Resource use analysis

An analysis of the NHS resources consumed by participants in each of the trial arms indicated that participants in the intervention arms had greater increases in NHS resource consumption at 9 months compared with baseline than participants in the control arm, as indicated in Table 10.

TABLE 10

Differences-in-differences NHS resource use estimates

Cost of the intervention

Estimates of the cost of the intervention were arrived at taking into account a range of factors including number of sessions per completer; number of completer sessions per RA; numbers of completers at 3 and 9 months in the mini and full booster intervention arms; numbers of participants in the mini and full booster arms; estimates of the duration of sessions in the mini and full booster arms; RA training and monitoring costs; venue hire (full booster) and telephone call (mini booster) costs; and the ratio of participants who completed a full or mini booster course to those assigned to the course.

Following discussion with RAs, the average duration of a mini booster session was estimated to be 20 minutes and the average duration of a full booster session was estimated to be 30 minutes. The cost per minute of a mini booster session was estimated to be lower than the cost per minute of a full booster session as the latter required venue hire and RA travel time. Rates of attrition were similar in both intervention arms at 3 months but were higher in the mini booster arm at 9 months. Because of this the average total cost per completer (someone who attended two booster sessions and provided valid Actiheart data) was estimated to be slightly higher in the mini booster arm at 9 months than in the full booster arm at 9 months. Because of the high level of uncertainty in estimating the cost of the intervention, eight different intervention cost estimates were produced for both the mini booster arm and the full booster arm using different plausible assumptions. The assumptions made to produce these estimates are shown in Appendix 7, Table 62; 2011 prices were assumed.

The range of estimates appeared to follow a log-normal distribution with a mean (SD) cost per completer of £216 (£88) for the mini booster and £205 (£76) for the full booster. Because of the similarity of these numbers the booster intervention was assumed to cost approximately the same irrespective of whether it was the full booster or the mini booster. The two pairs of eight estimates were combined to produce a mean (SD) cost per completer of £211 (£91), also assuming a log-normal distribution. Within the deterministic analyses the mean estimate of £211 was used. Within PSA 1000 values were drawn from an appropriately parameterised log-normal distribution.

Health-related quality of life data used

The trial outcome data used to populate the short-term model are patient-estimated utilities based on SF-12v2 plus 4 questionnaire data recorded at baseline and at the end of the trial. These data are presented in Figure 30 for participants who provided valid 9-month Actiheart data (‘completers’).

FIGURE 30

Estimated utility scores reported at 3 and 9 months by allocation group (trial completers only).

Within scatter plots of this form, an overall trend over time would be indicated by the scatter deviating from the diagonal line: above the line in the case of an upwards trend and below the line in the case of a downwards trend. A difference between groups would be apparent if the three sets of scatter, indicated by the three plotting symbols, tended to cluster in different places. It is seen from this plot that there does not appear to be either a trend over time or a difference between groups.

Table 11 shows the mean (SD) change in HRQoL from baseline to 9 months by allocation group. Differences-in-differences estimates, comparing the control group with either the mini booster group or the full booster group, are also presented. These differences are very close to zero in all cases but very slightly favour the control group over either of the intervention groups.

TABLE 11

Mean (SD) change in HRQoL scores from 3 to 9 months

This approach was adopted for each of the analyses.

Simple threshold analysis

Given an estimated cost of the intervention of £211 per participant, and assuming a willingness-to-pay threshold of £20,000 per QALY, the intervention would have to provide an additional 0.01055 QALYs to be considered cost-effective.

Short-term scenarios

Figures 31–42 present scatter plots, CEAFs and tables summarising mean scores and ICERs for each of the 12 scenarios described in Table 9.

FIGURE 31

Scenario 1. (a) Scatter plot of difference in costs (£) against differences in QALYs; (b) CEAF; and (c) mean costs, QALYs (to two decimal places) and ICERs.

FIGURE 42

Scenario 12. (a) Scatter plot of difference in costs (£) against differences in QALYs; (b) CEAF; and (c) mean costs, QALYs (to two decimal places) and ICERs.

FIGURE 32

Scenario 2. (a) Scatter plot of difference in costs (£) against differences in QALYs; (b) CEAF; and (c) mean costs, QALYs (to two decimal places) and ICERs.

FIGURE 33

Scenario 3. (a) Scatter plot of difference in costs (£) against differences in QALYs; (b) CEAF; and (c) mean costs, QALYs (to two decimal places) and ICERs.

FIGURE 34

Scenario 4. (a) Scatter plot of difference in costs (£) against differences in QALYs; (b) CEAF; and (c) mean costs, QALYs (to two decimal places) and ICERs.

FIGURE 35

Scenario 5. (a) Scatter plot of difference in costs (£) against differences in QALYs; (b) CEAF; and (c) mean costs, QALYs (to two decimal places) and ICERs.

FIGURE 36

Scenario 6. (a) Scatter plot of difference in costs (£) against differences in QALYs; (b) CEAF; and (c) mean costs, QALYs (to two decimal places) and ICERs.

FIGURE 37

Scenario 7. (a) Scatter plot of difference in costs (£) against differences in QALYs; (b) CEAF; and (c) mean costs, QALYs (to two decimal places) and ICERs.

FIGURE 38

Scenario 8. (a) Scatter plot of difference in costs (£) against differences in QALYs; (b) CEAF; and (c) mean costs, QALYs (to two decimal places) and ICERs.

FIGURE 39

Scenario 9. (a) Scatter plot of difference in costs (£) against differences in QALYs; (b) CEAF; and (c) mean costs, QALYs (to two decimal places) and ICERs.

FIGURE 40

Scenario 10. (a) Scatter plot of difference in costs (£) against differences in QALYs; (b) CEAF; and (c) mean costs, QALYs (to two decimal places) and ICERs.

FIGURE 41

Scenario 11. (a) Scatter plot of difference in costs (£) against differences in QALYs; (b) CEAF; and (c) mean costs, QALYs (to two decimal places) and ICERs.

Summary of the short-term model results

The results of the separate scenarios for the short-term, questionnaire-based cost-effectiveness model suggest that the estimated cost-effectiveness of the intervention is subject to a high degree of structural uncertainty and depends on factors such as which data are used and which time periods are compared. In the majority of the scenarios, the control intervention, no booster, appears both the optimal choice and the option with the highest probability of cost-effectiveness at all willingness-to-pay thresholds considered. However, if the assumptions involved in scenarios 5 and 6 are made, there is an indication that the full booster may be the optimal choice when assuming a willingness-to-pay threshold of ≥ £20,000.

Looking at the estimated mean costs and mean QALYs for each arm in each scenario, it is apparent that the differences in costs and QALYs are marginal for all scenarios. For QALYs, the differences in mean values between arms are typically around or < 0.01 QALYs. The cost of the booster intervention is just one of a number of costs to the NHS and society that the participant population incurred and, compared with the costs of other NHS resources accessed by the participants over the trial period, it is not large. As the mean costs and QALYs observed in all arms are very similar, and the ICER is a ratio of two numbers, even slight decreases in costs or increases in QALYs could lead to very different indications of the cost-effectiveness of either intervention and so all results presented are potentially very dependent on statistical noise.

As previously discussed, the approaches taken to estimate the causal effect of the intervention on mean resource use and mean utility should be considered cautiously. This is partly because of the small sample sizes involved but mainly because the population considered does not as a rule suffer from any particular NHS resource-consuming and quality of life-reducing disease that the intervention is specifically designed to treat. Because of this, changes in HRQoL or NHS resource consumption over the relatively short time horizon of the trial are unlikely to be directly related to the effect of the intervention. As the intervention is more preventative than curative, the effect of the intervention on NHS resource consumption and HRQoL is likely to be relatively indirect, mediated by the effect of increased physical activity on lifelong morbidity and mortality risks, and to operate over a much longer time horizon. The long-term model results presented in the following section attempt to take these factors into account.

Individual sampling model results

In the long-term model, 9-month and 3-month mortality RRs associated with individuals within each intervention arm are based on TEE scores sampled directly from individual participant trial data. Because of the power law mapping equation used to associate individual TEE scores with mortality RRs, and slight differences in the age and gender distribution of the trial arms, the mean incremental effects of the intervention on HRQoL may differ from the effects on TEE. This is particularly likely to be the case when the distribution of baseline levels of physical activity differs by group, as the mapping equation assumes that the most sedentary individuals have much higher mortality RRs than slightly less sedentary individuals.

As discussed earlier, a number of separate primary scenarios were considered to investigate the impact of structural uncertainty on the model results. These are shown in Table 12, which shows mean incremental life-years and QALYs in each arm, and Table 13, which shows, to two decimal places, estimated incremental differences in effectiveness in the intervention arms compared with the control arm in each of the scenarios.

TABLE 12

Summary of mean additional life-years and QALYs within different arms and scenarios

TABLE 13

Summary of estimates of effectiveness

As Table 12 indicates, all estimates for the number of additional life-years lived and QALYs accumulated are very similar, with all values differing by < 1 life-year and < 0.5 QALYs. As Table 13 indicates, this in turn leads to very small estimates for incremental differences, of < 1 life-year and < 0.33 of a QALY. Because of the stochastic qualities of the models, and the effect of initial ages on QALYs and annual QALY increments, the direction of the incremental differences in life-years and QALYs is different for some of the scenarios, highlighting how marginal the differences between the trial arms were and the resulting influence on random variation of the estimated results.

Secondary analyses

In addition to the main long-term model results, two supplementary analyses were also conducted. In the first series of analyses, the levels of physical activity of all participants were set to fixed values. These fixed values were varied from the first (lowest) quintile to the fourth quintile observed in the trial. The mean additional utility that resulted from shifting up by one quintile was estimated for each of these baseline levels of physical activity so that the relationship between additional physical activity and baseline activity could be explored. The second series of analyses adopted a similar approach but used a level of physical activity gain based on the mean differences-in-differences estimates for TEE in the full booster group compared with the control group.

Scenarios assuming gains of one quintile

Given the power law relationship that appears to exist linking physical activity with mortality RRs, it is important to consider the effect of the baseline level of physical activity in estimating the cost-effectiveness of an intervention. Given that the least physical active quintile has the highest mortality risk, it can be assumed that a given improvement in physical activity is likely to be disproportionately effective in terms of reduced mortality in this population compared with less sedentary baseline populations. Within the scenario analysis it was assumed for simplicity that the intervention led to an improvement in physical activity of one quintile for 2 years. The effect of this temporary increase in physical activity was modelled when assuming that the entire population was initially in the first (most sedentary) quintile, the second quintile, the third quintile and then the fourth quintile. The results of this analysis are shown in Table 14 and Figure 43. The maximum acceptable intervention cost for each of these quintiles is presented, assuming a standard willingness-to-pay threshold of £20,000 per QALY.

TABLE 14

Shift in physical activity quintile

FIGURE 43

Shift in physical activity quintile.

‘Value-added’ model

Analyses of available data comparing mean daily TEE levels at 3 months and 9 months, and in the control arm, mini booster arm and full booster arm, suggest that those in the control arm used on average 66.64 kcal less per day at the end of the trial than at 3 months. In comparison, those in the mini booster arm used 36.37 kcal less per day at the end of the trial than at 3 months and those in the full booster arm used 8.85 kcal less at the end of the trial than at 3 months. This indicates a difference of 30.27 kcal favouring the mini booster over the control and a difference of 57.78 kcal favouring the full booster over the control. These differences are very small but positive. Because of the non-linear relationship between TEE gain and baseline TEE level, the main economic model, which used individual-level data from all participants, produced utility estimates that very slightly favour the control arm over either of the booster interventions. Within this additional series of analyses, the implications of assuming that the mean relative gains of the intervention groups relative to the control group were applied equally to all participants was explored. To do this it was noted that 30.27 kcal is equal to 1.36% of the 3-month median TEE score and 57.78 kcal is equal to 2.53% of the 3-month median TEE score. Just as in the previous series of analyses an intervention was assumed to lead to a 20 percentage point increase in activity for 2 years relative to no intervention (i.e. shift activity levels from the first to the second quintile, the second to the third quintile, and so on), so in this series of analyses the full booster was assumed to result in a 2.53 percentage point increase and the mini booster was assumed to result in a 1.36 percentage point increase.

With these assumptions it appears that the full booster may be cost-effective, assuming a willingness-to-pay threshold of £20,000 per QALY, if the intervention costs < £332 per participant (95% credible interval dominated to £725 per participant), as shown in Table 15. However, the clinical differences between the control arm and the mini booster arm appear so marginal that the mini booster and control arms appear largely indistinguishable in terms of QALYs and so the central estimate suggests that the mini booster is ruled out by simple dominance compared with the control (although the 95% credible intervals of the maximum acceptable intervention cost vary from not acceptable/dominated to £299 per participant).

TABLE 15

Scenario analyses