• No results found

Baseline characteristics

Chapter 4 Health economic analysis

Introduction

As described inChapter 1, DMARDs are used in the early stages of management of RA. However, even when there is an initial positive response, treatment efficacy often reduces over time. bDMARDs are usually given to patients experiencing insufficient response to conventional DMARDs, but at a markedly higher cost of around £9500 per patient per year, compared with around £450 per year for conventional therapy.122During 20078, expenditure on bDMARDs for the treatment of RA alone ranged between

£0.8M and £3.5M per acute trust, with expenditure on bDMARDs accounting for the highest pharmaceutical spend within some trusts.122

Tumour necrosis factor inhibitor drugs are a type of bDMARD that have been found to be costly but highly effective.31–33However, NICE currently approves only rituximab following TNFi non-response, with the

use of alternative TNFi being permitted only when rituximab (and/or MTX that is co-prescribed) is contraindicated.

An economic evaluation was conducted to estimate the cost-effectiveness of alternative TNFi or abatacept compared with the current practice of rituximab in patients with RA who have failed treatment with an initial TNFi. The economic evaluation was conducted alongside the SWITCH clinical trial so that only the data collected within the (reduced) trial were analysed. Originally, the health economic analysis included a within-trial analysis and a decision analytical model. Given the early termination of the trial and the

consequent reduced period of follow-up, the health economic analysis was adapted to include a within-trial cost-effectiveness analysis over 48 weeks and a value of information analysis to inform future research.

Methods

Aim and end points

The primary aim of this analysis was to assess the cost-effectiveness of the use of abatacept or alternative TNFi compared with the current practice of rituximab in patients with RA who have failed treatment with an initial TNFi. The primary end point was the cost per quality-adjusted life-year (QALY) gained. The methods used for this within-trial analysis were guided by the recommendations from the NICE methods guide.123 Perspective and time frame

The study adopted a NHS and Public Social Services perspective for cost evaluation, but a broader societal perspective was adopted for secondary analysis to incorporate costs to patients and productivity costs. Costs and benefits for the base-case analysis were calculated for the study period of 48 weeks. As the time frame of the trial was<1 year, discounting of the costs and benefits was not required.

Measurement of effectiveness

This analysis used the QALY as the main outcome measure. QALYs are a generic measure of health state that take account of both quality and length of life such that 1 QALY is equal to 1 year in full health.124

Health-related quality of life was measured using the EuroQol 5 Dimensions, 3 levels (EQ-5D-3L). The EuroQol 5 Dimensions (EQ-5D) is a commonly used generic measure of health-related quality of life and is NICE’s preferred outcome measure for cost-effectiveness analyses.123The questionnaire comprises five

domains: mobility, self-care, usual activities, pain/discomfort and anxiety/depression. Each domain consists of three levels: no problems, some problems and severe problems.125

DOI: 10.3310/hta22340 HEALTH TECHNOLOGY ASSESSMENT 2018 VOL. 22 NO. 34

© Queen’s Printer and Controller of HMSO 2018. This work was produced by Brownet al.under the terms of a commissioning contract issued by the Secretary of State for Health and Social Care. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

The EQ-5D-3L was administered at baseline and at follow-up visits in weeks 12, 24, 36 and 48. Responses were converted to health state utility values using the UK general population time trade-off tariff values.126 Measurement of costs

Health-care resource utilisation data were collected using patient self-reported questionnaires covering primary care [e.g. general practitioner (GP) and nurse visits] and secondary care (e.g. hospital stays/visits) resource use over the trial period (seeAppendix 2). The questionnaires also captured personal costs to patients related to RA (e.g. travel to/from hospital and cost of aids) and any impact the disease had on their income over the trial period. The resource-use questionnaires were completed at weeks 12, 24, 36 and 48 by the research teams at the participating centres and were supplemented by case report forms (CRFs) capturing data on hospital inpatient or outpatient visits. When there were discrepancies between the CRFs and the patient-completed forms, the CRFs were given precedence. Unit costs for health service staff and resources were obtained from the Personal Social Services Research Unit (PSSRU) report in 2015, entitledUnit Costs of Health and Social Care 2015,127andNHS Reference Costs 2014 to 2015.128 For

medications, a unit cost per treatment received was assigned. The Commercial Medicines Unit’s electronic market Information Tool (eMit) was used to cost the drugs when possible.129However, when drugs were

not listed on eMit, costs were taken from theBritish National Formulary(BNF).130When unit costs were not

available, targeted literature searches were used to provide the relevant costs which were inflated to 2015 prices (pounds sterling) using an online inflator.131Unit costs are presented inAppendix 15, Table 76. Assumptions related to medication use

A number of assumptions were required in order to measure the costs related to the drugs used within the trial:

l When the patient’s weight was needed to calculate the dose of trial medication, the baseline weight was used. This applied to the one patient using infliximab and, in this case, the associated cost was not affected if the patient’s weight at each clinical assessment had been used instead.

As no stop date was recorded for the trial drugs, the number of doses was deduced from the CRFs making the following assumptions.

1. If the records showed that all infusions were received as per the protocol, the full protocol-defined allocation for the relevant time period was allocated to that patient.

2. In patients who were reported to have received some randomised treatment but the infusions received did not follow protocol because they were delayed, it was assumed that the patient received the full allocation of trial medication.

3. In patients who were reported to have received some randomised treatment but for whom infusions received did not follow protocol because treatment was stopped, it was assumed that the patient received half the allocation of trial medication for that time period.

4. If it was indicated that some of the randomised treatment had been received but it was not reported whether or not all the treatment had been received or whether or not there had been any

modifications to the treatment protocol, it was assumed that the full allocation of treatment for that time period had been received.

5. For treatments administered by injection, if the number of missing injections was recorded, the number of missing injections was taken from the full protocol-defined allocation for the relevant time period. 6. For treatments administered by injection, if some treatments were missed but the number of treatments

missed was not recorded, then it was assumed that half of the allocation for the relevant time period was received.

The following assumptions were applied in order to cost the concomitant drugs used within the trial period.

l When dose was not recorded, it was assumed that the standard dose was received.

l When it was indicated that it was an ongoing drug with no start or end date recorded, it was assumed that it was taken for the full 48 weeks.

l For those patients who took MTX, it was assumed that it was taken orally at an average dose of 15 mg per week based on the relevant literature and expert opinion.132

Missing data

The base-case analysis was conducted using only complete cases. That is, patients were included in the analysis if they had no missing resource use data as well as no missing quality-of-life data. In the case of resource use, no missing data were defined as a resource use form having been completed at all time points. For the resource use questionnaires, if a patient recorded that they had used a form of health care (e.g. GP visit) but did not record the number of visits, the mean number of visits was imputed. For quality of life, complete data were defined as a completed quality-of-life questionnaire returned at each time point. Sensitivity analyses were conducted using imputed data so that all patients were included in the analysis. Two imputation methods were explored: mean imputation and multiple imputation. For the mean imputation, when a QALY value for a given time point was missing, the mean of the non-missing QALYs for the trial arm at that time point was imputed. The same approach was taken to impute missing cost data for resource use. For the multiple imputation, costs for each follow-up and total QALYs were imputed by chained equations using predictive mean matching.133Forty-five data sets were imputed (reflecting the

percentage of incomplete cases), which were then combined using Rubin’s rules.119,134 Analysis

The primary analysis was a cost-effectiveness analysis of the three relevant treatment arms of the trial. A complete-case analysis was the primary method for analysing the trial data and an ITT analysis was undertaken as a sensitivity analysis.

Resource use and costs were quantified and analysed using analysis of variance and independent sample

t-tests. Owing to the small sample size and, subsequently, the potential violation of the underlying normality assumption when usingt-tests, the robustness of the results was checked using a non-parametric bootstrap. Health-state utilities were used to calculate QALYs using an area under the curve approach:

QALY=f½(EQ-5DBaseline+EQ-5D12)/2× 0:231g+f½(EQ-5D12+EQ-5D24)/2× 0:231g +f½(EQ-5D24+EQ-5D36)/2× 0:231g+f½(EQ-5D36+EQ-5D48)/2× 0:231g,

(6)

where EQ-5DBaseline, EQ-5D12, EQ-5D24, EQ-5D36and EQ-5D48are the EQ-5D scores at baseline, week 12,

week 24, week 36 and week 48, respectively; 0.230769 represents 12 weeks out of 52 for each time period:

t = 12

52 = 0:230769 (7)

Total costs and QALYs for each arm of the trial were calculated. For the secondary analysis, a wider cost perspective was adopted to include the total costs incurred by the patients.

Incremental cost-effectiveness ratios (ICERs) were calculated.135An ICER represents the additional cost per

QALY gained for each intervention compared with the next best alternative and is calculated as follows for treatment A relative to treatment B:

ICER = (CostA−CostB)/(QALYA−QALYB), (8)

DOI: 10.3310/hta22340 HEALTH TECHNOLOGY ASSESSMENT 2018 VOL. 22 NO. 34

© Queen’s Printer and Controller of HMSO 2018. This work was produced by Brownet al.under the terms of a commissioning contract issued by the Secretary of State for Health and Social Care. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

where CostAand CostBare the mean costs and QALYAand QALYBare the mean QALYs for groups A and

B. An intervention was judged to be cost-effective using the lower limit of the NICE acceptance threshold of £20,000 per incremental QALY (λ=£20,000) as the decision rule for the analysis.123

The level of sampling uncertainty around the ICER was determined using non-parametric bootstrapping (with replacement) to generate 10,000 estimates of incremental costs and benefits. These were then plotted on the cost-effectiveness plane to visualise the uncertainty around the mean incremental costs and effects. The expected ICERs for the primary analysis were estimated from the means of bootstrapped cost and outcome distributions.

Net monetary benefit (NMB) values were also calculated. Net benefit (NB) combines cost-effectiveness and willingness to pay to give an explicit monetary valuation of the health outcome. It is calculated by rearranging the ICER calculation and incorporating a proposed willingness-to-pay threshold value per QALY.135The

expected value of the NMB was calculated for each treatment. Treatments with positive NMBs provide more health benefit than is displaced by the associated opportunity costs and should be adopted. The treatment with the highest positive NMB is the most cost-effective.135The probability that the treatments were

cost-effective was evaluated by generating estimates of NMB for a range of cost-effectiveness thresholds (λ). This analysis was presented as a cost-effectiveness acceptability curve.136,137The cost-effectiveness acceptability

curve provides decision-makers with useful information regarding the risk of making a wrong decision; however, the decision to fund or not fund a treatment should be made on the expected value of the NMB.

Net monetary benefit is derived for each patient as:

NMB= (λ× QALYs)−costs. (9)

Sensitivity analyses

The following scenario sensitivity analyses were conducted to test the robustness of the conclusions drawn from the results.

1. Mean imputed data: an analysis was conducted using singly imputed data for missing QALYs and costs to enable an assessment of cost-effectiveness using data from all patients.

2. Multiple imputation: an analysis was conducted using multiply imputed data for missing QALYs and costs.

3. Adjust baseline: an analysis was conducted to evaluate the effect of adjusting for baseline differences in EQ-5D score (using an ordinary least squares regression and adjustment: total QALYs over 48 weeks were regressed on trial arm, EQ-5D score at baseline, age at baseline and sex).

4. Subcutaneous MTX: patients could have taken MTX orally or by subcutaneous injection but, as the method was not recorded, at baseline an assumption was made that MTX was taken orally by all patients. Therefore, sensitivity analysis was conducted that explored the alternative scenario that MTX was instead taken via subcutaneous injection by all patients.

Secondary analysis assessed the effect of taking a broader, societal cost perspective. The analysis uses health and social sector costs together with the addition of patient out-of-pocket costs plus values from the EQ-5D to estimate QALYs (replicating the primary analysis). As in the primary analysis the base case used complete cases. Sensitivity analyses using mean imputation and multiple imputation were undertaken.

Value of information analysis

Value of information analysis was conducted to estimate the potential gains from the elimination of uncertainty as a result of conducting additional research. As decisions based on current information are uncertain (because of imperfect information) there is a chance that the wrong decision will be made, resulting in costs being incurred in the form of health benefit and cost of resources forgone. Given the very small sample size, the decision uncertainty is large and, therefore, the value of information analysis is

especially important. The expected value of perfect information (EVPI) is derived from the expected costs associated with the uncertainty in decisions. The EVPI provides the maximum value that a health-care system should be willing to pay for additional evidence to eliminate uncertainty in parameter estimates to inform future decisions, and gives an upper bound for the value of additional research. As information is valuable to all patients with a disease (not just one patient) EVPI can be expressed for the population of patients who could benefit.138,139The EVPI was calculated for the population of the UK who have RA

as follows:

EVPI=EθmaxjNB(j,θ)−maxjEθNB(j,θ). (10)

The bootstrap simulation provides estimates of costs and benefits and, therefore, NB. EθmaxjNB(j,θ) is the

expected NB with perfect information, which is the mean value of NMB in the set when the intervention with the higher NMB is chosen for each simulation, and maxjEθNB(j,θ) is the expected NB with current

information, which is obtained when the intervention with the higher expected NB is chosen across all simulations.140

All of the analyses were conducted in Stata®(version 14, StataCorp LP, College Station, TX, USA) and

Microsoft Excel®(2013, Microsoft Corporation, Redmond, WA, USA).

Results

Sample

Of the 122 patients recruited to the trial, 70 patients with complete resource use data and EQ-5D results (25 rituximab, 24 abatacept and 21 alternative TNFi) were included in the base-case analysis.

Baseline characteristics of the 70 patients analysed in the complete case are presented inTable 21(see

Table 3for baseline characteristics of the ITT population). In all treatment arms more than two-thirds of the patients were female. The average weight was slightly lower in the abatacept group than in the other treatment groups. Fewer patients in the alternative TNFi arm were non-smokers and a higher percentage were past smokers than in the other treatment arms. There was some variation between arms in baseline EQ-5D scores, with the alternative TNFi group having the highest scores. However, the difference between the scores was not statistically significant.