Economic evaluation

Some parts of this chapter are reproduced with permission from Wiseman et al.1 Cooper et al.54 These are Open Access articles distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) licence, which permits others to distribute, remix, adapt and build upon this work, even for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The text below includes minor additions and formatting changes to the original text.

Some parts of the Methods section in this chapter are also reproduced with permission from Ahmed et al.55 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY NC 4.0) licence, which permits others to distribute, remix, adapt and build upon this work, for non-commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by-nc/4.0/. The text below includes minor additions and formatting changes to the original text.

Introduction

Economic evaluation helps to make better informed decisions to adopt new healthcare technologies by providing information on the relative efficiency of the technologies under consideration. Resources are scarce in the NHS and healthcare technologies should be adopted or retained only if they provide good value for money. Therefore, the economic analysis is interested in all relevant cost and consequences resulting from the compared care pathways. Consequently, a longer time horizon was selected compared to that used in the statistical analysis. This chapter reports on the economic evaluation of FURS as the first treatment option in comparison to ESWL for stones size ≤ 10 mm in maximum dimension (RCT1) and to PCNL for stones size > 10 and ≤ 25 mm in maximum dimension (RCT2).37 The chapter is divided into two main sections: (1) a within trial analysis that considers costs and consequences up to 12 months post randomisation and (2) a modelling extrapolation extending the analysis for a time horizon of 5 years.

Within-trial analysis

Objectives of the economic evaluation

The primary economic objective of the within-trial cost-effectiveness analysis was to assess the cost-effectiveness of FURS as the first treatment option in comparison to ESWL for stones ≤ 10 mm in maximum dimension or PCNL for stones > 10 and ≤ 25 mm in maximum dimension at 12 months post randomisation.

Methods

Established methods were followed for this economic evaluation56,57 as well as to report its findings58 with a prespecified protocol and health economics analysis plan (HEAP).

Study design and participants

The PUrE trial protocol37 and Chapter 2 provide details of the design. The within-trial economic analysis follows the ITT principle and is based on the same participants randomised and considered for the main trial analysis.

Cost and outcome assessment

Costs and outcomes were assessed via the trial CRFs and participant-completed questionnaires (see Figure 1 and Table 1). The research nurses completed a CRF at the time of surgery providing details of the operative procedures, complications and resource used in hospital. Costs of the initial intervention procedures were estimated from resource use data recorded on the CRFs coupled with routine unit cost data. Costs associated with subsequent contacts with primary care (due to kidney stones) were estimated from patient questionnaires at pre intervention, 12 weeks post intervention and 12 months post randomisation. QALYs were estimated from patients’ responses to the EQ-5D-5L at baseline, pre intervention, weekly up to 12 weeks post intervention and at 12 months post randomisation. EQ-5D-5L data were also collected pre and post intervention for additional interventions occurring outside the 12-week post-intervention window or any hospitalisation. The schedule for economic data collection is shown in Table 1.

Assessment of health service costs

NHS costs for health service use in both primary and secondary care for each trial participant were estimated. The economic evaluation seeks to inform the efficient allocation of the NHS budget; therefore, the base case analysis adopted a health service perspective. Costs are expressed in Great British pounds for the 2020–1 price year.

Cost of the primary interventions

Costs of the initial intervention were estimated from resource use data recorded in the PUrE Intervention CRF for each participant. The CRF captured date of admission, treatment and discharge, the type of procedure carried out ESWL, FURS or PCNL, procedure time (i.e. time procedure started and finished for ESWL, and time of entry into anaesthetic room, entry to and exit from the operating room, and time of exit recovery room for FURS and PCNL), intraoperative complications, and destination post operation or recovery room.

The primary costing approach assigned costs to individual components of resource use to capture patient-level variation in costs (Table 16). A number of assumptions have been made on discussion with the clinical colleagues in the project management team (three clinicians). In the UK, ESWL is conducted as a day case procedure by a nurse and a technician. The PUrE intervention CRF stated the total time of the procedure and to this, 20 minutes were added to allow for patient preparation and discharge. Details were collected on the type of lithotripter (i.e. mobile or fixed) and corresponding pro rata costs were allocated assuming throughput of 200 patients per year and a 5-year equipment depreciation. For FURS and PCNL, the time in the anaesthetic room was costed assuming a cost per hour (incorporating overheads) for a consultant anaesthetist and an anaesthetist nurse.59 For time in theatre, we defined the grade of the surgeon conducting the intervention through discussion with the clinical collaborators and costed using the unit cost per hour applied to the final total time in theatre.60 For PCNL, a radiographer was assumed for whole procedure for all cases plus 1 hour for an interventional radiologist for half of the cases. Nursing staff was costed at the requirement for general day surgery: one anaesthetic nurse, a scrub nurse, and two further theatre nurses. In addition, a published unit cost was applied for time in theatre to reflect the average cost of other staff, supplies and consumables, and allocated capital charges and overheads.61 This detailed unit cost of theatre time is only available for Scottish hospitals. Therefore, the average Scottish estimate (£596 per hour, excluding medical and nursing staff) was applied to the time in theatre for the base-case analysis. In addition to this, we applied the unit costs of major consumable items specific to the alternative procedures (see Appendix 9, Table 76).

Time in recovery following surgery was costed using the unit cost of a Band 6 nurse (inclusive of overheads) assuming one-to-one care and the time on the ward following recovery using an estimate of the cost per excess bed day following index procedure62 adjusted by inflation. Equipment costs were allocated assuming a 200-, 100- and 40-patient-per-year throughput for ESWL, FURS and PCNL, respectively (Samuel McClinton, Daron Smith and Oliver Wiseman, personal communication, 26 May 2023) and 5-year equipment depreciation (see Appendix 9, Table 76).

As an alternative approach to costing the initial procedure episodes, each patient record was mapped to the appropriate Health Care Resource Group (HRG) and costed using the relevant NHS reference cost.62 The core HRG codes LB36Z (Extracorporeal Lithotripsy) either as day case (elective with admission and discharge on the same day), elective inpatient, or non-elective short stay (if admitted as elective or emergency with overnight stay, respectively). For FURS and PCNL the HRG codes LB65E (Major Endoscopic, Kidney or Ureter Procedures, 19 years and over, with CC Score 0–2) and LB75B (PCNL with CC Score 0–1) were used, respectively. The unit costs for FURS and PCNL were applied as either a day case (patient admitted and discharged on the same day) or as an elective inpatient admission (stay ≥ 1 day).

Costs of hospital admissions, perioperative complications and hospital readmissions

The clinical management of any procedure resulting from perioperative complications leading to prolonged hospital stay as well as any hospital admissions were based on the NHS reference cost. These included events, such as urinary tract infection, haematuria, abdominal pain (Chapter 3, Complications and Chapter 4, Complications). Details for the type of complication experienced and the procedures undertaken (e.g., stent insertion, stent removal, ESWL, FURS) were obtained from the PUrE Intervention CRF, the second treatment session CRF, the follow-up CRF, the supplementary treatment CRF, and the SAEs forms. These events were costed using HRG-based reference costs.62

Costs of hospital outpatient and primary care healthcare utilisation

Secondary care outpatient visits incurred over the 12-month follow-up period were obtained from hospital records using supplementary CRFs costed using the NHS reference cost for standard surgery outpatient visit.62 Primary care visits over the follow-up period were collected from the patient questionnaires at 12 weeks post intervention and 12 months post randomisation. These contacts were costed using the Unit Costs of Health and Social Care60 by multiplying the number of visits by the appropriate unit cost.

While participants were asked about prescribed medications for their kidney stones condition, the questionnaire had no details of the actual medication prescribed. On discussion with the trial management team and due to the possible very low cost of these medications (i.e. mostly over the counter painkillers), a decision was made to exclude the medication prescribed in primary care from the costing exercise.

Indirect costs

Participants were asked about the number of sick days needed from the day of their intervention and up to 12 months post randomisation. The number of days were multiplied by the average hourly pay (gross) by age group for 2022 published by UK Office of National Statistics,63 assuming a full-time shift of 8 hours per day of work, to value the time off work. In addition, participants were asked about out-of-pocket expenditure on medications, private health care and the cost of travelling for their last hospital admission, hospital consultation and GP appointment. Participants were asked for the cost of the trip, for those travelling on public transport. Individuals using private cars were asked to estimate their trip mileage with the trip mileage valued at £0.45 per mile following the valuation of cost of using private cars for business by the UK government.64 As these costs are not costs to the NHS, these are reported separately and were not incorporated in the main economic analysis.

Outcome measures

Quality-adjusted life-years were estimated using the EQ-5D-5L questionnaire completed by participants at baseline, pre intervention, weekly up to 12 weeks post intervention, and at 12 months post randomisation. In addition, EQ-5D-5L was administered pre and post interventions or hospitalisations occurring outside the 12-week post-index intervention window (see Table 1). The methods recommended by NICE65 to map from the EQ-5D-5L to the EQ-5D-3L utility scores and based on Hernández-Alava et al. 202066 were used to assign a utility score to each participant response. QALYs were then estimated calculating the AUC approach, assuming linear change in utility between the observed follow-up time points.

Statistical analysis of trial economic data

Aggregating costs and effects

Following the ITT principle, resource use, cost and health outcome data were summed up for each participant for the 12-month follow-up period and compared by treatment allocation group. Continuous and count variables are reported as means (± SDs), and dichotomous and categorical variables as absolute numbers (and percentages).

Missing data

Unless the data are missing completely at random, reliance on complete case data for the cost-effectiveness analysis can introduce bias. The total estimated cost is the sum of numerous components over the observed follow-up period of the trial. Besides, QALYs can be computed only where participants have responded to the QoL questionnaires at every relevant follow-up point. Therefore, economic evaluations based on trial participant level data are likely to suffer missing data. We implemented multiple imputation as part of the primary analysis, using chained equations to generate multiple data sets with plausible fitted values assigned for missing cost and utility elements. The imputation was conducted at the cost category variable level (e.g., cost for anaesthetic, operating, and recovery rooms, equipment, consumables, length of stay, etc.) and the EQ-5D score level. The imputation model included all the variables in the analysis model. Rubin’s rules were used to pool estimates across the multiple imputation data sets.

Incremental cost-effectiveness analysis

The trial-based economic analysis estimated the incremental cost and QALYs between (1) FURS as the first treatment option in comparison to ESWL for stones ≤ 10 mm in maximum dimension (RCT1) and (2) FURS as the first treatment option in comparison to PCNL for stones > 10 and ≤ 25 mm in maximum dimension (RCT2) up to 12 months post randomisation. General linear regression models (GLM) with appropriate variance and link functions were fitted, with adjustment for baseline EQ-5D score. The selection of the family function was based on modified Park test, while a modified Hosmer and Lemeshow, Pearson correlation, and Pregibon link tests were used to select the link function57 (see Appendix 9, Table 77 for details). These test results suggested a Poisson family with identity link for costs and a Gaussian family with identity link for QALY for RCT1 and Gamma family with identity link for costs and Gaussian family with identity link for QALYs for RCT2. Adjusted mean values by treatment allocation group, and the incremental differences between groups were obtained using the methods of recycled predictions.57 The incremental cost-effectiveness ratio (ICER) for FURS and ESWL (RCT1) and FURS and PCNL (RCT2) were calculated as the difference in the (arithmetic) mean cost divided by the difference in (arithmetic) mean QALYs. Variance surrounding the joint incremental costs and effects was characterised using non-parametric bootstrapping. Bootstrapping involves sampling with reposition from the original trial participants a new sample of equal size of the initial sample (e.g. n = 230 and n = 231 for FURS and ESWL, respectively, for RCT1). These data are analysed in a similar way as for the original trial data and results for average cost and QALYs obtained. The process is repeated several times (i.e. 1000) and a distribution for the average cost, average QALYs, and average cost and QALY differences between trial groups are obtained. The simulation results are presented graphically using the cost-effectiveness plane (scatterplots) and cost-effectiveness acceptability curves (CEACs). The probability of FURS being cost-effective at 12 months is reported for threshold values of willingness to pay per additional QALY of £13,000, £20,000 and £30,000.67

Sensitivity analyses

Subgroup analysis was conducted to explore cost-effectiveness of FURS according to predefined subgroups: Normal/overweight/obese/extreme obesity defined according to BMI score; stone size (RCT2 only: 10–15 vs. > 15 mm); stone density (CTKUB Hounsfield units comparing ≤ 1000 vs. > 1000); skin-to-stone distance (< 120 vs. ≥ 120 mm) (see Appendix 2 for subgroup definition). Treatment allocation by subgroup interaction terms were defined in the regression models for this analysis. A further analysis was conducted using HRG-based costing for the index surgical intervention.

Results

RCT1

Resource use and costs

Resource use and cost from the NHS perspective by treatment allocation are reported in Table 17. Data for participants going through surgical procedures resulted in similar average time in anaesthetic room, theatre room and recovery room. As expected, fewer participants went through surgery in the ESWL group (n = 7) compared to the FURS group. Conversely, while ESWL were conducted using similar length of time in both study groups, fewer participants went through ESWL in the FURS allocation group (n = 17) compared with ESWL group (n = 191).

Table 17 also reports follow-up resource use up to 12 months post randomisation. A similar number of participants were readmitted into hospital in both groups (36%, n = 83 for FURS vs. 34%, n = 79 for ESWL), but those participants admitted in the latter group spent slightly more time in hospital. More participants attended outpatient appointments in the ESWL (30%, n = 69) compared with FURS group (24%, n = 55) although there was a higher mean number of visits for those in the FURS group. Participants reported a slightly lower mean number of GP and mean number of GP nurse contacts in the FURS group.

Costs for the index treatment episode were higher for FURS when either micro-costing or the HRG-based costing approach were used. However, additional procedures and hospital readmissions costs were higher for the ESWL group but the cost for outpatient appointments or GP visits are similar between the study groups. The 12-month care pathway mean cost for the NHS was £3408 and £2041 for the ITT with FURS and ESWL, respectively; giving an unadjusted difference of £1367 in favour of ESWL. Over 70% of the PUrE participants had complete data on costs (n = 330 of 461) and contributed to these mean costs calculation. A cost of zero for the index intervention was included for those individuals waiting for their index intervention at 12 months (20 for the FURS group and 6 for the ESWL group). Table 17 also shows the care pathway cost when HRGs were used to value the initial procedures. In this case, the 12-month care pathway mean cost for the NHS was £4084 and £2517 for the ITT with FURS and ESWL, respectively; resulting in FURS being £1567 more costly than ESWL.

Utility scores and quality-adjusted life-years

The EQ-5D mean health state utility scores based on the responses to the EQ-5D-5L and the mapping algorithm66 to the EQ-5D-3L scores by treatment allocation are reported in Table 18. QoL data were collected at baseline, pre intervention, weekly post intervention up to 12 weeks (3 months), and 12 months post randomisation. EQ-5D-5L data were collected when participants needed additional interventions beyond the 12-week window and treatment-related hospitalisations. Data attached to the index intervention are shown for observations within the 12-month trial follow-up only, with QALYs calculated with data from baseline, 12 months and any observation between these two time points. For the QALY calculation in Table 18, complete data were defined as those individuals with data for, at least, baseline and 12-month EQ-5D scores.

An imbalance in the mean EQ-5D scores favouring the ESWL group is seen at baseline and pre intervention. This difference favouring ESWL continues up to 6 weeks post treatment when the difference turns to favour FURS; probably reflecting the negative QoL impact of needing repeated ESWL sessions in the ESWL group. At 12 months post randomisation, the difference in the mean EQ-5D score favours ESWL. To be noted, the data include 20 and 6 participants that were waiting for their index intervention (see Table 6) at 12 months post randomisation. In the last row in Table 18 the average QALYs for the 12-month follow-up are reported. From a visual inspection of the SDs in Table 18, a small non-significant unadjusted mean QALY difference of 0.011 QALYs favouring ESWL emerges. However, this QALY calculation is based on data for 60% of participants and is not adjusted for baseline QoL.

Cost–utility analysis results

The base-case cost–utility analysis results based on the multiple imputation data set and adjusted by baseline EQ-5D utility score are reported in Table 19. The 12-month mean costs per participant were £2223 and £3362 for the ITT with ESWL and FURS groups, respectively, resulting in an adjusted cost difference of £1138 [95% CI £646 to £1631]. The mean QALYs per participant were 0.787 and 0.804 for the ESWL and FURS groups, respectively, producing an adjusted QALY difference of 0.017 QALYs (95% CI −0.008 to 0.043) for the 12-month follow-up period. Therefore, ITT with FURS results in significantly higher mean costs and non-significantly higher mean QALYs compared with ESWL. The ICER between FURS and ESWL was £65,163. Thus, ITT with FURS results, on average, in higher QALYs at a cost of £65,163 per QALY gained. This is above the usual cost-effectiveness threshold considered in the UK decision-making (i.e. below £30,000 per QALY gained).65 Moreover, given the cost and QALY differences seen in Table 19, either a difference of 0.057 QALYs or £349 are needed to reach an ICER of £20,000. These differences are outside the 95% CIs for QALYs (−0.008; 0.043) and costs (646; 1631) (see Table 19).

Figure 6 shows the cost-effectiveness plane with results from the 1000 bootstrapped iterations of the regression analysis nested multiple imputation. The incremental costs for FURS against ESWL are reported in the vertical axis and the incremental QALYs in the horizontal axis. All the iterations show positive mean costs differences, meaning that FURS was more costly than ESWL for all the iterations. For 11% of the iterations ESWL produced more QALYs resulting in a non-statistically significant difference of 0.017 favouring FURS. The red dashed line in Figure 6 represents the £20,000 cost-effectiveness threshold and separates iterations that result in either trial group being cost-effective. Those iterations sitting to the right and below the red dashed line indicate that FURS is cost-effective. Conversely, iterations to the left and above the red dashed line indicate that ESWL is cost-effective. All but five of the 1000 iterations sit above and to the left of the £20,000 threshold dashed line; that is just 0.5% of the iterations show QALY differences large enough (and cost difference small enough) for FURS to be considered cost-effective (see Figure 6). Increasing the threshold value (i.e. moving the red line anticlockwise from the zero point), would result in higher proportions of the iterations crossing the red line to the south and right in Figure 6. The effect of this is represented in Figure 7.

FIGURE 6 Trial-based incremental cost-effectiveness scatterplot for FURS vs. ESWL (base case; imputed data – 1000 bootstrap iterations). Note: Iterations below the red dashed £20,000 threshold line means FURS is cost-effective and above the line ESWL is cost-effective. (Figure not reproduced here; see the original article.)

FIGURE 7 Cost-effectiveness acceptability curves for FURS and ESWL groups (trial-based analysis; base case; imputed data – 1000 bootstrap iterations). (Figure not reproduced here; see the original article.)

The CEACs for FURS and ESWL are reported in Figure 7. The curves show the probability of either trial group being cost-effective at alternative cost-effectiveness threshold values. The CEAC for ESWL shows a 99.9% probability of being cost-effective at £20,000 threshold value. This probability decreases when the threshold value increases, reflecting the higher valuation of the additional QALYs resulting from FURS. However, the probability of ESWL being cost-effective is above 65% at relatively high cost-effectiveness threshold (i.e. £50,000) and the cost-effectiveness threshold has to be increased to £68,000 for a 50% chance of being cost-effective for both strategies (data not shown).

Alternative costing method

The results for the HRG-based alternative costing methodology for the index interventions are summarised in Table 20. Mean cost for ESWL and FURS are higher with the HRG-based costing compared with the base case using micro-costing. However, FURS cost increases more than ESWL; therefore, the analysis generates a larger cost difference between the study groups (£1372 compared to £1138) resulting in a higher ICER (£78,537) and an even lower probability for FURS being cost-effective at the usual threshold values used for decision-making in the UK NHS.

Indirect costs

Results on productivity costs (time off work) are reported in Table 78 in Appendix 9. These show that participants took an average of 2.3 and 6.6 days off work in the ESWL and FURS groups, respectively for the initial intervention and about half a day in both groups for any other admissions up to 12 months post randomisation. The total mean productivity cost associated to time off work was £416 (SD 2028) and £1012 (SD 4686) for ESWL and FURS groups, respectively. Table 78 also report the out-of-pocket expenditure. Only 45 individuals from 183 answering the question declared buying medications over the counter in the ESWL group and 52 of 173 in the FURS group, with the mean expenditure in medications declared being £4.60 and £7.60 for the ESWL and FURS groups, respectively. Only two participants declared buying private health care in the ESWL group and five in the FURS group, resulting in average costs of < £1. Finally, Table 78 presents out-of-pocket travelling costs for hospital admission, outpatient consultation, and GP visits. The mean costs of travel are all below £9 with no meaningful differences between the groups on visual inspection of the standard errors.

RCT2

Resource use and costs

In Table 21 the resource use and costs form the NHS perspective for RCT2 are reported. PCNL surgery took longer than FURS, with 20, 118 and 84 average minute times in anaesthetic, operating, and recovery rooms, against 15, 88 and 58 minutes for FURS. Length of stay in hospital for the initial operation was also longer for PCNL. However, more procedures were needed after the initial intervention in the FURS group and 53% (n = 39) participants needed to be readmitted into hospital (vs. 40%, n = 34 in PCNL group) with relatively longer average hospital stay (1.82 days for FURS vs. 1.22 fir PCNL). The FURS group also showed a higher proportion of participants attending hospital outpatient visits (33%, n = 24 for FURS vs. 27%, n = 23, for PCNL). Questionnaire respondents reported a higher number of average GP visits for PCNL and a similar average number of GP nurse visits.

The average costs of the initial intervention were higher for PCNL regardless of the method used to estimate costs (i.e., micro-costing or HRG-based costing). However, the average cost difference was larger when HRG-based costing was used (£2416 for HRG-based costing vs. £618 for micro-costing). The average costs of subsequent procedures, hospital readmissions and outpatient visits were higher for the FURS group, but the average cost of primary care visits was lower. These higher costs of subsequent procedures and hospital readmissions reverse the cost difference in favour of FURS for the initial intervention when micro-costing was used. The 12-month care pathway mean cost for the NHS was £5685 and £4708 for the ITT with FURS and PCNL, respectively; giving an unadjusted difference of £977 in favour of PCNL. Table 21 also shows the care pathway cost when HRGs were used to value the initial procedures. In this case, the 12-month care pathway mean cost for the NHS was £6091 and £6907 for the ITT with FURS and PCNL, respectively, resulting in FURS being £817 less costly than PCNL.

Utility scores and quality-adjusted life-years

The utility scores generated from responses to the EQ-5D-5L and mapped to the EQ-5D-3L scores are reported in Table 22. The mean EQ-5D scores are slightly higher for the FURS group at baseline and pre-index interventions. Mean EQ-5D scores reduced for both study groups at 1 week after their initial intervention and show a steady recovery up to week 6, when mean scores stabilise. However, week-by-week post-treatment scores are slightly higher for PCNL. At 3 months post treatment, the EQ-5D scores are higher for both groups compared with the scores at baseline and pre treatment. Mean EQ-5D scores converge to the baseline values at 12 months; however, scores at 12 months included individuals waiting for their index intervention (10 and 8 for the FURS and PCNL groups, respectively). QALYs reported in Table 22 were calculated for 66% for the participants (n = 105 of 159) resulting in 0.777 and 0.781 QALYs for FURS and PCNL, respectively, and an unadjusted mean difference of 0.004 QALYs.

Cost–utility analysis results

The cost–utility analysis results for the base case are reported in Table 23. Micro-costing was used for the index intervention and multiple imputation to deal with missing data. The 12-month mean cost per participant were £4565 and £5298 for ITT with PCNL and FURS, respectively, resulting in an adjusted non-significant mean difference of £733 (PCNL less costly). Mean QALYs were 0.775 and 0.773 for PCNL and FURS, respectively, generating an adjusted non-significant QALY difference of −0.001 QALYs in favour of PCNL. Therefore, ITT with FURS is on average more costly and does not produce additional QALYs compared with ITT with PCNL.

The results for the 1000 bootstrapped iterations of the regression analysis nested multiple imputation are shown in Figure 8. The mean QALY differences between FURS and PCNL are presented in the horizontal axis and mean cost differences in the vertical axis. For only 6.5% of the 1000 iterations PCNL is more costly than FURS (negative cost difference in Figure 8). Besides, 51% of the iterations result in FURS producing more QALYs than PCNL. The red dashed line in Figure 8 represents the £20,000 cost-effectiveness threshold. Iterations sitting to the right and below the red dashed line indicate that ITT with FURS is cost-effective and iterations to the left and above indicate that ITT with PCNL is cost-effective. Only 13% of the result are below the threshold line meaning that FURS is unlikely to be considered cost-effective compared with PCNL. The effect of increasing the threshold above £20,000, that is, moving the red dashed line clockwise with centre in the zero point, is shown in Figure 9.

FIGURE 8 Trial-based incremental cost-effectiveness scatterplot for FURS vs. PCNL (base case; imputed data – 1000 bootstrap iterations). Note: Iterations below the red dashed £20,000 threshold line means FURS is cost-effective and above the line that PCNL is cost-effective. (Figure not reproduced here; see the original article.)

FIGURE 9 Cost-effectiveness acceptability curves for FURS and PCNL groups (trial-based analysis; base case; imputed data – 1000 bootstrap iterations). (Figure not reproduced here; see the original article.)

The CEACs showing the probability of ITT with FURS or PCNL being cost-effective at alternative cost-effectiveness threshold values are reported in Figure 9. PCNL is less costly for the great majority of the simulations in Figure 8 and given the small differences in QALYs between the groups, the CEACs in Figure 9 show PCNL as highly likely to be cost-effective at the usual cost-effectiveness threshold values used for decision-making in the UK NHS (e.g. probability of 0.87 at £20,000 per QALY gained).67 The probability of PCNL being cost-effective reduces as the cost-effectiveness threshold increases. However, only 51% of the iterations result in FURS producing more QALYs than PCNL; therefore, even at very high threshold values, the probability of FURS being cost-effective cannot be higher than 51%.

Alternative costing method

Sensitivity analysis was conducted using the HRG codes to allocate costs for the index and additional interventions instead of the micro-costing approach. The result for this analysis is reported in Table 24. The QALY mean differences remain unchanged with respect to the base-case analysis as this analysis only varied the calculation of costs. ITT with PCNL mean total cost was £6703 and with FURS was £5769. These mean costs produce an adjusted non-significant cost difference of £934 in favour of FURS (PCNL more costly). These cost changes reflect in a downward shift of the simulated iterations in Figure 9 cost-effectiveness scatterplot. This is presented in Appendix 9, Figure 32. The great majority of the iterations have crossed to the south and right of the £20,000 threshold represented by the red dotted line, meaning that FURS is highly likely to be considered cost-effective. This is expected: given the very small difference in QALYs the probability of being cost-effective will follow the distribution of the difference in costs. The CEACs for this analysis are reported in Appendix 9, Figure 33. ITT with FURs has over 80% probability of being cost-effective at the usual threshold values used in the UK NHS for decision making (e.g. up to £30,000 per QALY gained).65

Indirect costs

Results on productivity costs (time off work) for RCT2 are reported in Table 79 in Appendix 9. These show that participants took an average of 2.4 and 7.6 days off work in the PCNL and FURS groups respectively, for the 12 weeks after the initial intervention, and 7 and 4 additional days, respectively, for the period up to 12 months post randomisation. The total mean productivity cost associated to time off work were £1200 (SD: 6593) and £2146 (SD: 6499) for PCNL and FURS groups, respectively. Table 79 also reports the out-of-pocket expenditure. Only 9 individuals from 69 answering the question declared buying medications over the counter in the PCNL group and 15 of 60 in the FURS group, with the mean expenditure in medications declared being £0.60 and £5.30 for PCNL and FURS groups, respectively. No participant declared buying private health care in the PCNL group and 2 in the FURS group, resulting in average costs of £1.80. Finally, Table 79 presents out-of-pocket travelling costs for hospital admission, outpatient consultation, and GP visits. The mean costs of travel are all below £10, with no meaningful differences between the groups on visual inspection of the standard errors.

Economic modelling extrapolation

The statistical analysis in Chapters 3 and 4 showed differences between the trial groups in the number of participants with residual fragments 12 weeks after the initial treatment and 12 months after randomisation (see Tables 6 and 13). The presence and size of residual fragments are associated with higher rates of recurrence,67 potentially leading to utility and resource use differences beyond the trial follow-up period. The aim of this section is to report on the modelling extrapolation performed to extend the time horizon for the economic analysis beyond the 12-month trial follow-up to 5 years.

Methods

Decision analytic model

A cohort Markov model was developed using TreeAge Pro software (TreeAge Software, Inc., Williamstown, MA, USA). The structure of the model was informed by discussions within the trial management group and clinical collaborators (Figure 10). A similar model structure was used for RCT1 and RCT2 and all the compared treatment strategies (i.e. ITT with ESWL, FURS and PCNL). Individuals referred to secondary care and with confirmed lower pole kidney stones enter the model in the ‘Waiting for index intervention’ health state and are assigned to the index intervention: either ESWL, FURS or PCNL. The model recreates the care pathways observed in the PUrE trial on an ITT basis incorporating the waiting times seen in the trial. Individuals undergoing treatment accrue the cost of the initial episode and move to the tunnel state ‘Index intervention’, where a reduction in QoL associated with the intervention is accounted for. Markov tunnel states are a series of temporary states ‒ each one lasting one cycle ‒ that must be visited in a fixed sequence.68 Mirroring the PUrE trial definition for stone cleared, on exit from the tunnel Markov state, individuals enter either of ‘Stone cleared/no fragments’, ‘Fragments ≤ 4 mm’ or ‘Fragments > 4 mm’ Markov states. Furthermore, simulated individuals experiencing a recurrence and needing further interventions are moved to the ‘Further intervention’ Markov tunnel state (lasting three-monthly cycles as for the ‘Index intervention’ Markov state).

FIGURE 10 Simplified schematic for the PUrE Markov model. (Figure not reproduced here; see the original article.)

For the Markov process to end, Markov models need at least one absorbing state; that is, a state that cannot be left. This is represented by the ‘Death’ state in Figure 10. The Death state can be entered from any other state based on age-specific mortality rates reported in the UK life tables.69 The state occupancy and estimated payoffs are updated on a constant monthly Markov cycle.

Population

The characteristics of the cohort of individuals entering the model match those of the participants in the PUrE trials. The starting ages for the model were 54 and 59 years for RCT1 and RCT2, respectively, and the proportion of male individuals in the simulated cohort were 62% and 63% for RCT1 and RCT2, respectively.

Time horizon and discounting

A time horizon of 5 years was adopted based on discussions within the trial management group and on the expectation that any events beyond 5 years were likely to be related to a new stone and not to fragments left from the initial intervention. Following economic evaluation method guidelines for the UK, costs and QALYs accrued beyond 12 months were discounted at an annual rate of 3.5%.65

Clinical input parameters

While the PUrE trial follow-up was 12 months, data were collected up to 12 weeks post intervention for all trial participants, including those that had their index intervention after 12 months post randomisation. Figures 11 and 12 show Kaplan–Meier curves estimated from the PUrE data for RCT1 and RCT2, respectively. Transition probabilities were estimated as 1 − (S(t₁)/S(t₀)) straight from these Kaplan–Meier data, where S denotes the survival proportion and t_i, the time (Markov cycle).

FIGURE 11 Kaplan–Meier plot of time to index operation by treatment allocation for RCT1. (Figure not reproduced here; see the original article.)

FIGURE 12 Kaplan–Meier plot of time to index operation by treatment allocation for RCT2. (Figure not reproduced here; see the original article.)

On a successful intervention, individuals can become stone-free. The RR for stone clearance at 12 weeks was obtained from the analysis of the PUrE data. Generalised linear models with binomial family and a log link, clustered by recruiting centre, were used to obtain the SFR risk ratios between FURS and ESWL for RCT1 (RR 1.99, 95% CI 1.61 to 0.2.45), and between FURS and PCNL for RCT2 (RR 0.65, 95% CI 0.53 to 0.81). The proportion of individuals with residual fragments of size ≤ 4 mm and above 4 mm, given the stone was not cleared, were obtained straight from the PUrE data (Chapters 3 and 4, Tables 6 and 13, and Appendix 2). Normal distributions were attached to risk ratios and beta distributions to the proportion of residual fragments of size ≤ 4 and > 4 mm (see Appendix 9, Table 80).

Data to estimate the proportion of individuals needing further treatment due to recurrence were retrieved from the retrospective study by Emmott et al. (2018)70 and based on 263 individuals who received postoperative CTKUB with residual fragments of at least 1 mm. The authors report the Kaplan–Meier survival curves for time to the need of PCNL surgery. Data from the original publication were extracted using a web plot digitiser (https://automeris.io/WebPlotDigitizer/) with individual data obtained following methods provided by Wei and Royston (2017) using the Stata command ipdfc.71 Several parametric survival functions were fitted to the reconstructed data (exponential, Weibull, Gompertz and loglogistic). However, from visual inspection it was clear that none of the parametric adjustments satisfactorily fitted the extracted data. Therefore, a decision was taken to build up the transition probabilities for fragments ≤ 4 and > 4 mm to the ‘Further intervention’ Markov state, based on the extracted Kaplan–Meier data from Emmott et al. (2018)70 (Figure 13) using the same formula used for the time to index intervention. Furthermore, the risk of recurrence for individuals with no fragments (stone cleared) was taken from D’Costa et al. (2019)72 based on the population-based study by Vaughan et al. (2019).73 The authors report a 17% 5-year recurrence rate for individuals that had experienced a first stone episode. This rate was converted to a monthly probability assuming a constant monthly rate.74

FIGURE 13 Kaplan–Meier plot of time to recurrence by size of residual fragment. (Figure not reproduced here; see the original article.)

Health state utilities

PUrE EQ-5D-5L trial data were analysed to estimate health state utilities for the model (Table 25). Pooled baseline EQ-5D mean scores for each trial were used as utility weights for the ‘Waiting for index intervention’ Markov state. EQ-5D score mean and standard error by week were obtained for ESWL (RCT1) and PCNL (RCT2) groups. These results were combined with weekly treatment effects for FURS. Separate regression models were estimated for RCT1 and RCT2. Mixed-effects regression models for repeat measures with adjustment for baseline EQ-5D score, and including centre as a random effect, were defined to estimate the EQ-5D utility score difference between trial groups by data collection time point. Average utility and treatment effects for weeks 1–4, 5–8 and 8–12 were calculated and used as utility weights for tunnel cycles 1, 2 and 3, respectively. In addition, EQ-5D score data at 12 months post randomisation were classified according to the stone being completely cleared, and residuals fragments ≤ 4 and > 4 mm. Using all data pooled for RCT1 and 2, a mixed-effects regression analysis adjusted by baseline EQ-5D score was estimated with EQ-5D score at 12 months as dependent variable and stone clearance as categorical independent variable. This showed no difference in utility score between those with the stone completely cleared and those with reported fragments ≤ 4 mm. Therefore, the classification was dichotomised into completely clear/residual fragments ≤ 4 mm and residual fragments > 4 mm, and the regression model re-estimated. The results of this analysis were used to estimate the utility score difference between stone cleared/residual fragments ≤ 4 mm, and residual fragments > 4 mm Markov states (see Table 25).

Health service resource use and costs

Cost for ESWL, FURS and PCNL were obtained from the National Schedule of NHS Costs for year 2020–1 (Table 26). Unit costs for day cases were assumed for ESWL and FURS for HRG codes LB36Z and LB65E, respectively. Unit cost for elective impatient using HRG code LB75B was assumed for PCNL. Gamma distributions were attached to all unit costs to explore parameter uncertainty by conducting probabilistic sensitivity analysis (PSA). Costs for the care pathways for the first year were obtained directly from the RCT within trial analysis; therefore, the HRG unit costs were applied for index and further interventions occurring after the first 12-monthly cycles.

Finally, no differences in the number of outpatient or GP contacts were observed by the study group in the within-trial analysis for PUrE, therefore, a simplifying assumption was made and no cost for these contacts were incorporated in the model.

Model verification

To assess the internal validity of the Markov model, model verification was conducted throughout the model implementation stage following the list proposed by Büyükkaramikli et al. (2019).75 For example, model formulae were verified using external software [Microsoft Excel^® (Microsoft Corporation, Redmond, WA, USA)]. The final model results were checked when varying the model input parameter values (e.g. defining all utility scores equal to 1 and 0 discount rate to check whether total QALYs equal total life years) to assess the consistency of the model performance given specific variation in the model input values (black-box testing). The validity of the projected estimates of surgical events was assessed by plotting the model Markov traces with the Kaplan–Maier curves for the PUrE trial with visual inspection showing Markov traces followed closely the Kaplan–Meier curves.

Model analysis

The Markov cohort model was run deterministically using expected costs and QALYs. Parameter uncertainty was characterised by running second-order Monte Carlo simulations using the probability distributions attached to the model mean parameter values. Beta distributions were defined for probabilities (risks) and utilities, gamma for costs and normal distributions for treatment effect for FURS. Details of the probability distributions are reported in Appendix 9, Table 80. PSA were conducted by running 10,000 random draws from the allocated distributions, producing 10,000 estimates of incremental costs and QALYs. Result tables also show the probability of ESWL, FURS, or PCNL being cost-effective at £13,000, £20,000 and £30,000 per QALY thresholds.65 CEACs and cost-effectiveness scatterplots are reported in Appendix 9, Figures 34 to 37.

The base-case analysis for RCT1 and RCT2 assumed that individuals needing further treatment for recurrence after 12 months were treated with the intervention they were initially allocated to (e.g. ESWL and FURS for participants allocated to ESWL and FURS, respectively for RCT1). In addition, different utility weights were assumed for individuals with residual fragments above 4 mm (0.916) and those with residual fragments smaller or equal to 4 mm (0.965). Therefore, the difference in utility after 12 months results from the additional treatments received due to recurrence, the waiting time to receive the initial intervention, and the alternative utility weights according to residual fragment size after treatment.

The time horizon for the base-case analysis was 5 years. The analysis was conducted from the NHS perspective with cost and QALYs beyond 1 year discounted at a 3.5% rate.65

The within trial analysis for RCT2 using a micro-costing approach showed PCNL to be less costly and more effective than FURS. It is expected that a higher proportion of individuals treated initially with FURS will need further interventions compared with PCNL. It is also expected that the additional treatment costs and the reduction in QoL will worsen the cost-effectiveness of FURS versus PCNL. For this reason, the base case for the extrapolation beyond the trial follow-up was performed with the HRG costing that resulted in FURS being cost-effective at 12 months post randomisation.

Sensitivity analysis

Key sensitivity analyses were performed:

• assuming alternative time horizons from 12 months to 5 years;
• one-way sensitivity analysis (threshold analysis) varying the RR of stone free/stone clear rate;
• a scenario analysis incorporating equal utility weights for all individuals after treatment regardless of stone being cleared or not (i.e. utility weight of 0.965 for stone cleared, and fragments below or above 4 mm).

Results

RCT1

Base-case analysis

The results for the base-case cost–utility analysis are reported in Table 27. Understandably, the expected costs and QALYs are higher for the 5-year time horizon compared to the 12-month follow-up for the within-trial analysis. These higher costs are explained by the proportion of individuals needing treatment due to recurrence in the longer time period considered. The difference in costs and QALYs increase compared to the within-trial base-case analysis (see Table 19). The cost and QALY joint effect generate a decrease in the ICER below £30,000 but above £20,000 cost-effectiveness threshold values, indicating that FURS is still not cost-effective when the longer time horizon of 5 years is considered. The PSA shows FURS is unlikely to be considered cost-effective (26% at £20,000 per QALY threshold) unless a higher cost-effectiveness threshold is considered (i.e. 55% at £30,000). The cost-effectiveness scatterplot (see Appendix 9, Figure 34) shows FURS produces higher expected QALYs for 99% of the PSA iterations but is also more costly in all cases. Finally, the CEACs (see Appendix 9, Figure 35) show FURS has over 80% chance of being cost-effective at £50,000 per QALY threshold.

Alternative time horizon

The cost–utility analysis results for alternative time horizons applied to the base case are reported in Table 28. Cost and QALY increase the longer the time horizon. However, the ICER decreases the longer the time horizon assumed, and the probability of FURS being cost-effective rises. This is explained by the higher proportions of individuals requiring further treatment in the ESWL group. While the ICER between FURS and ESWL decreases the longer the time horizon considered, the fall is not enough to cross the £20,000 cost-effectiveness threshold.76

Scenario analyses

Same utility weight for individuals with residual fragments below or above 4 mm

The results for the scenario analysis are also reported in Table 28. Assuming equal utility for individuals with residual fragments sized > 4 mm increases QALYs in both ESWL and FURS model strategies. However, the increase in expected QALYs is proportionally higher for ESWL. This is explained by the higher proportion of individuals remaining in the ‘Fragments > 4 mm’ Markov state for ESWL. Therefore, the ICER increases well above the cost-effectiveness threshold used in the UK (e.g. £20,000 per QALY gained). Also, the probability of FURS being cost-effective further reduces to zero at £20,000 cost per QALY threshold.

Health Care Resource Group-based costing for randomised controlled trial initial interventions

Costing the initial intervention in PUrE with HRG-based unit costs increases the difference in cost between FURS and ESWL maintaining the base-case QALY difference unaltered. This results in a higher ICER just above £30,000 and a lower probability of FURS being cost-effective (just 17% at £20,000 per QALY threshold) (see Table 28).

Relative risk of stone-free rate between flexible and extracorporeal shockwave lithotripsy

The results for the one-way sensitivity analysis where the SFR RR at 12 weeks after treatment was varied are reported in Figure 14. The stone-free RR is presented in the vertical axis while the ICER for FURS against ESWL is shown in the horizontal axis. ICERs above the cost-effectiveness threshold (e.g. £20,000) mean that, on average, FURS is not cost-effective. Conversely, ICERs below the cost-effectiveness threshold mean that FURS is, on average, cost-effective. In Figure 14, this is illustrated with a dashed vertical line representing the £20,000 per QALY threshold. The analysis shows that a stone-free RR above 2.2 is needed for FURS to be cost-effective.

FIGURE 14 One-way sensitivity analysis varying RR of stone-clear between FURS and ESWL (> 1 FURS better). (Figure not reproduced here; see the original article.)

RCT2

Base-case analysis

The results for the base-case analysis using HRG-based costing are reported in Table 29. Expected costs and QALYs are higher than the mean cost and QALYs for the within-trial analysis due to the extrapolation from 12 months to 5 years. The difference in costs of £1488 is higher compared to the cost difference for the within-trial analysis. While the proportion of individuals needing further interventions is lower for PCNL compared with FURS, the expected cost of the PCNL strategy rises due to the difference in the unit costs between PCNL and FURS. The difference in QALYs is much higher compared with the within-trial analysis (see Table 23). This is explained by the higher proportion of individuals moving to the stone-free Markov state in the PCNL group and the relatively lower QoL experienced for those going through FURS compared to those who experience PCNL. The cost and QALY differences result in an ICER of £24,382 just higher than the £20,000 threshold used in the UK.65 The higher the cost per QALY willing to pay, the higher the probability of PCNL being cost-effective (i.e. 58% at £30,000 threshold – Table 29). The incremental cost-effectiveness scatterplot and the CEACs for PCNL and FURS are reported in Appendix 9, Figures 36 and 37. These show PCNL producing more expected QALYs for 97% of the iterations but at higher cost in 98% of the iterations.

Sensitivity analysis

The results for the one-way sensitivity analysis varying the model time horizon are reported in Table 30. The 12-month time horizon correspond to the within-trial analysis result when the HRG-based costing was used to cost the index interventions. The statistically significant difference in cost together with the lack of significant difference in QALYs resulted in an ICER of £883,375, well above the usual cost-effectiveness thresholds for the UK NHS. This means that at 12 months, ITT with FURS is less costly with no evidence of a substantial impact, on average, in QALYs. Cost and QALYs increase the longer the time horizon considered. However, the difference in QALY increased proportionally more, reducing the ICER below the £30,000 threshold when a time horizon of 48 months is considered.

The results for the one-way sensitivity analysis varying the SFR RR are reported in Figure 15. The ICER between PCNL and FURS is presented in the horizontal axis and the SFR RR in the vertical axis (< 1 PCNL better). The £20,000 cost-effectiveness threshold is also shown and is represented by the vertical black dashed line. The vertical line divides cases where PCNL (left of the dashed line) or FURS (right of the dashed line) are considered cost-effective. The SFR RR has to be at least 0.6 for FURS to be considered cost-effective. For comparison, the RR obtained from PUrE data and used in the base-case analysis was 0.65 (95% CI 0.53 to 0.81). The analysis using micro-costing to cost the initial procedures is presented in Appendix 9, Table 81 and Figure 38. This shows a RR over 0.95 is needed for FURS to be considered cost-effective.

FIGURE 15 One-way sensitivity analysis varying RR of stone-clear between FURS and PCNL (< 1 PCNL better). (Figure not reproduced here; see the original article.)

The result for the scenario analysis assuming no difference in QoL between individuals with residual fragments above 4 mm is reported in Table 31. Expected costs do not vary with respect to those in the base case (see Table 29). However, the difference in utility reduces to a third of the difference seen in the base case (from 0.061 to 0.021). This increases the ICER over the usual cost-effectiveness threshold used in the NHS for decision-making.76

Summary and discussion

The within-trial cost utility analysis reported in this chapter indicates that ITT with FURS for kidney stones ≤ 10 mm in maximum dimension was more costly over a 12-month follow-up period than ESWL (the mean difference was £1138). This cost difference was driven by the high relative cost of the initial FURS intervention. While the mean costs of additional and subsequent hospital admissions were higher for the ESWL group, these higher costs did not overturn the initial difference in cost in favour of ESWL. A non-statistically significant QALY difference in of 0.017 QALYs in favour of FURS was observed for the 12 months after randomisation. This result is consistent with the main statistical analyses reported in Chapter 3 where non-statistically significant differences were observed for the AUC at 12 weeks post treatment for all but one analyses. The significant cost difference coupled with a relatively small non-significant QALY difference favouring FURS (FURS produced marginally more QALYs) resulted in an ICER of £65,163 per additional QALY. This is well above the usual cost-effectiveness threshold used in the UK NHS.76 That is, moving from an ITT with ESWL to ITT with FURS could produce, on average, more QALYs but at a very high cost per QALY gained. The probabilistic analysis showed a high probability for ESWL to be cost-effective (i.e. over 99% at a £20,000 cost-effectiveness threshold). These results were robust to the sensitivity analysis using HRG-based costing instead of micro-costing.

Intention to treat with FURS is more costly and produces fewer QALYs than ITT with PCNL to treat stones size > 10 and ≤ 25 mm in maximum dimension, but these differences are not statistically significant: the mean difference was £733 (95% CI: −£508 to £1973) for costs and −0.001 QALY (95% CI: −0.044 to 0.042 QALYs). While PCNL showed higher costs associated with the initial procedure, the cost due to hospital readmissions and subsequent procedures were higher for FURS, resulting in an overall higher average cost for FURS. Therefore, mean cost difference against FURS was driven by the similar cost for the index procedures together with the higher cost of subsequent health care for the FURS group. This, together with the QALY difference means that, on average, FURS was more costly and produced fewer QALYs than PCNL, having a 98% chance of being cost-effective at a £20,000 cost-effectiveness threshold.

The sensitivity analysis using HRG-based costing for the initial interventions radically changed the base-case results. This analysis shows a mean cost difference of £934 (95% CI: −£2582 to £714) in favour of FURS for the 12-month follow-up. This is explained by the cost associated with the HRG codes for FURS (LB65) and PCNL (LB75) that give weighted average costs of £4259 and £8114; a difference of £3855 against PCNL that is not levelled by the higher costs of subsequent health care in the FURS trial group. It is open for discussion which of the two costing approaches (micro-costing or HRG-based) reflect better the valuation of the resource use difference between FURS and PCNL. As the QALY difference between the trial groups is very small and non-significant, the cost-effectiveness results mirror the distribution of the difference in cost with the resulting ICER rising to £883,375. This means that ITT with FURS would save, on average, £883,375 per QALY forgone. This result, as well as the base-case analysis result, should be taken with caution given the non-significant and very small difference in QALYs.

We extended the 12-month trial follow-up analysis up to 5 years using a Markov model following best practice. The model reflects possible care pathways after the initial intervention and the potential need for further intervention due to recurrence. For stones ≤ 10 mm in maximum dimension, the model results showed that ITT with FURS generates more QALYs than ESWL but is also more costly, with incremental cost per QALY (£26,571) higher than the usual threshold used in the NHS and a low probability of being considered cost-effective (26% at £20K base case). These results were robust to a number of sensitivity analyses such as alternative time horizon, RR of SFR, and equal utility weights between those with residual fragments less than or equal to, and greater than 4 mm.

For kidney stones size > 10 and ≤ 25 mm in maximum dimension (RCT2), the cost-effectiveness of FURS was dependent on the method used to estimate costs, with PCNL and FURS being cost-effective when micro-costing or HRG-based costing were used, respectively. The extrapolation exercise showed PCNL is still cost-effective when micro-costing was used. Moreover, the results using HRG-based costing show PCNL cost-effectiveness improving for longer time horizons. However, while at 5 years PCNL generates more QALYs, it does this at an additional cost that is too high to be considered cost-effective. Finally, our analysis suggests the RR stone-free should be above 0.6 for FURS to be cost-effective when HRG-based costing was used (and above 0.95 when micro-costing was used).

Strengths and limitations

The main strength of the analysis reported in this chapter is related to the RCT design. RCTs provide the opportunity to collect data prospectively for individual participants and groups that are a priori comparable. The randomised treatment allocation increases the probability of unbiased cost and QALY calculations, and consequently a fairer estimate of the difference in cost and QALYs between the treatment groups. The PUrE study comprised two multicentre pragmatic trials facilitating the generalisability of the trial results to those individuals routinely treated in the UK NHS for kidney stones in the lower pole of the kidney. A further strength is that the economic analysis was conducted following a prespecified Health Economic Analysis Plan that followed established economic evaluation methods.77

An important limitation of the within-trial cost–utility analysis is the 12-month follow-up. One year might be a short period of time to allow for all relevant cost and consequences of the treatments under consideration. At 12 months post randomisation, a number of participants were still waiting for their index intervention and acceptable stone clearance rates differed between groups (see Tables 6 and 13). These issues could potentially trigger further differences in healthcare resource use and/or QoL between the study groups and provide a rationale for the extrapolation exercise. We used PUrE time to event data after 12 months to inform the proportion of the cohort needing their first surgical intervention, SFRs and RRs for FURS, and the proportion of the cohort with residual fragments of size ≤ 4 or > 4 mm after 12 weeks.

Our modelling extrapolation should inform the impact of using longer time horizons for the economic analysis.

As with any decision model, our modelling extrapolation is also subject to limitations. First, the data for the need for further treatment due to residual fragments (≤ 4 and > 4 mm dimension) were obtained from the Canadian study by Emmott et al. 2018, a retrospective hospital record review where the existence of residual fragments was assessed on day 1 after PCNL. These data might not accurately reflect the NHS care pathway after 12 months of the index intervention. However, the authors’ risk of further intervention (Kaplan–Meier curves) was applied on exit of the tunnel Markov state and disregarding the first 90 days of the Kaplan–Meier data. Arguably, this should better reflect the need for further intervention and less the differing standard clinical practice in Canada and the UK. Secondly, a lower utility weight was assumed for those individuals with residual fragments larger than 4 mm based on the analysis of the PUrE data. While there is some evidence of lower QoL due to residual fragments above 4 mm dimension,78 the evidence is mixed.78 Also, the drop in QoL might come as a result of subsequent interventions with the risk of partially double counting the reduction in QoL in our model. We conducted sensitivity analysis assuming equal utility weights between those with fragments smaller and larger than 4 mm dimension. The base case results are robust to this change: for RCT1 FURS was even less likely to be cost-effective and for RCT2 cost-effectiveness is still dependent on the costing method used (i.e. FURS and PCNL being cost-effective when HRG or micro-costing methods are used, respectively).