Background and methods
For the health economic evaluation, we compared the PEDAL intervention as an add-on to dialysis with dialysis alone. We estimated the mean between-group difference in costs of the intervention and the mean between-group difference in QALYs accrued by participants during the study, estimated as the area under the health utility curve. All costs and QALY differences were estimated over the 6-month period from randomisation; discounting future costs and effects for societal time preference is not relevant. Total costs and 1 – QALYs were modelled using a generalised linear regression model with a Gaussian distribution and a log-link, adjusting for age, sex and diabetes at baseline. The modified Park test was used to determine the distribution family for the model.
Intervention costs included costs for purchase and maintenance of exercise equipment (pro-rated to apply to the 6-month study period), initial training and supervision of staff, staff delivery of the exercise sessions and the number of sessions. Different scenarios for staff costs have been considered, varying situation on pay band, including/excluding London allowance and varying staff–patient ratios. The scenarios have been based on the staffing assumed to be required when implementing the exercise sessions in usual care, not within the trial. All costs were in Great British pounds; no currency conversions were required.
The cost of the equipment was assumed to be £1000 and the expected lifetime was assumed to be 10 years. This is the estimated cost of the modified Monark cycle ergometer if it were to be commercially available. Equipment costs over the 6-month study period were therefore assumed to be £50. The maintenance cost was £50 per year per bike. This is the cost of a call-out fee and any reasonable wear-and-tear replacements by the manufacturer. It was assumed that one bike covers four patients; therefore, equipment cost per patient was assumed to be £18.75 [(£50 + £25) ÷ 4] when doing the expected number of exercise sessions.
The level of staff required was assumed to be Agenda for Change band 4. This would be the level of staff required in the NHS to deliver this type of exercise intervention. Annual employer costs are £25,865.7180–30,114.3716 outside London and £30,065.7180–34,786.9716 in London. Cost per hour is calculated as annual staff cost ÷ 46 work weeks per year ÷ 37.5 work hours per week. The number of staff required to cover training sessions depends on the distances between satellite units of each study centre. It was assumed that 0.6 members of staff are required to cover training sessions for 12–20 patients. Values of 12 and 20 were used in the analysis. The level of staff required to supervise exercise sessions and train staff to deliver the sessions was assumed to be paid, Agenda for Change band 8. Annual employer costs are £55,078.005–64,755.960 outside London and £61,741.005–71,418.960 in London. It was assumed that one member of staff covers 80 patients.
The expected number of sessions over 6 months was calculated as 365.25 ÷ 2 ÷ 7 × 3 (three sessions per week). The number of sessions completed was calculated as number of expected sessions × percentage compliance reported in the eCRF ÷ 100. The cost per session was derived as (equipment cost + staff cost over 6 months) ÷ number of expected sessions for 6 months. The intervention cost per patient was derived as number of sessions actually attended × cost per session.
Costs in the control group were set to 0, because they received no additional intervention and, therefore, required no additional staff time or equipment.
The area under EQ-5D-5L health utility curve was derived as follows. Only the period from study entry (i.e. randomised and baseline visit complete) to 6 months was taken into account. If the baseline visit was on the date of study entry, the EQ-5D-5L health utility score collected at that visit was used in the calculation as EQ-5D-5L health utility score at entry. If the baseline visit was before randomisation, EQ-5D-5L health utility score was derived as follows:
- if there is no follow-up visit, starting EQ-5D-5L health utility score will be the baseline EQ-5D-5L health utility score
- if there is a follow-up visit, starting EQ-5D-5L health utility score will be determined through linear interpolation between baseline and the earliest follow-up visit thereafter.
The area under curve was derived as follows:
- If the patient was not dead and had no follow-up beyond baseline, the 6-month EQ-5D-5L health utility score was assumed to be the same as the EQ-5D-5L health utility score at entry. The area under the EQ-5D-5L health utility curve was derived as 365.25 ÷ 2 × EQ-5D-5L health utility score at entry.
- If the patient had their 6-month visit after 6 months from entry, EQ-5D-5L health utility score at 6 months was derived through linear interpolation between EQ-5D-5L health utility score at entry and EQ-5D-5L health utility score at 6-month visit. The area under the EQ-5D-5L health utility curve was derived as 365.25 ÷ 2 × (EQ-5D-5L health utility score at entry + EQ-5D-5L health utility score at 6 months) ÷ 2.
- If there was no follow-up available and the patient died after 6 months, EQ-5D-5L health utility score at the time of death was set to 0 and the EQ-5D-5L health utility score at 6 months was derived through linear interpolation. The area under the curve was derived as in the point above.
- If the 6-month visit occurred before 6 months, the EQ-5D-5L health utility score at 6 months was derived through linear interpolation between the 6-month visit and the next available information after that, either EQ-5D-5L health utility score at a following visit or at death, with EQ-5D-5L health utility score at death assumed to be 0. The area under the curve was then determined as time from entry to 6-month visit × (EQ-5D-5L health utility score at entry + EQ-5D-5L health utility score at 6-month visit) ÷ 2 + time from 6-month visit to 6 months × (EQ-5D-5L health utility score at 6-month visit + EQ-5D-5L health utility score at 6 months) ÷ 2. If there was no information after 6-month visit, EQ-5D-5L health utility score at 6 months was assumed to be the same as EQ-5D-5L health utility score at 6-month visit.
Estimated between-group differences in cost and QALYs were obtained by the method of recycled prediction in 5000 bootstrap samples. The distribution of these quantities was summarised and presented graphically in the incremental cost-effectiveness plane.
To assess whether or not the effect of treatment on QALYs differed between subgroups by age, sex and diabetes, interaction effects of these variables with treatment were added to the regression model.
Results
Table 9 shows means and CIs for costs from different sources under different scenarios for staff costs. Estimates shown assume that staff are on average pay for their pay band. Average total costs per patient over 6 months range from £231.70 (95% CI £204.30 to £259.10) to £423.90 (95% CI £373.70 to £474.10), depending on location and staff–patient ratio. The main cost factor is the staff cost for delivering the exercise sessions.
| Variable | Cost (£) per patient, mean (95% CI) | |||
|---|---|---|---|---|
| Outside London | London | |||
| Low staff–patient ratio | High staff–patient ratio | Low staff–patient ratio | High staff–patient ratio | |
| Equipment purchase and maintenance | 9.13(8.05 to 10.21) | 9.13 (8.05 to 10.21) | 9.13(8.05 to 10.21) | 9.13(8.05 to 10.21) |
| Staff delivering exercise sessions | 204.34(180.15 to 228.54) | 340.57(300.25 to 380.90) | 236.73(208.70 to 264.76) | 394.55(347.84 to 441.27) |
| Training and oversight | 18.23(16.07 to 20.38) | 18.23(16.07 to 20.38) | 20.25(17.86 to 22.65) | 20.25(17.86 to 22.65) |
| Total cost per patient over 6 months | 231.7(204.3 to 259.1) | 367.9(324.4 to 411.5) | 266.1(234.6 to 297.6) | 423.9(373.7 to 474.1) |
| Estimated difference in cost(recycled predictions) | 234.1(208.5 to 260.1) | 371.7(331.0 to 414.2) | 268.9(239.9 to 299.2) | 428.3(379.5 to 475.5) |
The mean of the area under the EQ-5D-5L curve was 0.665 (SD 0.248) in the control group and 0.653 (SD 0.269) in the intervention group. The mean difference between the treatment group and the intervention group, obtained using the method of recycled predictions, was –0.012, indicating lower QoL in the intervention group than in the control group, although the 95% CI (–0.069 to 0.043) included no difference. No significant subgroup effects were found for age, sex or diabetes at baseline.
Figure 2 shows the estimated differences in cost and QALYs on the ICER plane for the low staff–patient ratio outside London as an example.
FIGURE 2 Cost-effectiveness on ICER plane, 5000 bootstrap samples, low staff–patient ratio and outside London. Reproduced with permission from Greenwood et al.1 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original figure. (Figure not reproduced here; see the original article.)
Discussion
The economics results suggest that, over 6 months of implementation, the PEDAL intervention resulted in higher NHS costs with no gain in QoL measured with the EQ-5D-5L. A number of factors should be borne in mind; for example, the benefit from the PEDAL intervention could be avoiding cardiovascular events and other illnesses in future years, and the 6-month time horizon would not capture this. The PEDAL protocol could be refined to target different patients on dialysis and/or to include a time point at which the benefit of the intervention was assessed, with only those reporting a good QoL gain continuing; both of these could potentially improve the cost-effectiveness.