.

4.36 The full dose of adalimumab was restored for everyone on reduced doses when their disease flared (based on the mean time to first flare derived from INGEBIO).

4.37 Everyone who was switched back to the full dose continued on it for the rest of the model time horizon. The disutility of the flare and the cost of managing the flare were applied for the duration of the flare (7 days in the primary analyses; 3 months in the additional analyses).

4.38 The rates of serious adverse events in people on full and tapered doses were 3 per 100 and 2 per 100 patient-years, respectively.

Model inputs

4.39 The model was populated with data from INGEBIO and supplemented with information from secondary sources.

4.40 Costs considered in the economic evaluation included the costs of testing, the costs of treatments taken by people with rheumatoid arthritis, and healthcare costs, from the perspective of the NHS and personal social services.

4.41 The costs of testing comprised those of the test kits, staff time to perform the test and staff training, the cost of the testing service and sample transport. Based on the information submitted by Grifols, the assay cost was £8.75 per test per sample. In the primary analyses, it was assumed that tests for trough drug and antibody levels would be done at the same time (concurrent testing), each sample would be tested once (single testing), and testing would be done once a year. In the additional analyses, it was assumed that reflex testing would be done (see section 3.2), with single testing per sample. The number of tests per year was based on the INGEBIO full study report (commercial in confidence).

4.42 Adalimumab acquisition costs were based on the Humira list price in the British national formulary (BNF; £9,187). However, biosimilar versions of adalimumab are available in the UK. Because the actual prices paid by the NHS are confidential and subject to regional tendering processes, the EAG assumed a hypothetical minimum cost of adalimumab of £1,000 in the threshold analysis. Also, the EAG did one-way deterministic sensitivity analyses to explore the effect of an up to 80% discount on the adalimumab BNF list price on the incremental cost-effectiveness ratio (ICER).

4.43 Treatment wastage was assumed to be £370 per patient-year in people on a full dose; it was reduced proportionally to the reduction in dose.

4.44 Adalimumab is self-administered (usually at home), and, therefore, the administration cost was assumed to be zero.

4.45 The annual per-patient costs of managing health states used in the primary analyses were found to be incorrect after the first committee meeting. Health state costs used in the additional analyses were: remission, £902; remission or low disease activity, £1,089; active disease (low to high activity), £1,483; active disease (medium to high activity), £1,827.

4.46 The costs associated with flare management were £423 per flare for diagnostic investigations and £68 per month for treatment (excluding the cost of biological disease-modifying antirheumatic drugs [DMARDs]).

4.47 The cost of managing an adverse event was £1,622.

4.48 QALYs were estimated from the duration of time in each of the 2 health states, the rates and duration of flares and adverse events, and the corresponding utility values from published literature (see table 3).

Table 3 Model inputs: utilities

Assumption

Estimate

Source

Remission

0.718

Estimated from health assessment questionnaire scores for different health assessment questionnaire bands reported by Radner et al. (2014).

Low disease activity or active disease

0.568*

Estimated from health assessment questionnaire scores for different health assessment questionnaire bands reported by Radner et al. (2014).

Remission or low disease activity

0.665*

Estimated from health assessment questionnaire scores for different health assessment questionnaire bands reported by Radner et al. (2014).

Active disease

0.483

Estimated from health assessment questionnaire scores for different health assessment questionnaire bands reported by Radner et al. (2014).

Disutility of flare**

0.140

Markusse et al. 2015.

Disutility of adverse events***

0.156

NICE technology appraisal guidance 375, Oppong et al. (2013).

* Computed from a weighted average health assessment questionnaire score for the low, moderate and high disease activity health states reported by Radner et al. (2014) and mapped to EQ-5D values following Malottki et al. (2011).

** Rates of flares were based on the INGEBIO study. In the primary analyses, duration of flare was assumed to be 7 days.

*** In the primary analyses, the rates of adverse events in people on full and reduced doses were 3 per 100 and 2 per 100 patient-years, respectively. Duration of adverse events was assumed to be 28 days.

Cost-effectiveness results

4.49 The results of the primary analyses were discussed during the first committee meeting, but because of the error in the health state costs, they were superseded by results from the additional analyses.

4.50 In the additional analyses based on the INGEBIO full study report and corrected health state costs, the ICERs were:

  • £51,929 per QALY gained when the analysis was based on time in remission

  • £125,272 per QALY gained when the analysis was based on time in remission or low disease activity.

Cost-effectiveness results: scenario analyses

4.51 Sensitivity analyses were done to explore the effect of the following parametric and structural uncertainties on the model outcomes from the additional analyses:

  • differences in costs and QALYs are related to differences in rates of flares only (that is, when the effect of health states and adverse events is not considered)

  • the tapering strategy (dose halving: adalimumab 40 mg every 4 weeks rather than every 3 weeks assumed in primary analyses [and compared with every 2 weeks for standard dosing])

  • the total cost of treatment wastage (£0 rather than £370 assumed in the primary analyses)

  • the proportion of people with flares who increase their TNF‑alpha inhibitor dose (55% or 0% instead of 100%)

  • the frequency of testing (once or twice per year rather than the number of tests based on INGEBIO)

  • the cost of testing (including the effect of excluding the cost of the initial phlebotomy appointment, the effect of testing in duplicate, and the effect of concurrent testing or the effect of reflex testing assuming 35.8% have low drug levels).

4.52 In the additional analyses based on the INGEBIO full study report (commercial in confidence), in all except 2 scenarios, the ICERs ranged from £12,035 to £68,693 per QALY gained when analysis was based on time in remission, and from £33,082 to £164,009 per QALY gained when analysis was based on time in remission or low disease activity. When the frequency of testing was assumed to be 1 test per year, or when it was assumed that no additional phlebotomy appointment was needed, testing dominated standard care in both analyses (that is, testing was more clinically effective and cheaper than standard care).

Cost-effectiveness results: deterministic sensitivity analyses

4.53 Sensitivity analyses on the cost of adalimumab (20% to 80% discount) had little effect on the results of the additional analyses. In all analyses based on the INGEBIO full study report, the ICERs ranged from:

  • £55,249 to £65,207 per QALY gained when the analysis was based on time in remission

  • £132,942 to £155,954 per QALY gained when the analysis was based on time in remission or low disease activity.

  • National Institute for Health and Care Excellence (NICE)