Economic evaluation

The economic evaluation was from an NHS perspective as other significant care-related contacts outside the NHS were not anticipated. Private expenditure was incorporated within a sensitivity analysis that adopts a societal perspective. The evaluation uses both a within-trial time frame and a modelled patient-lifetime time frame; the modelled analysis was the primary focus of the economic evaluation. Methods are in line with those of NICE75 and have been previously described in the trial protocol1 and the study Health Economics Analysis Plan (HEAP; see Report Supplementary Material 1); deviations from these documents are summarised in the Discussion.

Methods

Resource use

Participant-level data were collected for the trial interventions [urodynamics plus CCA (referred to as the urodynamics arm) and CCA only (referred to as the CCA only arm)], plus subsequent treatments, investigations and other health service contacts. Data were collected on CRFs at 6 and 15 months post randomisation via a review of participant medical records. In line with the clinical trial, a 24-month follow-up was undertaken for participants who had their treatment delayed due to the COVID-19 pandemic. Other medical care and primary care contacts were collected via participant questionnaires. Provision of incontinence pads by the NHS is not universal across the UK, but, for the purposes of the analysis, we made the simplifying assumption that they were an NHS cost. All items of resource use are shown in Table 13.

Costs to participants of undergoing treatment and personal expenditure on products relating to their OAB symptoms were captured by a questionnaire at baseline, 6, 15 and 24 months, where appropriate. Likewise, time taken away from work due to treatment of symptoms and reduced productivity at work due to symptoms were also recorded on the participant questionnaires.

A full list of resource use items is given in Table 13.

Unit costs

The principal sources of unit costs were NHS Reference Costs, the British National Formulary (BNF) and Unit Costs of Health and Social Care. All costs are at 2020–1 price levels. A full list of unit costs is given in Table 13, with further details in Appendix 2 (Table 31). Costs and outcomes were discounted at 3.5% as recommended by NICE.75

A separate unit cost of AEs was not applied as these were identified and costed individually using the data and unit costs described above. So, for example, days in hospital, GP attendance and antibiotic prescriptions associated with AEs are included within the relevant categories of cost.

Quality-adjusted life-years

Health-related utility was assessed using the EQ-5D-5L measure at baseline, 6, 15 and 24 months post randomisation, where appropriate. QALYs were estimated using the EQ-5D-5L van Hout ‘cross-walk’ tariff as recommended by NICE at the time of the trial’s conduct76 using linear interpolation between the scheduled time points.

An exploratory analysis was undertaken to assess the QALY loss related to urodynamics (e.g. anxiety and discomfort) by estimating the degree to which EQ-5D-5L values at 6 months post randomisation were affected by time since urodynamic testing. If a robust estimate of QALY loss was produced, the impact of its incorporation into the cost-effectiveness analysis was examined using a sensitivity analysis. This analysis took the form of a regression using the difference between post-urodynamics and pre-urodynamics utility as the dependent variable, and time since urodynamics, its squared term, and other covariates as the independent variables. The estimated disutility associated with urodynamics was to be used as an adjustment to the QALYs of all participants undergoing urodynamics if p < 0.1 on either of the two time covariates.

Analysis

An incremental analysis was undertaken, together with plots on the cost-effectiveness plane (CEP) reflecting second-order uncertainty, and their associated cost-effectiveness acceptability curves (CEACs). Cost-effectiveness was assessed in relation to a threshold of £20,000 per QALY gained.

The following subgroup analyses, aligned with the clinical analysis, were planned:

• a comparison between participants with baseline diagnosis of OAB and urgency-predominant MUI
• a comparison between the clinical effectiveness of the different treatment pathways of those who started on BoNT-A and those who started with SNM treatment
• a comparison of the effectiveness of (1) SNM and (2) BoNT-A according to CCA only compared to treatment which was guided by urodynamics.

Deterministic sensitivity analyses were undertaken in relation to:

• complete case analysis
• societal perspective that includes patient costs and production losses
• utility loss relating to the use of urodynamics as estimated by exploratory analysis
• alternative, unpublished, cost for urodynamics taken from a UK study (personal communication)
• use of a £30,000 cost-effectiveness threshold
• use of an alternative utility tariff for the EQ-5D-5L76,77
• the inclusion of additional predictors within the multiple imputation.

Two frameworks were applied to provide complementary sets of analysis: firstly, a within-trial framework describing the trial results up to the 24-month follow-up; secondly, a decision-analytic model framework that supplemented the trial data with external evidence that captures clinical and patient events beyond the end of the trial.

The within-trial analysis followed the best-practice guidelines.78 The analysis calculated total costs and QALYs for each participant and estimated the incremental costs and QALYs using a regression model with the following baseline covariates; OAB dummy, age, age squared, follow-up time, follow-up time squared, number of deliveries, urgency perception dummies, CentreNo dummies and the QALY regression additionally included baseline utility score. Missing data were imputed using age, OAB dummy, 15-month dummy, number of deliveries and urgency perception as predictors.79,80 Costs and QALYs were calculated up to 24 months, with a dummy variable being added with a value of ‘1’ if the participant was not followed up to 24 months, and ‘0’ otherwise. As such, the estimated increment is assumed the same for all participants, with the coefficient on the dummy variable estimating the shortfall in costs associated with a shorter follow-up. A seemingly unrelated regression was planned, but other specifications that were better suited to the distribution of the data were also explored.78

The choice of regression model was based on an assessment of the distributions for the cost and QALY data and the appropriateness of the distribution family within a GLM specification. In the event of normally distributed data a seemingly unrelated regression was planned, but, if the data were skew, the most appropriate distribution family within a GLM would be assessed using the modified Park test.81

The data are assumed to be missing at random; that is, that the probability that data are missing is independent of unobserved values, given the observed data. The missingness is non-monotonic because those with missingness at one time point may subsequently return data at a later time point, therefore inverse probability weighting is not an appropriate method for dealing with missing data in this context. There are no statistically significant differences in missingness between arms of the trial. Based on the logistic regressions, the missingness is correlated with age, follow-up time, number of deliveries and previous therapies for both total costs and total QALYs.

The decision-analytic approach represents the primary analysis based on its recognised advantages over a purely trial-based approach.82 Of particular importance in the context of this study is the need to consider patient costs and outcomes beyond the end of the trial. Differences in the choice of initial therapy post randomisation could generate different success rates, retreatment rates, costs and patient outcomes. Such work inevitably requires the use of external data and some assumptions; however, the uncertainty that these generate will be explored in the analysis.

The structure of the model beyond 24 months was developed after a review of economic evaluations relating to urodynamics and/or treatments for OAB or SUI. A model structure was then conceptualised that was based around the NHS treatment pathways used within the trial design (see www.fundingawards.nihr.ac.uk/award/15/150/05), but also capturing the important features of clinical management and patient outcomes identified in the review. The model structure, a hybrid model with a decision tree describing short-term events and Markov processes describing long-term events, is shown in Figure 7. The conceptual modelling that led to the adoption of this model structure is given in Appendix 3.

FIGURE 7 Model structure. UDS, urodynamics. The nodes immediately to the right of ‘UDS’ and ‘No UDS’ are Markov nodes, with descriptors on the far right representing transitions to the relevant Markov node. Death is included in the model but is excluded from this figure for simplicity. (Figure not reproduced here; see the original article.)

In short, the model takes the estimated total costs and QALYs at 24 months directly from FUTURE trial, then applies a Markov process; participants with successful treatment remain on that treatment, with treatment failures moving to ‘other care’. The proportion of participants receiving BoNT-A or SNM, or having received surgery for SUI, or receiving ‘other care’ at 24 months is taken from FUTURE. The costs for ongoing BoNT-A, SNM replacement and SNM removals are taken from the table of unit costs (see Table 13), while the cost for ‘other care’ is estimated from FUTURE (using the observed cost of participants who have not received any of the alternative treatments). Events beyond that time point incorporated evidence drawn from a review of existing models. These reviews and the associated parameters are given in Appendix 3.

The model includes all-cause mortality using the UK lifetables and discounts costs and QALYs at 3.5% per annum. Utilities are not reduced to account for comorbidities associated with ageing. A half-cycle correction is applied.

A value-of-information analysis was planned in order to identify those parameters where there was greatest value in resolving outstanding uncertainty. This was to use the Sheffield Accelerated Value of Information tool, an annual patient population of 77,000 women,25,48 and a 10-year time horizon.83

Results

Costs

A description of the resources used by participants across the full follow-up period (which can be 15 or 24 months), without imputation, is given in Table 14. This shows that not all participants randomised to urodynamics received the intervention, while a small number of participants who were randomised to CCA only received urodynamics, although overall participants in the intervention arm received more urodynamic testing. The urodynamics arm also received more clinic visits (0.70 vs. 0.45) and more procedures relating to SUI (0.03 vs. 0.01). All other differences are not statistically significant.

When combined with the unit costs from Table 13 and Appendix 2, the distribution of costs across participants is highly skewed, with the vast majority having costs below £5000, but with smaller numbers having costs stretching up to around £20,000 (Figure 8). The disaggregated costs mirror the resource-use findings, with statistically significant differences limited to urodynamic visits, clinic visits and procedures related to SUI (Table 15).

FIGURE 8 Histogram of observed total costs by arm over full follow-up (15- and 24-month participants). (Figure not reproduced here; see the original article.)

Based on the distribution of costs, a GLM was adopted, with the modified Park test indicating that a gamma family and identity link was most appropriate to model total costs and estimate the differences (increment) between the two trial arms. Analysis was undertaken for all participants, assuming the same increment across all participants regardless of length of follow-up but presented in terms of estimated 24-month costs for consistency. For the primary analysis, which takes an NHS perspective and incorporating multiple imputation, the urodynamics arm has mean costs that were £463 higher (95% CI £48 to £877), as shown in Table 16.

Outcomes

Mean utilities for participants across the full follow-up period (which can be 15 or 24 months post randomisation), without imputation, are shown in Figure 9 and Table 17. Mean utilities are similar for all time points except for those collected at the 24-month visit, at which there is a difference of 0.078, although there are no statistically significant differences at any individual time point. When modelled with imputation, a small but not statistically significant difference in QALYs of 0.011 (95% CI −0.044 to 0.065) is estimated in favour of the urodynamics group (Table 18).

FIGURE 9 Histogram of observed QALYs over full follow-up (15- and 24-month participants), by arm. (Figure not reproduced here; see the original article.)

Cost-effectiveness

Based on the estimated incremental costs and QALYs of urodynamics (£463 and 0.011, respectively), the ICER is £42,643 per QALY gained. The CEP in Figure 10 has a red dot, the central estimate of incremental costs and QALYs from the analysis; urodynamics produces higher costs and higher QALYs. The ellipses represent the joint uncertainty around these estimates; there is a 50% chance that the true estimate sits within the dashed ellipse, for instance. If we want to be 95% certain of the area in which the true estimates sit, we have to accept the possibility that urodynamics could lie in any of the four quadrants of the plane (i.e. higher or lower costs, together with higher or lower QALYs).

FIGURE 10 Cost-effectiveness plane for primary analysis, urodynamics vs. CCA. (Figure not reproduced here; see the original article.)

A CEAC can be derived from those same data which attaches a probability to our estimate being below any given ICER. Shown in Figure 11, the CEAC indicates that urodynamics has a 34% probability of being cost-effective at £20,000 per QALY gained (see Figure 11).

FIGURE 11 Cost-effectiveness acceptability curve (CEAC). (Figure not reproduced here; see the original article.)

Subgroup analysis

When examining participants with a pre-randomisation clinical diagnosis of OAB and urgency-predominant MUI separately, the two groups of participants are found to have similar costs. However, participants with a diagnosis of urgency-predominant MUI have a notably greater gain in QALYs associated with urodynamics (0.053 vs. −0.010). This leads to a lower ICER than the full trial population (£8357 per QALY gained) and a commensurately higher probability of being cost-effective at £20,000 per QALY (72%), as shown in Table 19. For participants in the OAB subgroup, urodynamics is dominated by the CCA only group (with higher mean costs and lower mean QALYs).

A relatively small number of participants received PTNS (47 participants), surgery for SUI (21 participants) or SNM (19 participants); hence, a clinical and cost-effectiveness comparison of outcomes would be non-meaningful and was therefore not undertaken.

Sensitivity analysis

When a complete-case analysis is undertaken, the estimated incremental costs reduce to £385 and the incremental QALYs increase to 0.029 (Table 20). Neither of these differences is statistically significant (see Table 18). This leads to a lower ICER than the primary analysis of £13,281 and a higher probability of being cost-effective of 67%.

When a societal perspective is taken, there are very few changes from the primary analysis, with an ICER of £33,528 per QALY gained, associated with a probability of being cost-effective at £20,000 of 40%. When a cost-effectiveness threshold of £30,000 per QALY is used, the baseline analysis indicates that urodynamics has a 45% chance of being cost-effective.

An analysis to estimate the QALY loss related to urodynamics showed no difference in pre- and post-urodynamic EQ-5D-5L scores (see Appendix 4). Consequently, the QALYs for participants receiving urodynamics were not adjusted to account for any HRQoL impact of testing.

Modelling

Using the results of the within-trial analysis as a starting point, combined with data on the women’s treatment and utilities at 24 months, the model was parametrised (see Appendix 3, Table 40). The projected movement of women between treatments beyond the end of the trial is shown in the Markov traces in Figure 12. The four lines show the proportion of women, by year, who are assigned the costs and QALYs relating to BoNT-A, SNM, SUI, other treatments or death. This shows women moving from BoNT-A and SNM onto ‘other’ treatment for the first 5 years after the trial (see Figure 12). Beyond that point, the increasing mortality seen with ageing in the general population becomes the dominant factor (noting that the starting age for the modelled cohort, which is based on FUTURE, is 60 years). This is a natural consequence of the model structure, which directs all women to ‘other care’ after failure of the treatment that they are receiving after 2 years (as shown in Figure 7). The Markov traces in Figure 12 also highlight an important feature of the evidence used to extrapolate the results of the trial, namely, that those women who have successful treatment for 5 years continue treatment indefinitely; this is responsible for the ‘kink’ in the BoNT-A traces. Figure 12 also highlights two important results from the trial, which are: (1) urodynamics is associated with a lower rate of BoNT-A use at 24 months (which translates to the start of the Markov model) and (2) the rates of SNM and surgery for SUI at 24 months are very low.

FIGURE 12 Markov traces for the two patient groups. (a) CCA. (b) Urodynamics plus CCA. (Figure not reproduced here; see the original article.)

When costs and QALYs are calculated, the lifetime analysis shows that the urodynamics group remains more costly but is now also associated with fewer QALYs than the CCA only group. As such, urodynamics is dominated by the alternative, with only a 23.4% chance of being cost-effective (Table 21, Figure 13). This is a consequence of the urodynamics arm producing a lower proportion of participants on BoNT-A (which has a higher mean utility than the alternatives) and consequently a higher proportion of participants on ‘other treatments’ (which has a high cost and a lower mean utility).

FIGURE 13 Cost-effectiveness acceptability curve for the two participant groups. (Figure not reproduced here; see the original article.)

When the MUI subgroup is modelled over the lifetime of women, an alternative parameterisation is adopted using the within-trial and 24-month treatments/utilities for the sub-group (see Appendix 3, Table 40). Three points to note with this parameterisation are that, first, when entering the model at 24 months, women have higher QALYs in the urodynamics group (as already shown). Second, the distribution of women across treatments at 24 months is similar to that in the primary analysis (albeit with slightly fewer women receiving BoNT-A in the urodynamics group than before). Third, the utility of women receiving ‘other treatments’ is higher than that of women receiving BoNT-A, although this is not a statistically significant difference.

As a consequence of the second point above, the Markov traces for the MUI subgroup are similar to those of the primary analysis. As a consequence of the third point, the urodynamics group gains more QALYs than the CCA group, albeit at a higher cost. The ICER for urodynamics is £26,462 (Table 22).

The CEAC associated with this analysis is shown in Figure 14. It shows the probabilities of urodynamics being cost-effective at £20,000 and £30,000 per QALY gained as 45.3% and 53.8%, respectively.

FIGURE 14 Cost-effectiveness acceptability curves for the MUI subgroup. (Figure not reproduced here; see the original article.)

Discussion

The primary, model-based economic analysis shows that urodynamics has a low probability of being cost-effective at £20,000 per QALY gained (23.4%), producing modestly higher costs (£1380) and slightly lower QALYs (−0.002) per patient. There are no major differences in the pattern or costs of subsequent treatments. In the subgroup of women who had a preliminary diagnosis of MUI prior to randomisation, urodynamics generates an ICER of £26,462 and a higher probability of being cost-effective (45.3%), which increases to 53.8% when a threshold of £30,000 is used.

The ICER from the trial analysis suggests that urodynamics is not cost-effective due to a modest incremental cost and a small and uncertain gain in QALYs. The sensitivity analysis shows that when a complete-case analysis is undertaken, urodynamics appears cost-effective. The reasons for this are unclear, as a sensitivity analysis that used additional predictors had little effect on the results. The impact of the pandemic on missing data, treatment patterns and, by implication, costs and QALYs could be a factor. However, initial explorations of this did not reveal any clear systematic impact. Likewise, urodynamics appears cost-effective in participants with an initial diagnosis of urgency-predominant MUI, principally due to the higher QALY gains in that group compared to participants with an initial diagnosis of OAB.

There are several issues that require further consideration when interpreting these results. First, when assessing different model specifications, it was noted that the estimated incremental costs and QALYs were sensitive to the choice of covariates and predictors for imputation. The variables used within the analysis are aligned with those used in the clinical analysis in order to produce results that are consistent across the two parts of the evaluation. Alternative specifications can alter the results and conclusions; however, the relative validity of those results is unknown.

Second, the incorporation of the 24-month data into the overall analysis for a subset of participants needed to be accounted for within the statistical model used to estimate the incremental effects. The chosen pre-specified analysis was to add a dummy variable relating to not receiving a 24-month questionnaire; this assumes that the incremental costs and QALYs between the treatments are the same at 15 and 24 months. We have not formally tested this assumption, or explored alternative model specifications that could be adopted if the validity of that assumption was called into question. While including a time × increment interaction in the cost and QALY regressions is straightforward, we need to be wary of potential confounding effects (e.g. if delays were not consistent across recruiting sites or if they inadvertently led to prioritisation of treatment for some participants based on clinical history or prognosis). Consequently, as well as modelling a simple interaction, we may want to consider factoring in site and participant effects that help to explain how incremental costs and QALYs change between 15 and 24 months.84

Third, while there is good evidence that the EQ-5D-5L is sensitive to changes in severe OAB, it performs less well in milder cases.84 As such, it may underestimate the benefits of treatments. An alternative approach would be to use a condition-specific utility measure, such as the OAB-5D,84 but such measures are not widely accepted due to issues of validity and comparability.85 Likewise, the timing of data collection may not coincide with AEs, and so the full effect of these may not be fully captured. However, the weaknesses relating to our measurement of utility are not considered important given the clinical and economic findings; more specifically, the decision uncertainty is low.

Fourth, the perspective on costs in the primary analysis is the NHS perspective. While NICE requires an NHS and Personal Social Services perspective, the literature suggests that publicly funded care outside the NHS is extremely limited.86,87 Private expenditure is much more significant,86,87 and we have attempted to capture this within our societal perspective by the inclusion of patient-reported private expenditures.

The modelling of a lifetime horizon suggests that the urodynamics group is not cost-effective. This result is not unexpected given the trial results, the distribution of participants across treatments at 24 months and the structure/parameterisation of the model. The modelling makes urodynamics even less cost-effective due to fewer participants receiving BoNT-A (which has a high utility) and more women receiving ‘other care’ (which is costly).

There are several aspects of this analysis that need further consideration: firstly, the appropriateness of the model structure. While the chosen model structure is consistent with other previous studies, it is recognised that it is relatively simple, with no further lines of therapy beyond those observed at 24 months (except for transitions to ‘other care’). Other lines of therapy could be added to all four of the health states shown in Figure 7; however, the evidence base for the effectiveness and HRQoL of women following these treatments after previous treatments for refractory OAB symptoms is weak (as evidenced by the need for this trial).

Second, the parameterisation imposes an implicit assumption that all treatments beyond 24 months are equally effective, regardless of how women came about receiving them; success rates and HRQoL are the same for both groups of women. However, it is possible that the lower proportion of women receiving BoNT-A in the urodynamics arm could be due to the treatment being more appropriately targeted at women, and as such it could be that participants receiving BoNT-A after a urodynamic assessment have higher success rates and HRQoL. Arm-specific utilities could be used in the model, although sample sizes become prohibitively small for some treatments. Robust estimates of ‘arm-specific’ long-term success rates are not thought to be available.

Third, there is an inconsistency in the way that costs are estimated for the four health states shown in Figure 7, which may produce a bias against urodynamics. The costs for ongoing successful treatment are based solely on the treatments: the drug and visits for BoNT-A, revision and removal for SNM, and zero ongoing cost after SUI surgery. For women who are in the ‘other care’ health state (either after treatment failure or not having any treatment), their costs are based on those observed in FUTURE, which includes all care costs, including follow-up visits, medications and catheterisations. This inconsistency is due to a difficulty in identifying these ongoing costs in participants who are undergoing successful treatment in FUTURE; if this could be done, we could add in these costs to the other treatment options. Our inability to do this creates a potential bias against urodynamics because women transition to ‘other care’ (and its high costs) more quickly due to the lower proportion of women who receive BoNT-A in that group.

Fourth, the model is constrained by the evidence base on which it is built, which is heavily reliant on patient cohorts that do not match the focus of FUTURE. It should also be noted that our review of models and parameters was not systematic; however, the sources for our long-term treatment effectiveness parameters were reviewed by clinical experts. Consequently, it seems unlikely that we have missed any large, high-quality studies in our reviews. We feel that the greatest uncertainty over the validity of the available long-term estimates is in differences in patient characteristics within the cohorts, which will be brought about by differences in the assessment of refractory OAB and differences in the treatment pathways that lead to that diagnosis. Differences in the assessment of treatment success between studies/countries are also expected to have a significant impact on results and, as such, the validity of the estimates we have used.

An exploratory analysis was undertaken to examine whether there was any evidence of a QALY loss related to participants undergoing urodynamics. We examined this by looking at the change in EQ-5D-5L between pre- and post-urodynamic trial visits. Due to the timing of urodynamics and trial visits, the post-urodynamic EQ-5D-5L responses were rarely within 1 week of the intervention. We tried to take account of this by using two time-based covariates, but it is likely that the study results are dominated by observations that sit well outside the time frame during which any HRQoL impact of urodynamics persists. Limiting the sample to participants with much shorter pre–post time differences is possible, but the choice of that time period is subjective. Any identification of a disutility associated with urodynamics also requires the selection of a time period over which it extends in order to produce a QALY decrement, which is also subjective. However, the incorporation of any QALY decrement associated with the actual urodynamics procedure will necessarily increase the ICER for the urodynamics group (thereby making it less cost-effective).

Finally, the value-of-information analysis was not undertaken at this time as an extension study is under way, which will make the results based on this study obsolete. In effect, the decision about which further studies should be funded has already been made, and the next funding decision should be made after the results of that are known and incorporated into a value-of-information analysis at that time.

Conclusion

In the primary analysis, which models the results over a lifetime horizon, the cost-effectiveness of urodynamics is reduced further, such that there is only a 23.4% chance of it being cost-effective. This further deterioration is due to the lower rate of BoNT-A use at 24 months in the urodynamics arm, which leads to fewer QALYs and higher costs for other care, relative to the CCA only arm. In the subgroup of women who had a preliminary diagnosis of MUI, urodynamics generates an ICER of £26,462 with the probabilities of being cost-effective at £20,000 and £30,000 per QALY gained being 45.3% and 53.8%, respectively.

The model structure and the parameterisation of the long-term effectiveness of treatments are necessarily simplistic due to the lack of robust, long-term data relating to this patient population. We also identified a potential bias in one of the key model inputs – the annual cost of other treatments – which we could not resolve or quantify. While this is unlikely to be important for the primary analysis, given that urodynamics is not cost-effective at 24 months, it is likely to be important for the MUI subgroup, for which urodynamics is cost-effective at 24 months and is borderline cost-effective over the patient lifetime.

Within the trial, urodynamics is shown to be more costly, principally due to the intervention itself and more clinic visits. There is evidence of greater numbers of interventions for SUI in participants undergoing urodynamics, but all other effects are highly uncertain, and not statistically significant. There is no clear evidence of differences in HRQoL (as measured by the EQ-5D-5L) at any time point, nor in total QALYs. The higher mean costs and QALYs lead to urodynamics not being cost-effective at a funding threshold of £20,000 per QALY gained, with only a 34% chance of it being cost-effective. However, this is sensitive to imputation, with the complete-case analysis showing a 67% chance of urodynamics being cost-effective. The subgroup analysis suggests larger health benefits for participants with an initial diagnosis of urgency-predominant MUI, which is associated with a 72% chance of cost-effectiveness.

Further analysis of the effects that the 24-month data have on the results is warranted, as this leads to two methodological uncertainties. First, the pattern of missing data and its impact on costs and outcomes is more complex than is generally the case in RCTs. Second, the incorporation of both 15-month and 24-month data in the estimation of total costs and QALYs is based on a simple, additive specification. Alternative approaches to these issues are possible and need consideration.