Health economic results

Missingness of data at each follow-up time point

Table 25 shows missingness of data for both groups per resource use category at each follow-up time point. Data were missing in a non-monotonic pattern, as the proportion of missingness decreased between 6 weeks and 3 months and increased at subsequent time points. The percentage of missing data at 12 months was significantly higher than for other time periods, mainly due to the impact of the COVID pandemic on research. However, there was no significant interaction among participants randomised before and after the UK’s pandemic lockdown (i.e. 23 March 2020). Data were considered missing if a trial participant failed to complete a resource use section at a follow-up point despite all other resource use categories being completed. For example, data were classified as missing if outpatient care visits were not completed at 6-week follow-up period, but all other resource use categories were completed. Imputation was done at the aggregate resource use level at each follow-up time point. Similarly, an EQ-5D-5L response needed to be complete in all five domains at a time point to be classified as non-missing.

Resource use

The mean number of NHS and PSS resource uses at each follow-up time point are shown in Appendix 4. Resource use was similar comparing groups except for the use of hospital physiotherapy visits.

The ARTISAN Plus group had a higher number of physiotherapy contacts compared to the ARTISAN-only group. However, this difference gradually reduced over time, and use was similar at 12 months. Both the level and difference in physiotherapy visits comparing treatment arms were higher when reported by sites than when recalled by patients.

Economic costs

NHS and PSS costs were similar at each time point, except for outpatient services costs, which were higher in the ARTISAN Plus group, as shown in Table 26. The higher costs in the outpatient services were largely driven by the higher number of physiotherapy contacts. However, outpatient services costs gradually reduced and were similar when comparing groups at the 6-month follow-up point. Total NHS and PSS costs were higher in the ARTISAN Plus group at the 6-week time point (Table 27) but converged over subsequent time points, as shown in Figure 11.

FIGURE 11 Unadjusted NHS and PSS costs and EQ-5D-5L utility estimates at each time point per trial group. (Figure not reproduced here; see the original article.)

Health outcomes

Table 28 summarises the unadjusted EQ-5D-5L utility estimates and between-group mean differences at each time point and across the follow-up period for cases with complete data. There was a statistically significant increase in utility for the ARTISAN Plus group at the 3-month follow-up time point. Utility estimates were similar at subsequent time points and converged over time (Figure 11).

Cost-effectiveness analysis

Cost-effectiveness results are presented in Table 29 with ARTISAN as the reference treatment and ARTISAN Plus as the comparator. Analytic time horizon covers the period from randomisation to 6 months post randomisation. Point estimates of the incremental costs and QALYs are shown with 95% CIs. The probability of ARTISAN Plus being cost-effective is presented at three willingness-to-pay thresholds (£15,000, £20,000 and £30,000). The mean NMB is similarly presented at these thresholds.

Base-case analysis

Patients in the ARTISAN Plus group had a small, non-significant increase in quality of life of 0.019 QALYs (95% CI −0.0005 to 0.0375) at a small, significant increased cost of £64 (95% CI 33 to 207) over the follow-up period. The ICER for the base-case analysis of £3373/QALY suggests that ARTISAN Plus is a cost-effective alternative: the probability of ARTISAN Plus being cost-effective at a willingness-to-pay threshold of £30,000 is 0.946.

Figure 12 shows a graphical representation of the cost-effectiveness estimates on the ICER plane, the probability that ARTISAN Plus is cost-effective at different cost-effectiveness thresholds and the NMB of ARTISAN Plus at different cost-effectiveness thresholds (£0 to £100,000/QALY).

FIGURE 12 ARTISAN Plus vs. ARTISAN: cost-effectiveness estimates on the ICER plane. (Figure not reproduced here; see the original article.)

At a cost-effectiveness threshold of £30,000/QALY, the expected value or information38 (EVPI) is £7 per participant. Given an incidence of 8.2–23.9 per 100,000 people per year in the UK, the monetary value of removing all uncertainty from the cost-effectiveness results ranges from £39,032 to £113,764 per annum. However, the technological lifespan of the evidence for ARTISAN Plus (in years) is uncertain and so the total EVPI has not been estimated.

Sensitivity analysis

Incremental costs and QALYs were compared under different analytic scenarios. These sensitivity analyses support the base-case findings. In the complete case analysis, the ARTISAN Plus group had a reduction in costs of £61 (95% CI −357 to 201) and an increase in QALYs of 0.0189 (95% CI −0.0157 to 0.0546). An ICER of −£3245/QALY indicates dominance of the ARTISAN Plus group over the ARTISAN-alone group. While the reductions in costs and QALYs were not statistically significant, the probability of ARTISAN Plus being cost-effective at a £30,000 per QALY threshold is 0.846. For the societal perspective, the ICER increased to £22,073/QALY. ARTISAN Plus had a non-significant increase in costs of £422 (95% CI −240 to 1099) and a significant increase in quality of life of 0.0191 (95% CI 0.0002 to 0.0373). The probability of ARTISAN Plus being cost-effective at a £30,000/QALY threshold is 0.638. Limiting the analysis time window to 6 months’ follow-up, the ICER increased to £9163/QALY. The ARTISAN Plus group had a non-significant increase in quality of life of 0.013 QALYs (95% CI −0.0032 to 0.0293) at an increased cost of £120 (95% CI 33 to 207). The probability of ARTISAN Plus being cost-effective at a £30,000/QALY threshold is 0.845.

Findings are similar when site-reported physiotherapy data were used in place of patient-reported data. The ARTISAN Plus group had a non-significant increased cost of £95 (95% CI −41 to 231) and QALYs of 0.0188 (95% CI 0.0041 to 0.0417). The ICER increased to £5084/QALY. The probability of ARTISAN Plus being cost-effective at a £30,000/QALY threshold is 0.921. Including the costs of missed appointments in site-reported physiotherapy data, the ARTISAN Plus group had increased costs of £104 (95% CI −31 to 240) and QALYs of 0.0176 (95% CI −0.006 to 0.0415). The ICER increased to £5932 and the probability of ARTISAN Plus being cost-effective at a £30,000/QALY threshold is 0.873. Findings were also similar when the EQ-5D-5L responses were valued using the mapping algorithm developed by Alava et al. The ARTISAN Plus group had non-significant incremental costs of £69 (95% CI −88 to 226) and incremental QALYs of 0.0157 (95% CI −0.0061 to 0.0374). The probability of ARTISAN Plus being cost-effective at a threshold of £30,000/QALY is 0.867.

It is notable (Tables 27 and 28) that ARTISAN Plus is generally characterised by small and imprecise increases in cost and health (measured as QALYs). Examining the NMB values at any threshold, it is apparent that all sensitivity analyses are statistically similar to the base case.

Longer-term economic modelling

The protocol allowed for decision-analytic modelling to estimate longer-term cost-effectiveness if costs and outcomes do not converge over the analytic time horizon. As shown in Figure 11, Tables 28 and 29, incremental costs and QALYs converged at between 3 and 6 months. Due to their convergence within the follow-up time horizon, longer-term extrapolation of cost-effectiveness is unlikely to provide further insight or understanding.