Threshold analyses

In the threshold analyses, the cost of TNF-α testing at which adding TDM to SOC would result in zero net monetary benefit (NMB) was estimated, as described below.

The NMB represents the value of an intervention in monetary terms when a willingness-to-pay (WTP) threshold for a unit of benefit (e.g. QALY) is known. It is estimated by first assuming a WTP threshold (e.g. £20,000 or £30,000 per QALY gained) and then calculating the NMB as follows:

where incremental costs and incremental benefits represent incremental costs and QALYs for the health technologies under consideration.

In this study, NMB was estimated for a range of acquisition costs of ADL (from £1000 to £9187 per patient-year) at the WTP thresholds of £20,000 and £30,000 per QALY gained, which are the thresholds usually considered by NICE. In the threshold analyses, the costs of drug acquisition and administration and the costs associated with disease management were included; the latter comprised the costs of managing flares and AEs, and the costs of managing different health states. QALYs were estimated from the rates of flares and AEs, and the average duration of remission and LDA/active disease health states (for the analysis based on data from Ucar et al.⁴²) or remission/LDA and active disease health states (for the analysis based on data from Arango et al.⁴³) in patients from the intervention and control arms.

In the threshold analyses, the cost of TNF-α testing per patient-year, under which the test-based treatment strategy has zero NMB, was estimated in the following way:

where the total cost of testing comprises the costs associated with testing patient blood samples to monitor trough drug and antibody levels. The ICER threshold represents the NICE WTP of £20,000 or £30,000 per QALY gained, and Δcosts and ΔQALYs are incremental costs and QALYs across the intervention and control arms.

The costs incurred in each arm were estimated as follows:

For scenario 1 (with ‘remission’ and ‘LDA/active disease’ health states), QALYs were derived in the following way:

For scenario 2, QALYs were estimated from the duration of ‘remission/LDA’ and ‘active disease’ health states and their corresponding utilities.

Cost-effectiveness analyses

In addition to the threshold analyses, cost–utility analyses were conducted in which ICERs were estimated using the list prices of biologics and the cost of TNF-α testing; the latter was based on the prices of the Promonitor test kits (provided by Grifols–Progenika), and the other costs associated with TNF-α testing³⁷ and clinical advice.

Duration of flare

Substantial heterogeneity in the duration of flare has been reported⁷⁰ and observed in clinical practice. A flare may last from 2–3 days up to 2–3 months, depending on severity (Dr Jani, personal communication). The duration of flare was estimated in the dynamic cohort in the Brigham Rheumatoid Arthritis Sequential Study (BRASS),⁷⁰ which included 1105 people with established RA who had received usual care at the Brigham and Women’s Hospital in Boston⁷⁰ (Table 26).

TABLE 26

Flare duration in the BRASS study

The estimate of 7 days was adopted in the primary analyses based on Ucar et al.⁴² and Arango et al.;⁴³ this was consistent with the estimate used in NICE TA375.²³ The impact on the results of a longer duration of flare, 19 days, was evaluated in scenario analyses; this represents a weighted average of the estimates reported in the BRASS study⁷⁰ (see Table 26) and those provided by Dr Jani (personal communication).

Time to the first flare

Arango et al.⁴³ and Ucar et al.⁴² reported the median time to the first flare that was observed in the intervention and control arms of the INGEBIO study; however, according to the NICE Guide to the Methods of Technology Appraisal,⁶⁹ mean estimates should be utilised in economic analyses of health interventions. The mean time to the first flare in the intervention and control arms was calculated from Kaplan–Meier curves for the time to the first flare in the INGEBIO study, sourced from a poster presentation by Ucar and colleagues at the Annual European Congress of Rheumatology EULAR 2017 (Ucar and Osakidetza, personal communication, September 2018), by using the area under the curve approach.

The Kaplan–Meier estimates were available for 300 days (see Figure 8, Appendix 10) and were extrapolated for the duration of follow-up reported in Ucar et al.⁴² and Arango et al.⁴³ (see Table 23). Given that the proportion of participants who were on a tapered dose in the intervention and control arms levelled at around 240 days after dose tapering, it was assumed that these proportions remained the same until the end of the observational periods in the INGEBIO study and, therefore, no parametric model fitting was performed. The estimated mean time to the first flare was 208.07 days and 189.32 days in the intervention and control arms, respectively. These values were used in both scenario 1 (based on Ucar et al.⁴²) and scenario 2 (based on Arango et al.⁴³).

Flare rate

Treatment arm-specific flare rates (per patient-year) were reported in both Ucar et al.⁴² and Arango et al.⁴³ (see Table 23), and were the same in both sources despite the fact that the abstracts reported outcomes for different follow-up periods. These estimates were utilised in the primary and exploratory analyses.

Rate of serious adverse events

One study from the clinical effectiveness systematic review, Senabre Gallego et al.,⁷³ reported the rate of AEs experienced by patients who were treated with TNF-α inhibitor therapies. This study recruited 39 participants with RA who had achieved remission. The findings showed that one participant (3%) had septic arthritis (serious infectious arthritis) that was associated with TNF-α inhibitor therapies (ADL or ETN) in the 1-year follow-up period (see Table 55, Appendix 11).

Given that the evidence on SAEs in the population of interest was limited, additional searches were conducted. Lahiri and Dixon⁷⁴ indicated that there was a time-dependent increase in the risk of serious infections in people with RA who were treated with biologics, with the maximum risk in the first 6 months of biological therapy and a gradual decline thereafter. The authors argued that this time-dependent decrease in the risk of serious infections can be attributed both to ‘depletion of susceptibles’ (i.e. high-risk participants dropping out of the TNF-α inhibitor cohort because of death, stopping therapy or loss to follow-up), which accounted for two-thirds of the observed difference, and to reduction in the inherent infection risk resulting from an improvement in patient’s functional status and a decrease in the dose of glucocorticoid.

According to Bruce et al.,⁷⁵ the risk of Pneumocystis jirovecii pneumonia in people from the BSRBR-RA register who were treated with TNF-α inhibitors was low, with an incidence rate of 2 (95% CI 1.2 to 3.3) events per 10,000 person-years of follow-up (see Table 55, Appendix 11); the rate of tuberculosis was higher among those treated with ADL (144 events per 100,000 person-years) and IFX (136 events per 100 000 person-years) than among those treated with ETN (39 events per 100,000 person-years) (see Table 55 and Dixon et al.⁷⁶).

The rate of SAEs reported in Burmester et al.⁷⁷ was 4.7 per 100 patient-years (see Table 55, Appendix 11). This estimate was derived from 15,132 people with RA who were exposed to ADL in 28 global clinical trials. A SAE was defined as a fatal or immediately life-threatening event; an event necessitating hospitalisation or prolonging hospitalisation; an event resulting in persistent or significant disability/incapacity or congenital anomaly; or an event necessitating medical or surgical intervention to prevent a serious outcome. At baseline, participants considered in this study had a mean age of 53.5 years and a mean disease duration of 9.1 years; 78.8% were female, 16.5% had a treatment duration > 2 years and 10.9% had a treatment duration > 5 years.

The rate of serious infection that was adopted in TA375,²³ 35 out of 1000 patients, was based on Singh et al.⁷⁸ and was assumed to be independent of the bDMARDs used (i.e. all biological therapies were assumed to have similar safety profiles).

Consultation with clinical advisors confirmed that serious infections in people with RA from the population of interest are relatively rare.

In this study, the modelled AE rate for people who were receiving a full dose of biologics, three events per 100 patient-years, was adopted from Senabre Gallego et al.⁷³ The AE rate in tapered patients was estimated using an odds ratio (OR) for serious infections in people who were treated with low-dose biologics compared with people who were receiving the standard dose⁷⁹ (see Appendix 12). The resulting AE rate in tapered patients was two events per 100 patient-years.

Duration of serious adverse events

In TA375,²³ a serious infection in RA patients was assumed to persist for 28 days, on average. This estimate was adopted in all analyses reported here.

Parameters specific to the threshold analyses

Given that the patent for the ADL originator product (Humira) expired in October 2018 and the true costs of the ADL biosimilars to the NHS were not known to the EAG at the time of writing, in the threshold analyses the annual acquisition cost was varied from £1000 to £9187 per patient-year. The latter represents the annual cost of ADL (Humira), assuming a dose of 40 mg every 2 weeks delivered by subcutaneous injection using a prefilled pen and the NHS indicative price from the BNF²¹ (Table 28).

TABLE 28

Acquisition costs of the biologics

Parameters specific to the cost-effectiveness analyses

In the cost–utility analyses, ICERs were estimated using the list prices of the TNF-α inhibitors (in accordance with NICE guidelines⁶⁹) and the costs of testing based on the costs of Promonitor assays (provided by Grifols–Progenika), other testing costs outlined in the study conducted by Jani et al.³⁷ and clinical advice (see Table 28 and Cost of testing for more details).

The primary analyses were conducted for the list price of the ADL originator product, Humira^®, whereas, in the exploratory analyses for other TNF-α inhibitors described in Exploratory analyses: etanercept or infliximab and Promonitor, the list prices for the ETN originator product (Enbrel^®) and its biosimilar Erelzi^® and IFX biosimilars Flixabi^® or Renflexis^® were utilised.

Conversion to Great British pounds

Where conversion from other currencies to Great British pounds (GBP) was required, International Monetary Fund purchasing power parity (PPP) was used to convert within the year (e.g. from 2001 euro to 2001 GBP), after which inflation was applied. The Campbell and Cochrane Economic Methods Group–EPPI-Centre (Evidence for Policy and Practice Information and Coordinating Centre) Cost Converter was used for the PPP conversion.⁸⁶

Inflation to 2017–18 prices

Unit costs were inflated to 2017–18 prices by inflating to 2015–16 prices using the Hospital and Community Health Services (HCHS) pay and prices index,⁸¹ and then to 2017–18 prices using the average increase in the index for the previous 3 years (from 2013–14 to 2015–16), with the average rate of 1.1% per annum (see Appendix 14).

Treatment costs

Drug acquisition

Annual acquisition costs of the TNF-α inhibitors from the NICE scope⁴⁸ estimated using their list prices and assuming adherence to standard dosing regimen for each drug are shown in Table 28.

The estimated costs of treatment with ADL, ETN, GLM and CTZ were based on the price of the solution for injection in prefilled pens given that these biologics are administered subcutaneously and can be self-administered. Consultation with clinical experts confirmed that all of the TNF-α inhibitors considered in this study, except IFX, are usually self-administered by people with RA at home.

Consistent with acquisition cost calculations in TA375,²³ the cost per annum of IFX was estimated using a patient average weight of 70 kg. IFX is administered intravenously (the cost of intravenous administration is described in Drug administration).

As reported in TA375,²³ the manufacturers of GLM provided the 100-mg dose at the same price as the 50-mg dose under a patient access scheme arrangement. This discount would not affect the annual cost presented in Table 28, as that is based on the assumption that the average patient weight is < 100 kg.

The acquisition costs of the cheapest available pens for each drug are equivalent to the cost of the cheapest available dose. Therefore, the annual acquisition costs for the self-administration route are equivalent to those for biologics administered during outpatient visits.

Of note, the estimates for the additional acquisition costs for the first year (see Table 28) are presented for information only. They were not used in any analyses given that the population in this assessment are people experienced in biologics.

Dose tapering

According to EULAR recommendations for the management of RA with sDMARDs and bDMARDs,⁶⁶ tapering of biologics should be considered in people who are in persistent remission after having tapered glucocorticoids, especially if this treatment is combined with a conventional sDMARD. In this context, tapering means reduction of the dose, for example reducing ETN from 50 mg/week to 25 mg/week,⁸⁷ or increasing the interval between applications (‘spacing’), for example increasing the interval between ADL injections from 1 week to 10 days, as in the Exeter Biologic Clinic recommendations (see Appendix 13).

The EAG is aware that there is no gold standard on how dose tapering should be carried out. Studies evaluating dose tapering have used different approaches. In clinical practice, dose tapering varies extensively depending on the clinical opinion; for example, according to the Exeter Biologic Clinic recommendations (see Appendix 13), when tapering the ADL dose the dose should be reduced by one-third to 40 mg every 3 weeks and reduced further at 3 months to 40 mg every 4 weeks in people with LDA or remission. However, it may not be a representative strategy because of variations in clinical practice.

In the primary analyses, the assumption of reducing the dose by one-third (the first dose reduction in the Exeter Biologic Clinic recommendations; see Appendix 13) was implemented (see Table 27), while the assumption of halving the dose (the second dose reduction described in Appendix 13) was explored in sensitivity analyses (see Table 39).

TABLE 39

Sensitivity analyses conducted for the results of the cost-utility analyses

Treatment wastage

The dose-tapering strategy suggested in the Exeter Biologic Clinic recommendations (see Appendix 13) is spacing; therefore, when this tapering strategy is used, there is no wastage of the self-administered drugs resulting from partial use of the dose in the prefilled injection pens. Clinical advice indicated that wastage of IFX owing to partial use of vials is usually avoided (Dr Haigh, Royal Devon & Exeter NHS Foundation Trust, November 2018, personal communication).

In the primary analyses, however, wastage of £370 per patient-year was incorporated (see Table 27). This estimate was based on a survey conducted at the Royal Devon & Exeter NHS Foundation Trust (Dr Haigh, Royal Devon and Exeter NHS Foundation Trust, Exeter, December 2018, personal communication) and was derived from data on 119 people with RA who were treated with biologics, and included missed doses and oversupply (defined as delivery of treatment even if > 4 weeks’ supply was available at home). It was assumed that, on average, £370 per patient-year would be wasted in people who were on the full dose of ADL, whereas in people who were on tapered doses wastage would be reduced in proportion to the reduction in treatment dose. In scenario analyses considering other biologics (see Exploratory analyses: etanercept or infliximab and Promonitor), the treatment wastage was also assumed to be proportional to the drug acquisition cost. The effect on the outcome of the no-wastage assumption was explored in sensitivity analyses (see Table 39).

Drug administration

As stated above, ADL, ETN, GLM and CTZ are usually self-administered via subcutaneous injection using a prefilled pen. In this scenario, there is no administration cost for delivery. Alternatively, these drugs may be administered by a district nurse. The average administration cost that was assumed in TA375²³ (which was based on an estimate reported in TA247⁸⁸) was £2.61 (cost year 2012). Given that this cost is quite low and that self-administration of the drugs listed above is very common in clinical practice in England, the effect of the assumption that subcutaneous administration would be performed by a nurse was not evaluated.

The administration cost for IFX is considerably higher as it is administered intravenously over a 2-hour period. Patients may be pretreated with, for example, antihistamine, hydrocortisone and/or paracetamol, and the infusion rate may be slowed in order to decrease the risk of infusion-related reactions, especially if such reactions have occurred previously.⁸⁹ Patients are observed for at least 1–2 hours post infusion for acute infusion-related reactions. Based on clinical advice, IFX is typically administered in outpatient settings.

In DG22,⁹⁰ the administration cost for IFX was estimated to be £287.93 per infusion (2014 prices). In a more recent technology appraisal, TA329,⁹¹ the cost was estimated to be £297 per administration (2015 prices).⁹¹

Grant Smith (Specialist Pharmacist, Royal Devon & Exeter NHS Foundation Trust, December 2018, personal communication) advised us that in the Royal Devon & Exeter NHS Foundation Trust the cost of IFX administration is based on Healthcare Resource Groups (HRGs) for inflammatory bowel disease without interventions, with complications and comorbidities scores depending on patient type. The relevant HRGs from the NHS Reference Costs (2017–18)⁸⁴ are shown in Table 29.

TABLE 29

The HRG codes from the NHS Reference Costs 2017–18, day case

The weighted-average administration cost of £283 per administration (estimated across the unit costs for the HRG codes presented in Table 29) was adopted in scenario analyses considering people with RA who were treated with IFX (see Table 41).

TABLE 41

Cost-effectiveness results for other TNF-α inhibitors

Cost of testing

The costs of testing comprised the cost of the test kits, the staff time to perform a test, the cost of the testing service and the cost of sample transport. Based on the information provided by the companies and on clinical opinion, it was anticipated that minimal additional training would be required by health-care staff to use any of the testing kits that were considered in this assessment. Therefore, training costs were assumed to be negligible and were not considered in the model.

Dr McDonald advised us that laboratories that conduct TNF-α testing have previously negotiated arrangements with the manufacturers of bDMARDs to cover the cost of biological monitoring, including assays and personnel costs (Dr McDonald, Royal Devon & Exeter NHS Foundation Trust, Exeter, December 2018, personal communication). However, based on advice from Dr Jani (University of Manchester, November 2018, personal communication), that might vary by geographical area and may be relevant to certain biologics only (e.g. newer biosimilars).

Cost of managing different health states

Based on published literature, active disease in people with RA is more costly to manage than disease in people in remission or LDA. The major health-care costs (apart from drug acquisition costs) relate to joint replacement surgeries, hospital stays and doctor appointments.⁸²

A range of classification systems and scales have been developed to measure and monitor disease activity in patients with RA, and scales commonly used to measure other domains, such as disability or activity level (e.g. HAQ), are also administered.²⁶ Functional capacity measured with the HAQ was found to be the strongest predictor of costs.⁹⁵ Therefore, direct medical costs for hospitalisations, joint replacements and the number of outpatient visits were included by HAQ-dependency, as explained below.

Resource utilisation in rheumatoid arthritis patients stratified by HAQ score

Barbieri et al.⁸² reported resource utilisation in people with RA treated with IFX, stratified by four HAQ bands (see Table 62, Appendix 18). These estimates were used by the authors to calculate the costs of managing people with RA beyond the first year of therapy, and were based on data from the Norfolk Arthritis Register (NOAR). The NOAR cohort includes 1236 adults who had swelling of at least two joints that had persisted for > 4 weeks. This study reported that, on average, the number of outpatient visits, hospital days and the proportion of patients undergoing joint replacement surgery increased substantially with HAQ score (see Table 62, Appendix 18).

Average cost of an inpatient day, outpatient appointment and joint replacement surgery derived from the relevant HRG codes from the NHS Reference Costs 2017–18⁸⁴ are shown in Table 30 (the derivation of the cost of surgery for RA is explained in Cost of joint replacement surgery).

TABLE 30

The unit costs from the NHS Reference Costs 2017–18

Mean HAQ scores for different levels of disease activity (remission, LDA, MDA and HDA) in people with RA were estimated by Radner et al.⁸³ (Table 31): the mean HAQ score based on the SDAI was 0.39, the mean HAQ score for LDA was 0.72 and the MDA and HDA were characterised by a mean HAQ score of 1.24.

TABLE 31

The HAQ scores for the states of disease activity according to the SDAI, CDAI and DAS28

Using this classification and the cost estimates shown in Appendix 18, Table 62, the costs for managing remission (for scenario 1) and active disease (for scenario 2) were calculated from the corresponding probability density functions for HAQ scores weighted by the health management costs for different HAQ scores, whereas the costs of managing mixed-health states (LDA/active disease and remission/LDA) were derived from joint probability density functions for the relevant HAQ scores (see Appendix 18). The resulting average annual costs for managing remission, remission/LDA, LDA/active disease and active disease health states in people with RA were £902, £1089, £1483 and £1827, respectively.

Cost of managing flares

The cost of managing flares is another important consideration that needs to be parameterised in the model. A study by Maravic et al.⁸⁰ estimated the costs associated with managing flares in people with RA in a French setting. This study focused on investigational costs and treatment costs; rheumatology appointments were not considered (see Table 64, Appendix 20).

The costs of diagnostic investigations per flare and the monthly cost of treatment (excluding bDMARDs)⁸⁰ were converted to GBP based on PPP and inflated to 2017–18 prices using the HCHS pay and price index, resulting in costs of £423 and £68 for diagnostic investigations (per flare) and monthly treatment, respectively (see Table 27).

Cost of managing adverse events

In TA375,²³ the weighted average cost of serious infection in RA patients was estimated to be £1479, based on relevant NHS costs⁹⁷ and weighted by inpatient activity. Conservatively, HRG costs without complications and contraindications were used. This cost, inflated to 2017–18 prices using the HCHS pay and price index (£1622 per infection), was assumed in all analyses (see Table 27).

Health state utility values

The abstracts reporting the INGEBIO study provided results on the average duration of either remission⁴² or remission/LDA⁴³ in both the intervention and the control arms. However, none of the sources contain any definitions of remission.

A definition of remission was provided in Krieckaert et al.⁵³ In this study, health states were based on the categorisation of DAS28 as below:

• remission – DAS28 of < 2.6
• LDA – 2.6 ≤ DAS28 of < 3.2
• MDA – 3.2 ≤ DAS28 of ≤ 5.1
• HDA – DAS28 of > 5.1.

The DAS28 comprises four components: counts of tender joints and counts of swollen joints (both performed by a clinician), the visual analogue scale (VAS) score of the patient’s global health and the laboratory parameter ESR. It has been shown, however, that CRP is more accurate as an indicator of inflammation than ESR, and it is also more sensitive to short-term changes.⁹⁸ A modification of the DAS28, the DAS28-CRP,⁹⁹ that includes the level of CRP instead of ESR was used in Bykerk et al.⁷⁰ to define the disease activity types below:

• severe – DAS28-CRP of > 5.1
• moderate –3.2 ≤ DAS28-CRP of ≤ 5.1
• low –2.6 ≤ DAS28-CRP of < 3.2
• remission – DAS28-CRP of < 2.6.

In the study conducted by Bartelds et al.,⁵⁵ remission was defined as a DAS28 of < 2.6 at all consecutive measurements after a certain time point, with a minimum of two scores of < 2.6 in the case of participants who discontinued treatment prematurely.

In Barnabe et al.,¹⁰⁰ sustained remission was defined as DAS28 of ≤ 2.6 for more than 1 year, whereas non-sustained remission was defined as a DAS28 of ≤ 2.6 for less than 1 year.

In TA375,²³ non-responders, moderate responders and good responders were defined according to the EULAR response criteria (see Table 3).

Health state utility value estimated from the Health Assessment Questionnaire according to Simplified Disease Activity Index, Clinical Disease Activity Index and Disease Activity Score in 28 joints

There are several composite scores to assess disease activity in RA. The definitions of the disease states (i.e. remission, LDA, MDA and HDA) according to the SDAI, the CDAI and the DAS28 from Aletaha et al.¹⁰¹ are presented in Table 32.

TABLE 32

Cut-off points to separate remission and low, moderate and high disease activity states using composite indices SDAI, CDAI and DAS28

Radner et al.,⁸³ at the Medical University of Vienna in Austria, collected data on clinical and laboratory characteristics (including CRP, ESR, the number of swollen and tender joints, pain by VAS, patient’s global assessment of disease activity, evaluator’s global assessment of disease activity and physical function by HAQ) from 356 consecutive people with RA at routine clinic visits (every 3–4 months). In total, 716 visits were documented, with a median of two clinic visits per person (ranging from one to four clinic visits).⁸³ At baseline, 87 participants (24.4%) were in remission, 150 (42.1%) in LDA, 103 (28.9%) in MDA and 16 (4.5%) in HDA, as defined according to SDAI. Owing to the small number of participants in the HDA group, the MDA and HDA groups were combined in further analyses. The differences in functional disability measured by the HAQ scores at three levels of disease activity were evident, and similar conclusions were reached during a sensitivity analysis, when the disease states were assessed according to CDAI and DAS28 (see Table 31). Unless stated otherwise, in the remainder of this article, health states are assumed to be defined by SDAI.

The EAG is aware of several algorithms for converting the HAQ score to utility in RA, and that the estimates of utilities may vary when different mapping algorithms are used.¹⁰² In TA375,²³ a comparison of published relationships between utility and HAQ was conducted (see figure 115 in Stevenson et al.³⁴). Three of the eight compared studies reported data from the UK. Of these three studies, Bansback et al.¹⁰³ included data for UK and Canadian patients, and Kobelt et al.¹⁰⁴ included data for patients in the UK and Sweden; therefore, these were not considered relevant for the purposes of this analysis. Hurst et al.¹⁰⁵ included people with RA in Scotland only. Malottki et al.⁵⁹ used the data set from Hurst et al.¹⁰⁵ to estimate the coefficients of their mapping equation; therefore, there is little difference between their estimates, despite different algorithms being used.

Throughout this monograph, the EQ-5D utility values were mapped from the HAQ scores using the same formula as in Malottki et al.:⁵⁹

$$\text{EQ}‑5\text{D}=(a–b_1)\times (\text{HAQ}–b_2)\times {\text{HAQ}}^2,$$

(6)

where a = 0.804, b₁ = 0.203 and b₂ = 0.045.

Hernández Alava et al.¹⁰⁶ argued that pain should be included as an explanatory variable when estimating QALYs from HAQ scores in people with RA. This approach was used in TA375.²³ However, the estimates presented in this article were obtained without pain scores because the EAG did not have access to patient-level data. Table 33 presents the EQ-5D utility values mapped from the HAQ scores at three levels of disease activity from Radner et al.⁸³

TABLE 33

The EQ-5D utility scores for the states of disease activity according to the SDAI, CDAI and DAS28 mapped from HAQ scores from Radner et al.

Ucar et al.⁴² reported the mean duration of remission in the intervention and control arms. In the economic analysis based on this source, the EQ-5D utility value for remission, 0.718, was applied. The utility value of 0.568 used for a mixed-disease state (LDA/active disease) was approximated by the average of the estimates for LDA and MDA/HDA weighted by the proportion of patients in each health state from Radner et al.⁸³ Of note, when the weighted average of HAQ scores was computed instead and mapped to EQ-5D, the utility value was very similar, 0.571.

As the health states in Arango et al.⁴³ (remission/LDA and active disease) were defined differently from those in Ucar et al.,⁴² in analyses based on the former source,⁴³ the EQ-5D utility score of 0.483 for MDA/HDA was used as the utility value for active disease health state, and the weighted average of the estimates for remission and LDA, 0.665, was used to approximate the utility value for the mixed-health state. When the alternative approach (described above) was used, the resulting utility value was 0.666.

Health state utility values (HSUVs) obtained from HAQ scores reported in Stevenson et al.³⁴ (as described in the following section were assumed in scenario analyses.

Health state utility values estimated from the Health Assessment Questionnaire by European League Against Rheumatism response category

In TA375,²³ the model was based on a EULAR response category (good/moderate/none) to be consistent with the NICE guidance on biologics in RA³⁶ and to align more closely with UK clinical practice in terms of the assessment of response to therapies. The HAQ scores were estimated from the BSRBR-RA database,⁶⁷ which contains values measured at 6-month intervals for up to 3 years for all people with RA on the register. The analysis conducted in TA375²³ was restricted to those with the full set of baseline characteristics and at least two additional HAQ measurements while on bDMARDs. The database included data from 10,186 patients. Of these, 2417, 5492 and 2277 were classed as EULAR good responders, moderate responders and non-responders, respectively (see Table 3).

Figure 6 shows the HAQ trajectory in people with RA treated with bDMARDs. It was observed that the mean HAQ scores for patients with good, moderate or no response (according to the EULAR response criteria shown in Table 3) decreased during the first 6 months after the start of biological therapy (when the magnitude of decrease grows with the level of EULAR response), stabilised at around 6 months and remained rather flat over the remaining 2.5 years of measurement.

FIGURE 6

Mean HAQ score by EULAR response category for patients receiving biologics. Reproduced from Stevenson et al. Contains information licensed under the Non-Commercial Government Licence v2.0, which is available at www.nationalarchives.gov.uk/doc/non-commercial-government-licence/version/2/. (more...)

The HAQ scores that were measured after 6 months of therapy with biologics for all three categories of responders were mapped to EQ-5D utilities, which elicited the values shown in Table 34.

TABLE 34

Utility values based on BSRBR-RA data after 6 months of treatment with bDMARDs

The utility for the remission health state was based on the utility value for good responders (0.496, see Table 34), whereas the utility for the LDA/active disease health state was estimated as the average of utility values for moderate responders and non-responders weighted by their proportions in the BSRBR-RA database, resulting in the utility value of 0.302. These HSUVs were used in sensitivity analyses.

Disutility of flare

The values of utility losses owing to flares were obtained from the Dutch multicentre clinical study ‘BeSt’,⁷² which involved 508 participants who were treated to target for 10 years to achieve a DAS28 of, at most, 2.4 (follow-up data that suffice to establish presence or absence of a flare during at least a single visit were available for only 480 patients).⁷² The BeSt study⁷² considered three types of flares, which were named as ‘A’, ‘minor B’ and ‘major B’ (where ‘major B’ is a subcategory of ‘A’), with the number of occurrences of each (observed during a total of 11,458 rheumatology visits of all patients) shown in Appendix 21, Figure 10, and the definitions, frequencies and HAQ scores of each described in Table 35. The mean HAQ score of patients with no flare at a visit was estimated as 0.53 (SD 0.56).

TABLE 35

The definition of flares from Markusse et al. and the corresponding HAQ and utility values

Functional mobility of patients with these types of flares was measured using HAQ scores (mean and SD values are also included in Table 35). The loss of QALYs was computed as the difference between the mapped HSUVs of patients with each type of flare and the mapped HSUVs of patients in the absence of flares. The estimated disutility values are shown in Table 35.

Disutility of serious adverse events

People with RA have increased susceptibility to serious infections owing to the features of RA, comorbidity and immunosuppressive treatment.¹⁰⁷ It has been shown that TNF-α inhibitors increase the risk of serious infection up to two-fold.¹⁰⁸ The EuroQol-5 Dimensions three-level version (EQ-5D-3L) disutility value for England of 0.156 over 4 weeks (equivalent to the loss of QALYs of 0.012) that is associated with severe infections was reported in the observational study ‘Genomics to combat Resistance against Antibiotics in Community-acquired lower respiratory tract infections (LRTI) in Europe’ (GRACE) of the management of patients with acute cough/LRTI in primary care.⁸⁵ Data were collected in 13 European countries (including England and Wales) from adults (aged ≥ 18 years) who reported to their primary care clinicians with cough and LRTI.⁸⁵ EQ-5D-3L scores were generated using the country-specific UK value set, in which the original data were collected from non-institutionalised adults in England, Scotland and Wales (with a total of 2997 participants) between August and December 1993.

The effect of SAEs on costs and QALYs was modelled in the primary analyses. It should be noted, however, that in the analyses, assuming that TDM affects the duration of remission/LDA and the rates of flares and AEs, there is a risk of double-counting the effect of flares and AEs on HRQoL given that it is possible that the disutilities have already been incorporated into health-state utility values.

Consistency between utility values

As shown in Gülfe et al.,¹⁰⁹ there may be discrepancies between utility values that are measured in different countries (in our case Spain, Austria, the Netherlands and the UK), which may occur owing to differences in distinct preference sets for those countries. Figure 11 (see Appendix 21) shows EQ-5D-3L scores obtained using British and Swedish preference sets for people with established RA being treated with TNF-α inhibitors.

The population considered in the INGEBIO study was mixed. This trial recruited 169 people, 63 with RA (37.3%), 54 with PsA (32%) and 52 with AS (30.8%). Gülfe et al.¹¹⁰ also studied a mixed population, with two (RA and PsA) out of the three diseases the same as in the INGEBIO study; the third disease was SpA, which is usually considered as a phenotypically heterogeneous disease with PsA and AS as its best-studied manifestations.¹¹¹ One of the aims of this study¹¹⁰ was to analyse trends in health utilities in people diagnosed with three types of arthritis: 2554 people with RA (who constituted 68.8% of the total population), 574 with PsA (15.5%) and 586 with SpA (15.8%), who started treatment with TNF-α inhibitors. Data for the period from May 2002 to December 2008 were obtained from the Southern Sweden Arthritis Treatment Group register, which was set up in 2002 and collects health utility data from routine clinical follow-up. Treatment courses are classified as first, second or third or further TNF-α inhibitor. Among the three subpopulations, people with RA were typically older, had tried more DMARDs, were more often treated with a concomitant DMARD and were more often female than the other populations. Figure 12 (see Appendix 21) shows similar response patterns in people with RA, PsA and SpA at 6 months after the start of the first TNF-α inhibitor treatment course.¹¹⁰

In Arango et al.,⁴³ 19 patients who discontinued treatment were excluded from the analysis, although those patients were included in the ITT analysis reported in Ucar et al.⁴² As shown in Gülfe et al.,¹¹⁰ RA patients who terminated therapy for any reason had demonstrated lower utility gain by the time of withdrawal, which is illustrated in Appendix 21, Figure 13.

Although the use of all available data increases the generalisability of the study, it may also lead to lower utility estimates than when using data for only those participants for whom complete follow-up information is available (see Figure 14), as incomplete records may be a result of, for example, withdrawals from treatment owing to adverse effects of the intervention.

Of note, the utility values reported in this section were not used in the economic analyses.

Mortality

Although there is evidence of an association between HAQ improvement and reduced mortality risk, the impact of TNF-α testing on mortality was not considered owing to the short-term time horizon adopted in this study and the relatively small difference in the mean duration of remission⁴² and remission/LDA⁴³ across the treatment arms in INGEBIO.

Impact of therapeutic drug monitoring on flare rate only

In the sensitivity analyses, assuming that TNF-α monitoring affects the rate of flares only in patients treated with biologics (as in Gavan¹⁷), the ICERs in scenario 1⁴² and scenario 2⁴³ were £95,070 and £29,599 per QALY, respectively (see Table 39).

When this assumption was implemented in exploratory analyses for the other TNF-α inhibitors (see Exploratory analyses: etanercept or infliximab and Promonitor and Table 41), ICERs were either very close to £30,000 per QALY gained or well above this cost-effectiveness threshold.

Impact of the cost of the initial phlebotomy appointment

Scenario analyses were conducted that assumed that trough samples are taken at the time of existing doctor appointments (i.e. a phlebotomy appointment would not be required). The costs for reflex or concurrent, single or duplicate testing implemented in these analyses are shown in Table 61. In scenarios with reflex testing, it was assumed that the proportion of patients who would need to undergo antibody testing was either 4.7% or 35.8% (see Reflex versus concurrent testing).

When the cost of phlebotomy appointments was implemented together with the other assumptions on testing (as described above), the ICERs were under £20,000 per QALY gained in all analyses for Ucar et al.,⁴² whereas SOC dominated the intervention in the analyses parameterised from Arango et al.⁴³ (Table 39).

However, when this cost was excluded, TDM dominated SOC in all analyses based on Ucar et al.,⁴² whereas the intervention was less costly and produced fewer QALYs than SOC in all analyses for Arango et al.⁴³ (see Table 39), with ICERs of under £20,000 per QALY gained.

Proportion of flared patients on tapered doses, whose treatment dose would be restored to full

A US study⁷⁰ reported statistics on flare that showed that at least 45% of treatment strategies for coping with flares did not involve a dose increase or any other change of medication. Dr Haigh (our clinical advisor) (Royal Devon & Exeter NHS Foundation Trust, Exeter, 2018, personal communication) confirmed that in about only two-thirds of all flared patients on tapered doses would the dose be switched back to full.

Therefore, the effect of the flare management strategy outlined in Bykerk et al.,⁷⁰ that is the assumption that in only 55% of flared patients would the dose of ADL be fully restored, was evaluated. Another assumption, that all patients who flared while on tapered doses would stay on the same dose,^112,113 was also tested. The resulting ICERs were under £20,000 per QALY gained in the analyses for scenario 1,⁴² whereas SOC dominated TDM in the analyses for scenario 2 (see Table 39).⁴³

The number of tumour necrosis factor alpha tests per patient-year

Under the assumption of 6-monthly testing, SOC was dominant in scenario 2 and the ICER in scenario 1 was £36,756 per QALY gained (see Table 39).

Discounts for the price of Humira^®

One-way deterministic sensitivity analyses were conducted based on data from Ucar et al.⁴² and Arango et al.,⁴³ in which the Humira acquisition cost was reduced by 20–80% (Table 40).

TABLE 40

One-way deterministic sensitivity analyses for the acquisition cost of Humira

Regardless of the assumed reduction in the ADL acquisition cost, SOC was estimated to dominate the intervention when data from Arango et al.⁴³ were used, whereas the ICERs in the analyses based on Ucar et al.⁴² were under £20,000 per QALY gained (see Table 40).

Discounts for the price of Promonitor assays

The costs for the Promonitor test kits assumed in the economic analyses are shown in Appendix 15. Grifols–Progenika also offers price discounts, which depend on the uptake of testing, single or duplicate testing, concurrent or reflex, with different number of tests per year. Therefore, additional cost–utility analyses for the levels of discounts proposed by the company were also conducted (the results are not reported here).

Other scenario analyses

The other sensitivity analyses conducted are listed below:

• tapering strategy of dose halving (see Dose tapering)
• cost of treatment wastage assumed to be zero (see Treatment wastage)
• mean flare duration of 19 days (see Duration of flare)
• health-state utilities estimated from TA375 (see Health state utility values estimated from the Health Assessment Questionnaire by European League Against Rheumatism response category)
• disutilities for major B and minor B flares as defined by Markusse et al.⁷² (see Disutility of flare).

The resulting ICERs were under £20,000 per QALY gained in all analyses for scenario 1 (based on Ucar et al.⁴²), whereas SOC dominated the intervention in all analyses for scenario 2 (parameterised from Arango et al.⁴³) (see Table 39).