Introduction
The purpose of this chapter is to report the methods and results of the within-trial economic evaluation of the TRIUMPH trial. Details on site and participant recruitment are described in Chapter 2 in General practice recruitment and selection and Participants, respectively. A cost–utility analysis comparing a manualised and standardised non-pharmacological intervention with usual care for men experiencing bothersome LUTS was conducted from a UK NHS perspective. The analysis compared costs and quality-adjusted life-years (QALYs) from randomisation to 12 months’ follow-up, the time horizon of the trial. A secondary analysis compared mean differences in change in IPSS with mean differences in costs.
Methods
Measurement and valuation of relevant resource use
Resource use was captured for each individual patient’s 12-month follow-up period, through four primary sources: trial management records, case report forms, primary care EMRs and self-report questionnaires. Data extraction from GP EMRs is described in Chapter 2, Electronic medical record data extraction.
The resources required for training were obtained through primary sources including trial management records and expert opinion from the lead nurses. These resources included staff time, and travel and teleconference expenses. Staff delivering the intervention prospectively recorded on case report forms their time spent in the delivery of the intervention, both at the initial clinic visit and during subsequent follow-up contacts.
Primary care EMRs were used to obtain resource use in relation to (1) primary care consultations for any reason with a GP, nurse, HCA or pharmacist; (2) LUTS-related prescribed medication; and (3) for some of the participants, LUTS-related secondary care activity [outpatient, day case, inpatient and accident and emergency (A&E) visits].
Secondary care LUTS-related healthcare use (outpatient, day case, inpatient and A&E visits) was obtained from self-completed questionnaires 6 and 12 months post intervention. The questionnaires were either completed online or sent and returned by post. This information was supplemented by a review of secondary care letters (received by primary care practices) of participants who had a urology referral recorded in their primary care EMR data but had not recorded this in the follow-up questionnaires.
Details of all resource use were measured; the unit cost sources used to value them are reported in Table 32. Resource use was valued using 2018/19 costs. When a unit cost was not available for the year of analysis, it was inflated to current prices using the NHS cost inflation index as published in Curtis and Burns’ Unit Costs of Health and Social Care 2019.27
| Resource | Unit cost (£) | Source of unit cost |
|---|---|---|
| GP surgery visit | 34a,b | Curtis and Burns, 201927 |
| GP telephone call | 26.27a,b,c | Curtis and Burns, 201927 |
| GP home visit | 87.28a,b,d | Curtis and Burns, 201927 |
| Practice nurse surgery visit | 10.85b,e | Curtis and Burns, 201927 |
| Practice nurse telephone call | 8.25b,f | Curtis and Burns, 201927 |
| HCA surgery visit | 6.67e | Curtis and Burns, 201927 |
| HCA telephone call | 5.07f | Curtis and Burns, 201927 |
| Pharmacist surgery visit | 11.63e | Curtis and Burns, 201927 |
| Pharmacist surgery telephone call | 8.84f | Curtis and Burns, 201927 |
| NHS 111 call | 12.26 | Pope et al.28 2017 |
| Trainers (professor, lead nurses) | Varies | gUniversity of Bristol pay scales, 2018/1929 |
| hCurtis and Burns, 201927 | ||
| Intervention booklet | 6.15 | University of Bristol printing service charge30 |
| Intervention visit time per minute | Variesi | Curtis and Burns, 201927 |
| Follow-up contact time per minute | Variesj | Curtis and Burns, 201927 |
| Teleconference call per minute | 0.15 | BT conference call service charge31 |
| Car mileage | 0.45k | HM Revenue and Customs32 |
| Outpatient procedures | Varies | NHS National Cost Collection, 2018/1933 |
| Outpatient visits | 110 | NHS National Cost Collection, 2018/1933 |
| Day cases | Varies | NHS National Cost Collection, 2018/1933 |
| Inpatient admissions | Varies | NHS National Cost Collection, 2018/1933 |
| A&E attendances | 159 | NHS National Cost Collection, 2018/1933 |
| Medications | Varies | Prescription Cost Analysis, 201934 |
| NHS Drug Tariff, 201935 |
The printing of the intervention booklets, which were given to participants at their initial intervention visit, was valued using the actual price charged by the University of Bristol’s printing services for printing all booklets. From this aggregate price, it was possible to calculate the unit cost per booklet. As all booklets had to be prepared in advance, it was assumed that all participants randomised to the intervention arm incurred the booklet cost.
The NHS staff time used for training and for the intervention was valued using Curtis and Burns’ Unit Costs of Health and Social Care 2019.27 The University of Bristol staff time used for training was valued using the university’s 2018/19 pay scales29 and included basic salary, National Insurance and superannuation. All primary healthcare resource use was valued using Curtis and Burns’ Unit Costs of Health and Social Care 2019.27 Medications were assigned a unit cost based on the prescription cost analysis for 2019.34 If a unit cost was not available in the prescription cost analysis, which was rare, the unit costs reported in the 2019 NHS Drug Tariff were used.35
Secondary healthcare resource use was valued using NHS costs published in the 2018/19 National Cost Collection.33 For outpatient visits a weighted average of the unit costs for consultant- and non-consultant-led visits was used.
Outcome data collection and valuation
The primary outcome for the economic evaluation was the QALY, as recommended in NICE’s reference case for interventions funded by the NHS.37 Participants completed the EQ-5D-5L at baseline and at the 6- and 12-month follow-ups. At baseline the questionnaire was administered via post; at the 6- and 12-month follow-ups it could be completed online or via post. Participants’ EQ-5D-5L scores were mapped to the EuroQol-5 Dimensions, three-level version (EQ-5D-3L), valuation set using a validated mapping function by van Hout et al.38 This approach is currently recommended by NICE. The EQ-5D-3L valuation set enables a utility score to be calculated for each patient based on published UK population utility values. Utility scores were then used to calculate the QALYs for each patient using an area-under-the-curve approach. The approach accounted for any deaths that occurred during the 12-month trial period.
The primary outcome of the effectiveness analysis (i.e. change in IPSS) was also evaluated in a secondary cost-effectiveness analysis.
Data cleaning and missing costs and outcomes
The EMR data were derived from two different information technology (IT) systems; therefore, a standardised cleaning process was implemented to ensure that the final analysis data set was comparable across the two systems (detailed in Appendix 2).
Clinician opinion was used to ascertain whether or not a secondary care visit was related to LUTS. In terms of missing data within questionnaires, if other sections of the questionnaire had been completed but the resource use questions had been left blank, then it was assumed that zero resource was incurred. If any part of a resource use question was unanswered or any items of the EQ-5D-5L were left blank, the participants were excluded from the complete-case analysis.
Simple mean imputation was used to impute data in the case of the few participants whose case report forms did not include duration of the follow-up visit at week 1 (n = 1) or week 4 (n = 7). If a participant did not attend the follow-up visit, then the duration of the follow-up visit was coded as zero.
Analysis
All economic analyses were based on ITT, whereby all participants were analysed in accordance with the allocation group they were assigned to at the point of randomisation. As costs and benefits were not assessed beyond the 12-month follow-up period, discounting was not required. Similarly, as this trial was a definitive trial, it was not within the scope of the work package to extrapolate the costs and outcomes beyond the 12-month trial period.
Resource use quantities were multiplied by their relevant unit costs to calculate the total mean resource use and total mean cost by resource category and allocation group, as well as total mean NHS costs. Similarly, mean EQ-5D-5L domain scores for each time point were calculated and reported by allocation group.
A multilevel model (MLM) was used to account for the clustered data and to adjust for the minimisation variables (practice size, IMD and centre) and for the general practice IT system and baseline utility for costs and QALYs, respectively. The MLM approach used has been recommended by El Alili et al.,39 who found MLMs to perform better than ordinary least squares regression for trial-based economic evaluations that are based on clustered data. The MLM was combined with bias-corrected and accelerated bootstrapping to calculate the adjusted mean cost and QALYs, and the adjusted mean differences in costs and QALYs (and their associated 95% CIs) between groups. This type of bootstrapping was used to account for the skewness in the cost and QALY data.
Cost-effectiveness analysis
Outputs from the MLM were used to estimate the incremental net monetary benefit (INMB) statistic and the associated CIs at NICE’s recommended willingness-to-pay thresholds of £20,000 and £30,000 per QALY. A cost-effectiveness plane was created to graphically show the differences in costs and health outcomes between the trial arms. A cost-effectiveness acceptability curve was constructed for a range of willingness-to-pay thresholds using the parametric p-value approach, a method employed by El Alili et al.39 and Hoch et al.40
A secondary analysis examined mean differences in IPSS (adjusted for baseline IPSS and minimisation variables) with the adjusted mean differences in costs (adjusted for minimisation variables and general practice IT system).
Sensitivity analyses
Several sensitivity analyses were conducted to explore the uncertainty in the methodological choices made in the base-case analysis.
A seemingly unrelated regression (SUR) model was used as an alternative approach to analysing the data.
Alternative approaches to handling missing data were performed; these included the following:
- Multiple imputation by chained equations using predictive mean matching was used in combination with the SUR model. The covariates in the multiple imputation model were site identification, baseline utility, centre, practice size, IMD and general practice system. The model was run by allocation group and a randomisation seed was set to ensure reproducible imputations. A total of 25 imputations were performed and combined with Rubin’s rule in Stata 16.1.41
- Participants whose secondary care data were based on letters rather than self-report were excluded from the analysis.
- Participants who had missing questionnaire data, first, had their costs imputed to zero and their QALYS imputed to a value 10% higher than the mean QALY value for their respective trial arms. Second, they had their costs imputed to the highest participant cost for secondary care per arm, and their QALYs to a value 10% lower for their respective trial arms. These analyses were conducted to examine the impact of missing questionnaire data being potentially missing not at random.
- The most frequently reported quantity of tablets per medication item in data from practices using EMIS Web were used to impute the missing quantities in data from practices using SystmOne. It had not been possible to extract information in relation to quantity of medications from the SystmOne practices. In the base-case analysis the mean quantity of tablets per medication item from EMIS Web practices were used to impute these missing data.
Uncertainty in the way HCPs code the mode of delivery for their consultations was explored by assuming that all contacts with the same type of HCP have the same unit cost (face-to-face unit cost) regardless of mode of delivery. In terms of intervention costs, training costs were excluded to test the assumption that these costs would be captured in the published unit costs. A further analysis excluded all intervention costs. Finally, the model was adjusted for GP consultation costs for the 12-month pre-consent period.
Subgroup analysis
An analysis was performed in which participants who completed follow-up from 11 March 2020 (11 March 2020 is the date when the World Health Organization declared the COVID-19 outbreak a pandemic) were compared with those who completed follow-up before this date.
Results
Complete NHS cost data were available for 902 (83.8%) of the 1077 consented trial participants, and complete QALY data were available for 904 (83.9%) participants. The complete-case analysis was conducted on 866 (80.4%) participants, 413 (78.8%) in the intervention arm and 453 (81.9%) in the usual-care arm.
Unadjusted mean resource use and costs are present in Table 33. Table 33 also reports the adjusted mean costs for primary care consultations, medications and secondary care visits, which were similar between the arms. Disaggregated data for unadjusted EQ-5D-5L domain scores, utility scores and total QALYs are reported in Table 34. Overall, mean unadjusted domain scores, utility scores and QALYs were similar across both groups and all time points. Intervention costs for the intervention arm equated to £39 per patient, excluding intervention training costs reduced this to £31 per patient.
| Resource category | Measurement | Intervention arm (N = 524) | Usual-care arm (N = 553) | ||||
|---|---|---|---|---|---|---|---|
| n | Resource use, mean (SD) | Cost (£), mean (SD) | n | Resource use, mean (SD) | Cost (£), mean (SD) | ||
| GP consultations, all types | Number of consultations | 478 | 4.60 (4.67) | 145.86 (162.62) | 544 | 5.00 (5.68) | 158.17 (180.05) |
| Practice nurse consultations, all types | Number of consultations | 478 | 2.78 (3.64) | 28.77 (38.58) | 544 | 3.61 (5.75) | 38.41 (58.48) |
| HCA consultations, all types | Number of consultations | 478 | 1.16 (2.05) | 7.70 (13.67) | 544 | 0.59 (2.07) | 3.91 (13.82) |
| Pharmacist consultations, all types | Number of consultations | 478 | 0.15 (0.54) | 1.60 (5.59) | 544 | 0.15 (0.90) | 1.36 (7.18) |
| Total primary care consultations | Number of consultations | 478 | 8.69 (7.57) | 183.93 (182.04) | 544 | 9.35 (10.73) | 201.84 (219.49) |
| Tamsulosin | Items prescribed | 478 | 2.29 (4.02) | 15.37 (25.04) | 544 | 3.06 (4.59) | 18.61 (28.09) |
| Solifenacin | Items prescribed | 478 | 0.35 (1.68) | 12.66 (60.50) | 544 | 0.24 (1.47) | 7.65 (47.75) |
| Mirabegron | Items prescribed | 478 | 0.21 (1.35) | 9.39 (61.77) | 544 | 0.18 (1.31) | 7.95 (60.75) |
| Other LUTS medications | Items prescribed | 478 | 0.42 (1.74) | 8.35 (48.55) | 544 | 0.81 (2.96) | 8.00 (40.38) |
| Total medications | Items prescribed | 478 | 3.27 (4.74) | 45.77 (100.50) | 544 | 4.29 (5.75) | 42.22 (98.97) |
| NHS 111 encounters | Number of calls | 459 | 0 | 0.00 | 479 | 0.01 (0.14) | 0.08 (1.72) |
| Outpatient visits | Number of visits | 459 | 0.07 (0.34) | 8.04 (36.75) | 479 | 0.08 (0.37) | 8.96 (40.82) |
| Outpatient procedures | Number of procedures | 459 | 0.03 (0.20) | 4.00 (27.78) | 479 | 0.05 (0.31) | 6.43 (39.65) |
| Day case | Number of cases | 459 | 0 | 0.00 | 479 | 0 | 0.00 |
| Inpatient stay | Number of stays | 459 | 0 | 0.00 | 479 | 0.00 (0.09) | 7.10 (155.49) |
| A&E visits | Number of visits | 459 | 0.00 (0.09) | 0.69 (14.84) | 479 | 0.01 (0.08) | 0.99 (12.56) |
| Total secondary care visits | Number of visits/encounters | 459 | 0.11 (0.46) | 12.73 (54.71) | 479 | 0.15 (0.66) | 23.56 (171.33) |
| Total intervention delivery costs | 524 | 31.48 (10.16) | – | 0.00 | |||
| Total intervention training costs | 524 | 7.66 | – | 0.00 | |||
| Total intervention costs | 524 | 39.14 (10.16) | – | 0.00 | |||
| Total NHS costs (unadjusted) | 429 | 272.37 (189.06) | 473 | 275.63 (321.36) | |||
| Total adjusted primary care consultation costsa | 478 | 174.48 (£142.80 to £206.16) | 544 | 205.61 (£166.05 to £245.18) | |||
| Total adjusted medication costsa | 478 | 40.49 (£28.28 to £52.71) | 544 | 46.39 (£34.90 to £57.88) | |||
| Total adjusted secondary care consultation costa,b | 459 | 13.78 (£0.37 to £27.19) | 479 | 25.83 (£0.06 to £51.59) | |||
| Domain | Baseline | 6 months | 12 months | Total QALYs | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Intervention arm | Usual-care arm | Intervention arm | Usual-care arm | Intervention arm | Usual-care arm | Intervention arm | Usual-care arm | |||||||||
| Mean (SD) | n | Mean (SD) | n | Mean (SD) | n | Mean (SD) | n | Mean (SD) | n | Mean (SD) | n | Mean (SD) | n | Mean (SD) | n | |
| Mobilitya | 1.42 (0.77) | 523 | 1.38 (0.74) | 551 | 1.37 (0.72) | 486 | 1.38 (0.73) | 513 | 1.38 (0.74) | 461 | 1.41 (0.76) | 485 | ||||
| Self-carea | 1.09 (0.37) | 523 | 1.09 (0.38) | 551 | 1.09 (0.37) | 486 | 1.10 (0.41) | 512 | 1.08 (0.36) | 460 | 1.08 (0.39) | 484 | ||||
| Usual activitya | 1.42 (0.73) | 523 | 1.39 (0.70) | 550 | 1.38 (0.70) | 550 | 1.35 (0.70) | 511 | 1.42 (0.77) | 460 | 1.41 (0.72) | 484 | ||||
| Pain/discomforta | 1.69 (0.82) | 523 | 1.71 (0.80) | 549 | 1.65 (0.78) | 486 | 1.66 (0.80) | 513 | 1.70 (0.80) | 459 | 1.65 (0.78) | 484 | ||||
| Anxiety/depressiona | 1.40 (0.65) | 522 | 1.41 (0.67) | 549 | 1.38 (0.66) | 484 | 1.40 (0.65) | 512 | 1.37 (0.63) | 459 | 1.46 (0.70) | 485 | ||||
| Utility scoreb | 0.83 (0.17) | 522 | 0.83 (0.16) | 547 | 0.84 (0.17) | 483 | 0.84 (0.16) | 509 | 0.83 (0.17) | 457 | 0.83 (0.17) | 482 | ||||
| Total QALYs | 0.84 (0.16) | 444 | 0.84 (0.15) | 460 | ||||||||||||
The adjusted mean incremental differences in costs and QALYs were similar between the two arms. Compared with the usual-care arm, the intervention arm had slightly lower mean costs (adjusted mean difference of –£29.99, 95% CI –£109.84 to £22.63) and a small gain in QALYs (adjusted mean difference of 0.001, 95% CI –0.011 to 0.014). At a threshold of £20,000 per QALY the INMB was £48.01 (95% CI –£225.83 to £321.85), suggesting that the intervention is cost-effective, but there is uncertainty around this estimate, which is also reflected in the intervention having only a 63% probability of being cost-effective at this threshold (Figures 6 and 7).
FIGURE 6 Cost-effectiveness plane from an NHS perspective. (Figure not reproduced here; see the original article.)
FIGURE 7 Cost-effectiveness acceptability curve from an NHS perspective. (Figure not reproduced here; see the original article.)
The sensitivity analysis (Table 35) applied to the complete-case population shows similar results, with a small positive INMB and wide CIs. Two of the sensitivity analyses that addressed missing data (multiple imputation, and zero costs and higher QALY imputation for missing questionnaire data) led to small negative INMBs and wide CIs. The sensitivity analysis considering the impact of using the highest costs and a lower QALY value for missing questionnaire data led to a positive INMB and wide CIs. It is of note that removing the intervention costs from the analysis led to a significant reduction in cost (–£70, 95% CI –£154 to –£18).
| Trial arm | n | Adjusted,a mean (95% CIb) | Incremental adjusted, mean (95% CIb) | ICER (£ per QALY) | INMB (£) at £20,000 per QALY (95% CI) | ||
|---|---|---|---|---|---|---|---|
| Costs (£) | QALYs | Costs (£) | QALYs | ||||
| Base case: complete-case analysis with MLM | |||||||
| Intervention | 413 | 253.53 (215.85 to 291.22) | 0.84 (0.83 to 0.84) | –29.99 (–109.84 to 22.63) | 0.001 (–0.011 to 0.014) | Intervention dominates usual care | 48.01 (–225.83 to 321.85) |
| Usual care | 453 | 283.52 (235.40 to 331.64) | 0.84 (0.83 to 0.84) | ||||
| Alternative analysis: complete-case analysis with SUR | |||||||
| Intervention | 413 | 258.05 (229.90 to 286.21) | 0.84 (0.83 to 0.84) | –20.00 (–62.50 to 22.50) | 0.001 (–0.010 to 0.011) | Intervention dominates usual care | 35.54 (–182.20 to 253.28) |
| Usual care | 453 | 278.05 (251.42 to 304.68) | 0.84 (0.83 to 0.84) | ||||
| Multiple imputation with SUR | |||||||
| Intervention | 524 | 275.03 (248.63 to 301.43) | 0.83 (0.83 to 0.84) | 1.42 (–38.61 to 41.46) | –0.002 (–0.012 to 0.009) | Usual care dominates intervention | –35.42 (–248.71 to 177.87) |
| Usual care | 553 | 273.61 (248.26 to 298.96) | 0.84 (0.83 to 0.84) | ||||
| Complete-case analysis with MLM: applying unit cost for face-to-face contacts used for telephone contacts | |||||||
| Intervention | 413 | 262.52 (222.73 to 302.32) | 0.84 (0.83 to 0.84) | –34.81 (–114.27 to 19.65) | 0.001 (–0.011 to 0.014) | Intervention dominates usual care | 52.83 (–222.33 to 327.99) |
| Usual care | 453 | 297.33 (248.37 to 346.29) | 0.84 (0.83 to 0.84) | ||||
| Complete-case analysis with MLM: excluding participants whose secondary care data are based on letters | |||||||
| Intervention | 399 | 249.83 (212.76 to 286.90) | 0.84 (0.83 to 0.84) | –26.68 (–96.34 to 25.11) | 0.001 (–0.011 to 0.014) | Intervention dominates usual care | 30.59 (–234.61 to 295.79) |
| Usual care | 434 | 276.52 (229.62 to 323.42) | 0.84 (0.83 to 0.84) | ||||
| Complete-case analysis with MLM: medication quantities for SystmOne based on most frequently reported, rather than average | |||||||
| Intervention | 413 | 259.39 (219.57 to 299.21) | 0.84 (0.83 to 0.84) | –29.89 (–118.22 to 24.25) | 0.001 (–0.011 to 0.014) | Intervention dominates usual care | 47.92 (–226.82 to 322.66) |
| Usual care | 453 | 289.28 (239.48 to 339.08) | 0.84 (0.83 to 0.84) | ||||
| Complete-case analysis with MLM: removal of training costs | |||||||
| Intervention | 413 | 245.87 (208.19 to 283.56) | 0.84 (0.83 to 0.84) | –37.65 (–117.50 to 14.97) | 0.001 (–0.011 to 0.014) | Intervention dominates usual care | 55.67 (–218.17 to 329.51) |
| Usual care | 453 | 283.52 (235.40 to 331.64) | 0.84 (0.83 to 0.84) | ||||
| Complete-case analysis with MLM: removal of all intervention costs | |||||||
| Intervention | 413 | 213.55 (176.34 to 250.77) | 0.84 (0.83 to 0.84) | –70.28 (–153.84 to –18.45) | 0.001 (–0.011 to 0.014) | Intervention dominates usual care | 88.31 (–185.50 to 362.12) |
| Usual care | 453 | 283.84 (235.59 to 332.09) | 0.84 (0.83 to 0.84) | ||||
| Complete-case analysis with MLM: adjusting for 12-month pre-consent GP consultation visits | |||||||
| Intervention | 413 | 277.53 (245.93 to 309.14) | 0.84 (0.83 to 0.84) | –3.02 (–49.52 to 44.02) | 0.001 (–0.011 to 0.014) | Intervention dominates usual care | 21.50 (–246.93 to 289.03) |
| Usual care | 453 | 280.56 (244.16 to 316.96) | 0.84 (0.83 to 0.84) | ||||
| MLM: simple imputation of £0 for secondary care per arm and a QALY value 10% higher than the mean QALY value per arm | |||||||
| Intervention | 456 | 254.12 (216.02 to 292.21) | 0.84 (0.83 to 0.85) | –20.19 (–91.44 to 30.56) | –0.006 (–0.020 to 0.008) | Intervention less costly, but fewer QALYs | –96.06 (–394.16 to 202.04) |
| Usual care | 516 | 274.31 (231.20 to 317.42) | 0.85 (0.84 to 0.86) | ||||
| MLM: simple imputation of the highest participant cost for secondary care per arm and a QALY value 10% lower than the mean QALY value per arm | |||||||
| Intervention | 456 | 330.66 (273.59 to 387.73) | 0.82 (0.82 to 0.83) | –322.19 (–435.12 to –232.31) | –0.004 (–0.015 to 0.010) | Intervention less costly, but fewer QALYs | 243.74 (–39.00 to 526.48) |
| Usual care | 516 | 652.85 (582.44 to 723.25) | 0.83 (0.82 to 0.84) | ||||
| Complete-case analysis with MLM: subgroup analysis for participants completing 12 months’ follow-up before 11 March 2020 | |||||||
| Intervention | 236 | 255.12 (133.88 to 376.35) | 0.83 (0.81 to 0.84) | –33.04 (–209.94 to 438.39) | –0.001 (–0.027 to 0.010) | Intervention less costly, but fewer QALYs | –155.66 (–730.36 to 419.03) |
| Usual care | 169 | 288.16 (23.20 to 553.11) | 0.84 (0.82 to 0.85) | ||||
| Complete-case analysis with MLM: subgroup analysis for participants completing 12 months’ follow-up from 11 March 2020 | |||||||
| Intervention | 177 | 229.31 (119.77 to 338.85) | 0.86 (0.81 to 0.90) | –67.77 (–219.86 to 74.92) | 0.029 (0.002 to 0.133) | Intervention dominates usual care | 640.30 (–1445.07 to 2726.67) |
| Usual care | 284 | 297.07 (204.54 to 389.61) | 0.83 (0.76 to 0.90) | ||||
A secondary analysis comparing difference in mean IPSS (see Table 14) and mean costs (see Table 35) found that costs were lower and improvement in IPSS was greater in the intervention arm, indicating that the intervention was dominant; therefore, no incremental cost-effectiveness ratios were created.
Just over half (57.1%, n = 236) of participants from the intervention arm and just over one-third (37.3%, n = 169) of usual-care participants completed 12 months’ follow-up before 11 March 2020 (see Table 35). Costs in each arm were similar for this subgroup compared with the base-case analysis, whereas QALYs were slightly lower in the intervention arm than in the usual care when compared with the base-case analysis, resulting in an INMB of –£156 (95% CI –£730 to £419). Among those participants who completed 12 months’ follow-up on or after 11 March 2020, costs were slightly lower in the intervention arm, and slightly higher in the the usual-care arm, than in the base-case analysis. The intervention arm also had a higher number of QALYs, which result in an overall higher INMB of £640 (95% CI –£1445 to £2727).
Discussion
From an NHS perspective, the mean costs and QALYS were similar across both trial arms. A slightly lower mean cost in the intervention arm (a £30 cost difference) led to a small positive INMB at the NICE willingness-to-pay threshold of £20,000 per QALY, and a 63% probability of the intervention being the cost-effective option. There is therefore some uncertainty in the finding of the intervention being cost-effective at the NICE £20,000 per QALY threshold. With the exception of two sensitivity analyses, both alternative methods for dealing with missing data, the initial results were robust.
This is the first economic evaluation carried out on a definitive trial in which the value for money of a manualised and standardised non-pharmacological intervention, compared with usual care, for men experiencing bothersome LUTS has been evaluated.
A key strength of this trial was the use of routine data, which minimised the number of missing data and the burden on participants. Participants typically struggle to recall and name the type of HCPs they have engaged with.42 At each practice, EMR data were available from one of England’s two main general practice IT systems (EMIS Web or SystmOne). There were, however, differences between the IT systems: SystmOne was unable to extract data on the reason for a consultation and the quantities of medication prescribed. Consequently, a standardised approach was applied to reduce the potential for bias that may arise from using a different IT system. First, all consultations were included in the analysis and, second, medication quantities were based on the average quantities prescribed for the EMIS Web cohort. The IT system was also adjusted for in the regression analysis.
It was not possible to retrieve EMR data for any participants from one of the sites allocated to the intervention arm, which provided 5.5% (n = 29) of the intervention participants. The site attributed this to time constraints. Similarly, owing to time constraints at the general practices, it was not possible to request secondary care letters for all participants for whom data were missing. Prioritising participants with a urology referral in their EMR data was deemed a pragmatic approach. It was assumed that no secondary care resources were consumed if (1) no urology referral was recorded in the patient’s EMR data set and (2) the patient had answered other questions in the follow-up questionnaire booklet.
Future cluster randomised studies using EMR data to capture healthcare resource use should consider stratifying for the general practice IT system within the randomisation process, as well as controlling for this variable, as seen in this trial. Differences within the general practice IT systems, if arms are imbalanced, could lead to bias if there are differences in the number of resource use data available within each system. In this trial, a standardised approach was applied to reduce the impact of the IT systems; this meant that it was not possible to mask the second health economist, as the cleaning of the EMR data required the removal of intervention-related consultations which had already been captured in the case report form. Nevertheless, the lead health economist was masked to the randomisation allocation until all data had been cleaned and valued.
The primary analysis was a complete-case analysis that included 80.4% (n = 866) of all participants. Data were likely to be missing completely at random for the 3.5% (n = 38) of participants for whom EMR data or secondary care letters were not returned by sites or who had changed practice within the trial time frame. In the case of the remaining participants who were excluded from the complete-case analysis (16.1%, n = 173), all or some questionnaire data were missing. It is therefore possible that the complete-case analysis excluded participants who had better or worse outcomes or used more or fewer resources than the participants for whom we have observed data. Although the mechanism of missingness was assessed and the explanatory variables indicated that missingness was at random, it is not possible to rule out the mechanism ‘missing not at random’.
It may be the case that participants whose 12-month follow-up was after 11 March 2020 used less healthcare resource because of the impact of the COVID-19 pandemic on primary and secondary care services in the UK. This is important to consider given that the proportion of participants reaching their 12-month follow-up during the pandemic was one-fifth (19.8%) higher in the usual-care arm than in the intervention arm. That is to say, it could be possible that resource use was underestimated in the usual-care arm; however, the costs in this arm were higher for those participants whose 12-month follow-up occurred during the pandemic, and it is unclear why this is the case.
A further potential limitation is that costs and outcomes have been evaluated over only a short time horizon. It was beyond the scope of this work to compare the arms over a longer time horizon.
The cost-effectiveness analysis showed that costs and outcomes were similar in both arms. The removal of the intervention costs from the analysis (mean cost per person of £39) led to a significant difference in costs of £70 in favour of the intervention arm, indicating that, if the costs of implementing the intervention could be reduced, then the roll-out of this intervention could potentially lead to cost savings for the NHS.