Cost-effectiveness of hydroxychloroquine versus placebo for hand osteoarthritis: economic evaluation of the HERO trial

Overview

The HERO trial was a randomised, double-blind, placebo-controlled trial with an economic evaluation conducted alongside. Patients with symptomatic and radiographic hand OA aged 18 years or over were recruited from 13 primary and secondary care centres in England; description of the study design and interventions are available in full elsewhere.^11,12 A total of 248 participants were included in the trial; participants were aged 62.7 years on average (ranging from 40 to 88 years), with 81.9% being female. The trial involved 12-month follow up of the use of HCQ (200–400 mg) or placebo, in addition to ongoing usual care, where participants were randomised on a 1:1 basis. The study was registered with International Standard Randomised Controlled Trial Number ISRCTN91859104 on 17 October 2012, and received approval by Leeds East Research Ethics Committee (12/YH/0151) and the UK Medicines and Health Regulatory Authority. All participants provided written informed consent prior to screening and involvement in the trial.

A within-trial economic analysis was undertaken from the perspective of the UK NHS and PSS, with results presented in terms of the incremental cost per QALY (cost–utility analysis), and the cost per unit of reduction in pain score (cost-effectiveness analysis), which used the trial’s primary outcome of hand pain severity. Patient-level data were collected for costs (healthcare resource use, medication use, and of HCQ drug) and health outcomes (EuroQol EQ-5D-5L and hand pain severity) over the 12-month follow up period, using questionnaires completed by participants at baseline, six months and 12 months. The base-case analysis was undertaken using multiple imputation to deal with missing data, with complete case analysis explored as a sensitivity analysis. A secondary analysis from the broader societal perspective was also conducted.

Health outcomes

QALYs were estimated by participants’ completion of the EQ-5D-5L (EuroQol Group, Rotterdam, The Netherlands)^13,14 in self-completed questionnaires at baseline, six and 12 months follow up. The EQ-5D-5L is a preference-based measure that provides a descriptive profile of an individual’s health state¹⁵ and comprises five dimensions (mobility, self-care, usual activities, pain/discomfort and anxiety/depression) which each have five possible levels of response (no, slight, moderate, or severe problems, or unable to/extreme problems). The crosswalk value set developed by van Hout et al.¹⁶ was utilised for the estimation of utilities. A utility score of one indicates perfect health, a utility of zero indicates states equivalent to death, and negative scores indicate states considered to be worse than death. QALYs were calculated by plotting the utility scores at each of the three time points and estimating the area under the curve,^17,18 with adjustment made for baseline utility.¹⁹ In addition to the evaluation of QALYs via a cost-utility analysis, the primary clinical outcome of the trial, hand pain severity, was also investigated via a cost-effectiveness analysis. Specifically, the reduction in hand pain severity was measured using a NRS (0 to 11, where higher scores represent worse levels of pain) and assessed by calculating the difference in hand pain NRS at 6 months and hand pain NRS at baseline.

Resource use and unit costs

Participants were asked about their utilisation of healthcare services in relation to their hand or hand pain via self-completed questionnaires at baseline, six and 12 months. Resource use items included in the questionnaire aimed to represent relevant services used by patients with hand OA, specifically within primary care (visits to the general practitioner (GP), nurse, and other primary care services), community care (physiotherapist and occupational therapist visits, and other community care services) and in the hospital setting (outpatient attendances, day case visits, and accident and emergency attendances). Total resource use per participant for the duration of the 12-month trial period was calculated by multiplying each resource use item by the corresponding unit cost (Table 1); unit costs were derived from established national costing sources.^20,21 Participants were also asked via the study questionnaires about personal expenses, namely travel and childcare costs relating to healthcare appointments and over-the-counter medication costs, which fed into the secondary analysis, undertaken from a broader societal perspective. Unit costs were not required for these costs, as participants were asked to specifically state the amount of money spent on these, rather than the number of resources used.

Medications

Participants received a daily dose of HCQ in either 200, 300 or 400 mg (for 300 mg, alternating doses of 200 mg and 400 mg were taken); for costing purposes, participants were assumed to continue on the same dose throughout the study period (which was, in fact the case, with the exception of one participant). Unit costs of HCQ and medications used by participants, as recorded by study investigators, were obtained from the British National Formulary.²² The HCQ costing was based on the net price of HCQ sulfate 200 mg (60-tablet pack) being £5.15.²³ The cost of HCQ use over the 12-month period was applied, unless there was information recorded regarding the participant stopping/withdrawing from treatment. The cost of placebo tablets represented a research cost and hence excluded from the costings. Information regarding participants’ use of oral and topical medication was recorded by study investigators at regular intervals during the trial period; investigator forms at 3, 6, 9 and 12 months were used for the medication costing. To attach costs to participants’ medication use, each medication was categorised (e.g. oral opioid, topical NSAID, antidepressant/neuropathic therapy), and within each category the average cost of the most commonly occurring medications was applied. A total medication cost per participant was generated by summing the cost of all medications used by the participant.

Data analysis

The base-case analysis was conducted on an intention-to-treat (ITT) basis; the two groups (HCQ versus placebo) were compared according to their initial random allocation, regardless of whether protocol deviations or withdrawals took place. Cost and outcome data were collected prospectively during the study and compared for the two groups over 12 months, hence discounting was not required. Costs are presented in UK £ sterling at 2015 prices and the analysis was undertaken in Stata 13© (StataCorp 2013, TX, USA) (RRID:SCR_012763); an open-access alternative is R (RRID:SCR_001905). The base-case analysis utilised a dataset generated via multiple imputation with chained equations and predictive mean matching to deal with missing data, based on the assumption that data were missing at random. The imputation model included baseline hand pain severity, concomitant analgesic use, average grip strength, body mass index (BMI), age and gender (all consistent with the clinical primary analysis model), baseline costs and baseline utility. A complete case analysis was also undertaken as a sensitivity analysis, where patients with any missing data were excluded, and available case analysis was used for initial exploration of the data.

Mean differences in costs and QALYs were compared for the two groups to assess the cost-effectiveness of HCQ versus placebo; estimates were produced using seemingly unrelated regression equations, with 95% confidence intervals estimated using bias corrected and accelerated bootstrap methods. The regression model adjusted for baseline utility and covariates consistent with those used in the trial’s statistical analysis. Differences between the groups were found to be statistically significant if P<0.05 and are presented alongside confidence intervals around the differences in costs and outcomes. Incremental cost-effectiveness ratios (ICERs) were used to present the findings, where appropriate, thereby indicating the additional cost per extra unit of benefit, calculated by dividing the mean difference in costs between trial groups by the mean difference in health outcomes. An ICER is compared with cost-effectiveness threshold values, to determine whether the intervention represents value for money. The NICE cost-effectiveness threshold for the UK ranges between £20,000 and £30,000 per QALY,²⁴ hence both of these thresholds featured in the analysis. Results are also presented in terms of net health benefit (NHB)²⁵ to aid interpretation of the findings; a positive NHB indicates an intervention is cost-effective, based on the cost-effectiveness threshold under consideration. The decision uncertainty surrounding the cost-effectiveness findings was explored using cost-effectiveness planes and cost-effectiveness acceptability curves (CEACs).²⁶ A total of 5,000 estimates of incremental costs and incremental effects were generated, using non-parametric bootstrapping, and plotted on the cost-effectiveness plane. CEACs have been used to depict the probability of HCQ being cost-effective relative to placebo for a range of cost-effectiveness thresholds.

In addition to the cost-utility analysis described above, the cost-effectiveness analysis that explored the cost per unit of reduction in hand pain score utilised a mixed model, in line with the statistical model used for the clinical effectiveness analysis. This model accounted for baseline covariates and used an exchangeable covariance structure to account for the correlation of observations from the same patients over time.

Resource use and costs

Mean healthcare resource use is summarised in Table 2 by group and timepoint, for all available cases. Most notable health services used by participants over the duration of the trial (sum of 6 month and 12 month resource use) were for: 2.7 GP visits (at GP practice), 2.2 outpatient clinic visits, 1.1 nurse visits (at GP practice), and 0.7 physiotherapist visits per participant, on average. At 12 months, participants in the HCQ group had fewer hospital outpatient visits, GP visits and nurse visits, but higher physiotherapist, occupational therapist, and accident and emergency (A&E) visits. In terms of the total mean cost for each resource use item, Table 3 summarises this according to group and the mean difference between groups, based on all available cases. Costs were found to be lower for the HCQ group, on average, for visits to the GP (at GP practice), nurse, A&E and hospital outpatient attendances, but were higher for GP visits (at home), physiotherapist and occupational therapist visits. However, none of the differences between groups were statistically significant.

The mean (standard deviation, SD) cost of HCQ for the 12-month period was £42.14 (£16.23), with HCQ costs estimated based on the following: seven participants prescribed a daily dose of 200 mg, 85 receiving 300 mg and 32 receiving 400 mg HCQ daily. A total of 90 HCQ participants received HCQ for the 12-month trial duration, whilst 34 received HCQ for durations less than this (ranging from 15 days to 308 days). Mean (SD) medication costs over the 12-month period were £282 (£235) for HCQ participants and £300 (£232) for placebo participants, for all available cases over the 12-month follow up. Hence, larger components of the total cost derived from hospital outpatient clinic appointments, GP visits (at GP practice) and medication costs, with costs of the drug, occupational therapist and physiotherapist visits being smaller cost drivers.

Health outcomes

Utility and quality-adjusted life years

Estimation of the proportion of participants who reported the EQ-5D-5L levels (1 to 5) by dimension and group identified that the majority of participants (at least 95%) reported having problems in terms of pain/discomfort at all time points in both groups, whereas problems with self-care were reported for approximately a quarter to a third of participants (Table 4). Baseline utility was slightly higher in the HCQ group (0.615 versus 0.612 for HCQ and placebo, respectively) and the differences found between the groups at both 6 and 12 months were found to be very small (Table 5). The difference in QALYs at 12 months (HCQ−placebo), controlling for baseline utility, was 0.0012 (95% CI: −0.0251 to 0.0276), for available cases (n = 88 HCQ, n = 95 placebo). EQ-5D visual analogue scale (VAS) scores were found to be similar at baseline on average (72.4 for HCQ; 73.9 for placebo), and by 6 months, there was an increase for the HCQ group (to 74.5) compared with a slight reduction for the placebo group (73.6). However, by 12 months, the VAS score increased for both groups: 74.1 and 75.1 for the HCQ and placebo groups, respectively.

Hand pain score

The differences between treatment groups in terms of hand pain score were found to be small at each follow up time point and not statistically significant. At the primary endpoint of 6 months, patients in the HCQ group scored on average 0.16 points higher on the pain scale than those in the placebo group, therefore indicating worse pain for HCQ than placebo patients. As there was not a reduction found in terms of pain, it was not considered meaningful to calculate the ICER as originally intended, as it would have instead represented the cost per additional unit of pain.

Cost-utility analysis and uncertainty

The use of HCQ versus placebo was associated with a cost saving of £11.80 per participant, on average, in the base-case analysis (Table 6). In terms of the effect of HCQ on health-related quality of life, only marginal differences in QALYs were demonstrated. The base-case analysis found 0.005 fewer QALYs for participants in the HCQ group compared to the placebo group, on average, over the 12-month time horizon. It was therefore appropriate to report the results in terms of the cost per QALY lost, rather than per QALY gained that is more commonly seen; the resulting ICER showed cost savings of £2,267 per QALY lost, which implies that £2,267 would need to be saved in order to justify a loss of one QALY, i.e. costs are being saved but health is being lost. The NHB was found to be negative: −£92.30 (95% CI: −£102.11 to −£82.49) and −£144.34 (95% CI: −£158.67 to −£130.02) for the £20,000 and £30,000 cost-effectiveness thresholds, respectively.

Figure 1 illustrates the 5,000 bootstrap sample estimates, which are spread across the four quadrants of the cost-effectiveness plane quite evenly. The probability of HCQ being cost-effective for different willingness to pay thresholds is shown in Figure 2. For the NICE threshold of £20,000 per QALY, the probability of HCQ being cost-effective (under the base-case scenario) is 0.40, and similar at 0.39 for a threshold of £30,000 per QALY.

Figure 1. Scatter plot on the cost-effectiveness plane: incremental costs and incremental QALYs (base-case).

Figure 2. Cost-effectiveness acceptability curve (base-case).

Secondary analysis

The secondary analysis undertaken from a societal perspective (incorporating personal costs for over-the-counter medication, childcare and travel for appointments) showed participants in the HCQ group incurred marginally lower mean costs (−£2.66; 95% CI: −£6.14 to £0.81), and slightly fewer QALYs (−0.010; 95% CI: −0.011 to −0.010) compared with placebo. The marginal cost saving for the HCQ group was not statistically significant. The resulting ICER was £265 per QALY lost, with a negative NHB at both the £20,000 threshold (−£199) and the £30,000 threshold (−£299).

Sensitivity analysis: complete case analysis

The findings of the cost-utility analysis remained in the same direction for the complete case analysis, which comprised 76 participants (31% of total participants) who had complete economic data profiles, although a larger cost difference was found than in the base-case. Participants in the HCQ group incurred costs that were £50.95 lower than those in the placebo group, and 0.005 fewer QALYs, i.e. similar to the QALY difference in the base-case. However, the cost and QALY differences were not found to be statistically significant for the complete case analysis.

Underlying data

Full underlying (non-aggregated) data cannot be made publicly available since the ethics approval of this study does not cover openly publishing non-aggregated data.

In order to access this data, it must be requested from the corresponding author. Data requestors will have to provide: i) written description and legally binding confirmation that their data use is within the scope of the study; ii) detailed written description and legally binding confirmation of their actions to be taken to protect the data (e.g., with regard to transfer, storage, back-up, destruction, misuse, and use by other parties), as legally required and to current national and international standards (data protection concept); and iii) legally binding and written confirmation and description that their use of this data is in line with all applicable national and international laws (e.g., the General Data Protection Regulation of the EU).