Surgical treatments for women with stress urinary incontinence: the ESTER systematic review and economic evaluation

Types of studies

Randomised controlled trials or quasi-RCTs (using alternate allocation) were eligible for the assessment of clinical effectiveness. There was no restriction on the trials’ publication status (published or unpublished) or on the year or the language in which they were reported.

Population

The type of participants considered for this assessment were adult women with SUI or stress-predominant MUI. Either a clinical or a urodynamic diagnosis of SUI was considered suitable. We accepted the diagnoses as defined by the authors of the included trials. Women who underwent continence surgery as a primary or repeated procedure, women with SUI and concomitant prolapse and women who received concomitant prolapse surgery were deemed suitable for inclusion. Studies that did not specify the type of incontinence (stress, urge, mixed) or the predominant MUI symptoms (stress or urge) were excluded.

Interventions

The interventions detailed below were deemed suitable for inclusion. There are a number of variations on most surgical techniques, but, essentially, they fall into nine distinct categories, each of which has been previously evaluated in systematic reviews published in The Cochrane Library:

1. retropubic MUS operations

2. transobturator MUS operations

3. open retropubic colposuspension

4. laparoscopic colposuspension

5. traditional suburethral sling procedures

6. single-incision sling operations

7. anterior vaginal repair

8. bladder neck needle suspension

9. urethral injection therapy (periurethral injections/injectable bulking agents).

To be eligible for inclusion, studies had to compare two or more of the surgical interventions listed above. Studies that compared a surgical intervention with PFMT were also considered suitable for inclusion as they were useful for the development of a NMA treatment diagram for assessing direct and indirect treatment comparisons. For the purpose of this assessment, we considered eligible for inclusion PFMT programmes taught and monitored by health professionals, clinicians or instructors (defined as ‘supervised PFMT’). 22 PFMT programmes for which instructions were delivered using written material only (e.g. leaflets or flyers delivered to women), without any interaction or face-to-face contact with health-care professionals, were considered to be equivalent to no treatment and, therefore, excluded from this assessment.

We excluded, post hoc, studies that compared surgery with pharmacological treatments or no treatment, as they were not considered useful for the development of the treatment network diagram. We did not include studies comparing specific technical variations of the relevant surgical techniques (e.g. inside-out vs. outside-in transobturator MUS operations, or one injectable agent vs. another) as it was beyond the scope of this review to assess specific technical approaches within each surgical category.

Outcomes

The following outcome measures were considered.

Selection of studies

The selection of studies included in the Cochrane systematic reviews was established by one reviewer (SW) and checked by a second reviewer (MI). Screening of the titles and abstracts of all citations identified by the updated literature searches was conducted by one reviewer (SW). Full text copies of all potentially relevant reports were retrieved by the same reviewer for eligibility and checked by a second reviewer (MI or MS) to ensure that they met the prespecified inclusion criteria.

Data extraction

Studies selected from Cochrane systematic reviews

We relied on the original risk-of-bias assessments performed by the authors of the Cochrane systematic reviews. Assessment criteria varied across different Cochrane systematic reviews, which were accepted as reported. We updated the risk-of-bias assessments using the new criteria described below only for those studies that were initially included in the Cochrane systematic reviews as abstracts but subsequently published in full (eight studies). 25–32

Studies identified by the updated literature searches

For the studies identified by the updated literature searches, the Cochrane risk-of-bias assessment tool was used. 15 Critical judgments were made for the following domains: selection bias (random sequence generation, allocation concealment), performance bias (blinding of participants and personnel), detection bias (blinding of outcome assessor), attrition bias (incomplete outcome data), reporting bias (free of selective reporting) and other bias. Each included study was judged to be at ‘low risk of bias’, ‘high risk of bias’ or ‘unclear risk of bias’. The assessment was conducted by one reviewer (MI or MS) and checked by a second reviewer (MI or MS). Any disagreements were resolved by consensus or arbitration with a third reviewer (MB). Studies were not included or excluded on the basis of their methodological quality.

Network meta-analysis

Network meta-analysis aims to synthesise all of the available evidence within a distinct framework. It enables the integration of direct evidence with indirect evidence from a network of studies involving all possible intervention comparisons. Direct evidence is obtained from all possible head-to-head comparisons between the different interventions, whereas indirect evidence is obtained from comparisons of two or more interventions that share a common comparator. For example, direct evidence from comparing intervention A with B and B with C can be combined to provide indirect evidence for intervention A versus C.

We conducted a NMA to assess the effects of the surgical interventions for SUI in terms of the number of women who were cured or experienced an improvement in their incontinence symptoms (primary outcomes). Studies that reported 100% events in all arms (all participants were cured or improved) were excluded from the analysis as they provide no evidence for the analysis. The NMA included outcomes measured at 12 months or at a time point closest to 12 months (studies with a time point of < 2 weeks or > 36 months were excluded).

Convergence was assessed using Brooks–Gelman–Rubin plots, trace plots and autocorrelation plots. The NMA analysis was undertaken within a Bayesian framework, using WinBUGS 1.4.3 (MRC Biostatistics Unit, Cambridge, UK)33 and using code provided by Dias et al. 34 (see Appendix 3). The analysis used vague uniform prior and adjusted for multi-arm trials using conditional distributions. Consistency, which is where direct and indirect evidence are in agreement, was assessed by comparing the individual data point’s posterior mean deviance contributions for the consistency and inconsistency model and node splitting analysis35 using R 3.4.1 (R Foundation, Vienna, Austria).

In the results section, network diagrams are presented for each outcome and effect sizes are reported as posterior median odds ratios (ORs) and 95% credible intervals (Crls). We also present rankograms for all surgical interventions, which gives probabilities of an intervention being ranked 1 (the highest) to 9 (the lowest) as well as reporting the surface under the cumulative ranking curve (SUCRA), which is a numerical presentation of the overall ranking and presents a single number associated with each intervention. SUCRA values range from 0% to 100%. The closer to 100% the SUCRA value, the more likely that an intervention is in the top rank or one of the top ranks; the closer to 0 the SUCRA value, the more likely that an intervention is in the bottom rank or one of the bottom ranks. 36

Direct pairwise (head-to-head) meta-analyses

Adverse event outcomes as well as cure and improvement were summarised as direct head-to-head comparisons using a random effects model. Effect sizes are reported as ORs with 95% confidence intervals (CIs) for dichotomous outcomes and as mean differences with 95% CI for continuous outcomes. When data were available, meta-analyses assessing adverse events were performed at different time points (e.g. 6 months, 12 months, 24 months). Heterogeneity was assessed using the I² statistic. Analyses were performed using Stata version 14 (StataCorp LP, College Station, TX, USA).

Subgroup analysis

We planned to undertake subgroup analyses for the following groups: women with stress-predominant MUI versus women with SUI alone; repeated surgery (after failed previous continence surgery) versus primary procedures; women with and without coexisting vaginal prolapse/having concomitant prolapse surgery. However, these were not performed owing to the lack of available data. For example, lack of clarity in some trial reports made it difficult to decipher whether study participants had SUI, MUI or both. Moreover, outcome data for each patient subgroup were not reported separately in all trials. Owing to such inconsistencies and limited reporting of suitable data in the included studies, we were unable to perform these analyses.

Studies identified from Cochrane systematic reviews

From the eight published Cochrane systematic reviews that provided data suitable for this assessment, a total of 166 studies (in 443 reports) were initially selected as potentially relevant. Of these, 147 studies (in 406 reports) were included as they met the prespecified inclusion criteria for this assessment, 17 studies were excluded and one was merged with another study. 37 The number of included studies from each Cochrane systematic review is shown in Table 1.

Studies identified from updated literature searches

The updated literature searches identified 591 potentially relevant titles and abstracts from which 216 reports were selected for full-text evaluation. Of these, 65 reports (describing 28 new studies) were considered suitable for inclusion. In addition, 100 reports were identified as additional references of studies already included in the relevant Cochrane systematic reviews and two reports were additional references of studies originally included in the Cochrane systematic reviews but subsequently excluded from the effectiveness and cost-effectiveness of surgical treatments for women with stress urinary incontinence (ESTER) report. A total of 26 reports of 23 ongoing trials were also identified and two further reports were included as ‘awaiting assessment’ and 21 reports were excluded.

In total, 175 studies were included in the assessment of clinical effectiveness. Of these, 43 studies were available only as conference abstracts. 38–80 A PRISMA flow chart summarising the study selection process is provided in Appendix 4. The list of included studies and associated references is reported in Appendix 5. Reasons for exclusion for a sample of excluded studies are described in Appendix 6. The list of ongoing trials is provided in Appendix 7.

Selection bias (adequate sequence generation/allocation concealment)

For all 175 included studies, risk of selection bias was assessed in two domains: random sequence generation and allocation concealment. A total of 43 studies (out of 175; 25%) reporting both adequate sequence generation and adequate allocation concealment were considered to have an overall low risk of selection bias. 25,27–29,32,51,63,64,83,86,87,92,93,96,97,100,101,116,120,122,124–126,128,135–137,140,144,147,148,150,157,159,160,167,174–176,182,183,185 A further six studies (out of 175; 3%) reported adequate allocation concealment but they did not describe the method used for random sequence generation. 77,85,127,156,173,199 A total of 40 studies (out of 175; 23%) reported adequate sequence generation but did not provide information on allocation concealment. 26,30,31,37,41,52,60,65,72,79,82,88,89,91,94,99,101,114,119,129,130,138,141,143,146,162,164,172,178,180,181,186–189,194–196,198,203 Two studies (out of 175; 1%) with adequate random sequence generation specifically mentioned that the treatment allocations were not concealed. 95,193

There were eight quasi-randomised trials (out of 175 included studies; 5%) in which treatment allocation was based on a method of alternation103–107,109,110 or on date of birth. 111 These were considered to be at high risk of selection bias in terms of both allocation sequence generation and concealment of allocation. Two further studies (out of 175; 1%) allocated participants by alternation108 or by medical history numbers112 and were considered to be at high risk of selection bias, although allocation concealment was considered to be at unclear (rather than high) risk of bias by the authors of the Cochrane systematic reviews. 5,9 The remaining studies (74 out of 175; 42%) did not provide this information.

Performance and detection bias (blinding)

Blinding of patients and personnel provides a safeguard against performance bias and blinding of outcome assessors protects against detection bias. It is worth noting that, owing to the nature of the interventions, blinding of participants and personnel, especially the surgeon performing the operation, is not possible. Blinding of outcome assessment for patient-reported outcomes would similarly be difficult in this clinical context (with unblinded patients being the assessors), although it should be possible to blind health-care professionals who assess clinical outcomes.

In seven studies identified from the Cochrane systematic reviews assessing traditional slings7 and bladder neck needle suspension,11 risk of performance bias and detection bias was assessed using a single criterion (whether or not a lack of blinding could introduce performance and detection bias). One study (out of 7; 14%) was judged to be at low risk of bias92 and the other six studies (86%) at high or unclear risk of bias. 38,81,102,166,198,199

Risk of performance and detection bias was assessed separately in the 168 included studies identified from sources other than the two Cochrane reviews mentioned above. With respect to the assessment of whether or not lack of blinding of patients and personnel could introduce performance bias, eight studies (out of 168; 5%) were judged to be at low risk of bias87,89,117,128,129,151,157,191 and the other 160 studies (95%) were judged to be at high or unclear risk of bias. 25–32,37,39–80,82–86,88,90,91,93–101,103–114,116,118–127,130–150,152–156,158–165,167–190,192–197,201–203

Risk of detection bias in the 132 studies identified from Cochrane systematic reviews on MUS,8,18 open colposuspension,5 laparoscopic colposuspension6 and single-incision slings9 was assessed in terms of whether or not lack of blinding of outcome assessors could affect results. A total of 20 studies (out of 132; 15%) were judged to be at low risk of bias37,86,87,89,94,117,122,125,127,130,136,141,142,144,148,156,174,178,191,204 and the remaining 112 studies (85%) were judged to be at high or unclear risk of bias. 39–42,44–47,49,51,53–61,63,64,67–71,73,74,76–78,80,82–85,88,90,93,96–100,103–112,116,118–120,123,124,126,128,129,131–135,137–140,145–147,149–155,157,160–163,165,167–170,172,173,180,182–190,192–196,201–203

Detection bias in the 36 studies identified from other sources (e.g. new studies from the update literature searches) was assessed on a per outcome basis in terms of whether or not lack of blinding of outcome assessors could affect effect estimates of patient-reported outcomes such as subjective perception of cure of UI (where outcomes are assessed directly by women) and of clinician-measured outcomes such as urodynamic tests (where outcomes are assessed by care providers). One study (out of 36; 3%) was judged to be at low risk for patient-reported outcomes,159 five studies (14%) at low risk for clinician-measured outcomes,25,115,143,175,181 and three studies (8%) at low risk for both patient-reported and clinician-measured outcomes. 29,32,91 The remaining 27 studies (75%) were judged to be at high or unclear risk for either outcome. 26–28,30,31,43,48,50,52,62,65,66,72,75,79,95,101,113,114,121,158,164,171,176,177,179,197

Attrition bias (incomplete outcome data)

In 139 studies included in the relevant Cochrane systematic reviews, risk of bias was assessed for all outcomes concerned. A total of 54 were judged to be at low risk of attrition bias, with missing data either balanced between treatment groups or imputed using appropriate methods, or with no missing data;37,40,41,54,67,81–84,88,92–94,96–98,100,102,106,116,119,120,122,123,126–128,130,133,135–137,139,141,142,144,147–150,155,157,160,162,166,168,170,174,178,184,185,193,198,199 nine were judged to be at high risk of bias;86,87,112,140,156,172,173,180,189 and 76 were judged to be at unclear risk of bias. 38,39,42,44–47,49,51,53,55–61,63,64,68–71,73,74,76–78,80,85,89,90,99,103–105,107–111,116–118,124,125,129,131,132,134,138,145,146,151–154,161,163,165,167,169,182,183,186–188,190–192,194–196,201–203

In 36 studies identified from other sources (e.g. new studies from the updated literature searches), risk of attrition bias was assessed on a per outcome basis in terms of patient-reported outcomes and clinician-measured outcomes. Of these, 18 studies were judged to be at low risk for both patient-reported and clinician-measured outcomes,25,27,30,52,91,95,101,113,114,121,143,159,164,171,176,177,179,181 three studies were judged to be at low risk for clinician-measured outcomes but unclear for patient-reported outcomes,66,72,75 two studies were judged to be at high risk for both patient-reported and clinician-measured outcomes29,175 and the remaining 13 studies were judged to be at unclear risk of bias for both outcomes. 26,28,31,32,43,48,50,62,65,79,115,158,197

Reporting bias (free of selective reporting)

Few of the relevant Cochrane systematic reviews assessed the risk of reporting bias. Information was available from one study identified from the Cochrane systematic review assessing laparoscopic colposuspension6 and from 36 further studies (28 new studies and the eight studies originally included as abstracts in the Cochrane systematic reviews and subsequently published in full). Of these, 24 studies (out of 37; 65%) were judged to be at low risk of reporting bias on the basis that all outcomes specified in the methods section were reported in the results section of the study report,25,27–32,52,62,65,79,95,101,113–115,143,159,171,175,177,179,181,197 six studies (16%) were judged to be at high risk of bias on the basis that they did not report the results for all the outcomes specified in the method section of the study report,26,43,72,75,91,176 and the remaining seven studies (19%) did not provide sufficient information to formulate a judgement. 48,50,66,121,158,163,164

Other sources of bias

Information on which to assess ‘other sources of bias’ was available for 82 of the included studies. In these studies, there was no clear evidence that other sources of bias were present. 25–32,40,43,46–53,55,57,60,62,65,66,68,69,72,75–79,82,85,88,89,91,93,95,100,101,103,105,108,109,111,113–115,121,143,145,158,159,161–164,171,175,177,179,181–195,197,200–203

Network meta-analysis of primary outcomes: number of women cured and number of women who experienced an improvement in their incontinence symptoms

Technical information about the model

The NMA included eight surgical procedures for SUI:

• retropubic MUS operations

• transobturator MUS operations

• open colposuspension

• laparoscopic colposuspension

• traditional sling operations

• single-incision sling operations

• bladder neck needle suspension

• anterior vaginal repair.

The ninth procedure, urethral injection therapy, did not add any information to the network and was excluded from analysis (for the cure outcome analysis, urethral injection therapy was only connected to one intervention). Separate models were developed for the two primary outcomes: the number of women cured (defined as resolution of symptoms) and the number of women improved (defined as women experiencing an improvement in their incontinence symptoms, including cure). Figure 1 shows the network diagrams for the number of women cured and improved, respectively. The size of the circles reflects the number of participants and the line width reflects the number of direct comparisons. Figure 1 also includes the total number of women for each surgical intervention included in the NMA.

FIGURE 1.

Network plot for (a) the number of women cured; and (b) the number of women improved. Note: circle size reflects the number of participants; line width reflects the number of direct comparisons. Ant repair, anterior vaginal repair; bladder neck needle, bladder neck needle suspension; open colpo, open colposuspension; lap colpo, laparoscopic colposuspension; retro-MUS, retropubic MUS; single incision, single-incision sling; trad sling, traditional sling; transob-MUS, transobturator MUS.

Cure or improvement of incontinence symptoms was assessed subjectively (by the woman), objectively (by a clinician) or by a combination of subjective and objective measures (composite measure). When more than one of these outcome measures were reported within a study, only one measure was included in the NMA. This measure was selected in the following order: women’s self-report assessment was given priority and, if this was unavailable, the composite outcome measure followed by the objective outcome measure were chosen as a proxy. The NMA included outcomes measured at 12 months or at a time point closest to 12 months.

The number of studies providing direct head-to-head (pairwise) evidence for each intervention that contributed data to the NMA is shown in Tables 5 and 6. The searches identified 125 trials that reported cure or improvement, or both. Three trials67,76,105 were excluded from the cure data set and four trials67,76,105,124 from the improvement dataset because they reported ‘100% events’ in all treatment arms (i.e. all participants in the study were cured or improved), providing no information for the NMA. Thus, 105 trials contributed to the NMA for assessing the number of women cured, and 120 trials were included in the analysis assessing the number of women improved. For ‘cure’, the analysis included four three-arm trials and 101 two-arm trials and generated a total of 17 direct comparisons, whereas for ‘improvement’ the analysis included five three-arm trials and 115 two-arm trials and generated a total of 18 direct comparisons. Appendix 10 shows the total number of trials for each treatment comparison for the assessment of both primary outcomes.

TABLE 5 - Results for number of women cured

Ant repair, anterior vaginal repair; bladder neck needle, bladder neck needle suspension; injectable, urethral injection therapy; open colpo, open colposuspension; lap colpo, laparoscopic colposuspension; N/A, not applicable; retro-MUS, retropubic MUS; single incision, single-incision sling; trad sling, traditional sling; transob-MUS, transobturator MUS.

An OR of > 1 favours the first treatment, i.e. more events (cure) occur; an OR of < 1 favours the second treatment, i.e. fewer events occur.

These analyses are also informed by three-arm trials, including one comparing retropubic MUS, transobturator MUS and single-incision sling, one comparing retropubic MUS, open colposuspension and traditional sling, and two comparing open colposuspension, bladder neck needle and anterior vaginal repair.

Treatment		Direct evidence			NMA		GRADE
1	2	Number of trials	ORa	95% CI	ORa	95% CrI
Transob-MUS	Retro-MUS	36b	0.83	0.71 to 0.97	0.74	0.59 to 0.92	Moderate
Open colpo	Retro-MUS	6b	0.95	0.68 to 1.32	0.85	0.54 to 1.33	Low
Lap colpo	Retro-MUS	2	0.40	0.11 to 1.45	0.58	0.31 to 1.05	Low
Trad sling	Retro-MUS	6b	0.87	0.58 to 1.29	1.06	0.62 to 1.85	Very low
Single incision	Retro-MUS	6b	0.42	0.20 to 0.87	0.50	0.36 to 0.70	Low
Bladder neck needle	Retro-MUS	N/A	N/A	N/A	0.34	0.15 to 0.75	Low
Ant repair	Retro-MUS	N/A	N/A	N/A	0.22	0.10 to 0.45	Very low
PFMT	Retro-MUS	N/A	N/A	N/A	0.12	0.04 to 0.32	Low
Open colpo	Transob-MUS	1	0.90	0.30 to 2.69	1.16	0.72 to 1.86	Low
Lap colpo	Transob-MUS	N/A	N/A	N/A	0.79	0.42 to 1.46	Low
Trad sling	Transob-MUS	1	2.00	0.17 to 23.96	1.44	0.81 to 2.62	Very low
Single incision	Transob-MUS	21b	0.74	0.54 to 1.00	0.68	0.51 to 0.91	Low
Bladder neck needle	Transob-MUS	N/A	N/A	N/A	0.46	0.21 to 1.02	Very low
Ant repair	Transob-MUS	1	0.50	0.15 to 1.62	0.30	0.14 to 0.62	Very low
PFMT	Transob-MUS	1	0.20	0.12 to 0.33	0.16	0.06 to 0.43	Low
Lap colpo	Open colpo	9	0.74	0.43 to 1.30	0.68	0.42 to 1.08	Low
Trad sling	Open colpo	3b	2.47	0.73 to 8.40	1.24	0.66 to 2.45	Very low
Single incision	Open colpo	N/A	N/A	N/A	0.59	0.34 to 1.01	Low
Bladder neck needle	Open colpo	3b	0.41	0.25 to 0.68	0.40	0.20 to 0.78	Low
Ant repair	Open colpo	3b	0.20	0.07 to 0.60	0.26	0.14 to 0.48	Very low
PFMT	Open colpo	1	0.08	0.01 to 0.51	0.14	0.05 to 0.39	Low
Trad sling	Lap colpo	N/A	N/A	N/A	1.83	0.86 to 4.04	Very low
Single incision	Lap colpo	N/A	N/A	N/A	0.87	0.44 to 1.70	Low
Bladder neck needle	Lap colpo	N/A	N/A	N/A	0.59	0.26 to 1.33	Very low
Ant repair	Lap colpo	N/A	N/A	N/A	0.38	0.18 to 0.82	Very low
PFMT	Lap colpo	N/A	N/A	N/A	0.21	0.07 to 0.63	N/A
Single incision	Trad sling	N/A	N/A	N/A	0.47	0.25 to 0.88	Very low
Bladder neck needle	Trad sling	1	1.00	0.05 to 18.57	0.32	0.13 to 0.79	Very low
Ant repair	Trad sling	N/A	N/A	N/A	0.21	0.09 to 0.49	Very low
PFMT	Trad sling	N/A	N/A	N/A	0.11	0.04 to 0.34	Very low
Bladder neck needle	Single incision	N/A	N/A	N/A	0.67	0.29 to 1.56	N/A
Ant repair	Single incision	N/A	N/A	N/A	0.44	0.20 to 0.96	Very low
PFMT	Single incision	N/A	N/A	N/A	0.24	0.08 to 0.65	Low
Ant repair	Bladder neck needle	1b	0.92	0.55 to 1.55	0.65	0.30 to 1.36	Very low
PFMT	Bladder neck needle	N/A	N/A	N/A	0.35	0.10 to 1.17	Low
PFMT	Ant repair	N/A	N/A	N/A	0.55	0.17 to 1.77	Very low

TABLE 6 - Results for number of women improved

Ant repair, anterior vaginal repair; bladder neck needle, bladder neck needle suspension; injectable, urethral injection therapy; open colpo, open colposuspension; lap colpo, laparoscopic colposuspension; N/A, not applicable; retro-MUS, retropubic MUS; single incision, single-incision sling; trad sling, traditional sling; transob-MUS, transobturator MUS.

An OR of > 1 favours the first treatment, i.e. more events (improvement) occur; an OR of < 1 favours the second treatment, i.e. fewer events occur.

These analyses are also informed by two three-arm trials comparing retropubic MUS, transobturator MUS and single-incision sling.

Treatment		Direct evidence			NMA		GRADE
1	2	Number of trials	ORa	95% CI	ORa	95% CrI
Transob-MUS	Retro-MUS	40b	0.86	0.70 to 1.06	0.76	0.59 to 0.98	Moderate
Open colpo	Retro-MUS	6b	0.83	0.55 to 1.24	0.65	0.41 to 1.02	Low
Lap colpo	Retro-MUS	4	0.49	0.18 to 1.35	0.52	0.29 to 0.91	Low
Trad sling	Retro-MUS	6b	0.62	0.38 to 1.02	0.69	0.39 to 1.26	Low
Single incision	Retro-MUS	6b	0.42	0.20 to 0.89	0.50	0.35 to 0.71	Moderate
Bladder neck needle	Retro-MUS	N/A	N/A	N/A	0.25	0.11 to 0.58	Low
Ant repair	Retro-MUS	N/A	N/A	N/A	0.18	0.08 to 0.39	Very low
PFMT	Retro-MUS	N/A	N/A	N/A	0.43	0.14 to 1.37	Low
Open colpo	Transob-MUS	1	0.90	0.30 to 2.69	0.85	0.52 to 1.41	Low
Lap colpo	Transob-MUS	N/A	N/A	N/A	0.69	0.37 to 1.26	Low
Trad sling	Transob-MUS	1	2.00	0.17 to 23.96	0.91	0.49 to 1.72	Very low
Single incision	Transob-MUS	28b	0.74	0.57 to 0.96	0.66	0.49 to 0.89	Moderate
Bladder neck needle	Transob-MUS	N/A	N/A	N/A	0.33	0.14 to 0.79	Very low
Ant repair	Transob-MUS	1	1.00	0.26 to 3.89	0.24	0.10 to 0.53	Very low
PFMT	Transob-MUS	1	0.18	0.10 to 0.33	0.56	0.19 to 1.78	Low
Lap colpo	Open colpo	9	0.93	0.58 to 1.48	0.81	0.49 to 1.31	Low
Trad sling	Open colpo	3b	2.47	0.73 to 8.40	1.07	0.54 to 2.15	Low
Single incision	Open colpo	N/A	N/A	N/A	0.78	0.44 to 1.36	Low
Bladder neck needle	Open colpo	3b	0.38	0.22 to 0.63	0.38	0.18 to 0.81	Low
Ant repair	Open colpo	3b	0.20	0.07 to 0.60	0.28	0.14 to 0.55	Very low
PFMT	Open colpo	1	8.87	1.66 to 47.25	0.66	0.21 to 2.16	Low
Trad sling	Lap colpo	N/A	N/A	N/A	1.32	0.62 to 2.98	Low
Single incision	Lap colpo	N/A	N/A	N/A	0.97	0.50 to 1.87	Low
Bladder neck needle	Lap colpo	N/A	N/A	N/A	0.47	0.20 to 1.17	Very low
Ant repair	Lap colpo	N/A	N/A	N/A	0.34	0.15 to 0.79	Very low
PFMT	Lap colpo	N/A	N/A	N/A	0.82	0.25 to 2.88	Very low
Single incision	Trad sling	1	1.92	0.65 to 5.64	0.73	0.37 to 1.39	Low
Bladder neck needle	Trad sling	1	1.00	0.05 to 18.57	0.36	0.13 to 0.95	Very low
Ant repair	Trad sling	N/A	N/A	N/A	0.26	0.10 to 0.65	Very low
PFMT	Trad sling	N/A	N/A	N/A	0.62	0.18 to 2.18	Very low
Bladder neck needle	Single incision	N/A	N/A	N/A	0.49	0.20 to 1.24	Very low
Ant repair	Single incision	N/A	N/A	N/A	0.36	0.15 to 0.82	Very low
PFMT	Single incision	N/A	N/A	N/A	0.84	0.28 to 2.78	Low
Ant repair	Bladder neck needle	1b	0.92	0.55 to 1.55	0.72	0.31 to 1.63	Very low
PFMT	Bladder neck needle	N/A	N/A	N/A	1.72	0.45 to 6.89	Low
PFMT	Ant repair	N/A	N/A	N/A	2.38	0.65 to 9.30	Very low

The direct pairwise analyses showed some heterogeneity in four direct comparisons for the number of women cured (single-incision sling vs. retropubic MUS; single-incision sling vs. transobturator MUS; laparoscopic colposuspension vs. open colposuspension; and anterior vaginal repair vs. open colposuspension) and one comparison for the number of women improved (laparoscopic colposuspension vs. retropubic MUS) (see Appendices 10 and 11).

Ranking of treatment effectiveness

Figure 2 shows the SUCRA values for all surgical interventions. The rankograms are reported in Appendix 13.

FIGURE 2.

Surface under the cumulative ranking curve values for (a) the number of women cured; and (b) the number of women improved. Ant repair, anterior vaginal repair; bladder neck needle, bladder neck needle suspension; open colpo, open colposuspension; lap colpo, laparoscopic colposuspension; retro-MUS, retropubic MUS; single incision, single-incision sling; trad sling, traditional sling; transob-MUS, transobturator MUS.

Traditional sling and retropubic MUS (see Figure 2a) are the treatments most likely to result in the highest proportion of women cured (89.4% and 89.1%, respectively), followed by open colposuspension (76.7%), transobturator MUS (64.1%), laparoscopic colposuspension (48.9%), single-incision sling (39.8%), bladder neck needle suspension (26.9%), anterior vaginal repair (12.5%) and PFMT (2.6%). On the other hand, retropubic MUS (97.0%) and transobturator MUS (76.1%) are the most likely treatments to result in the highest proportion of women with an improvement in their incontinence symptoms (see Figure 2b), followed by traditional sling (67.7%), open colposuspension (63.8%), laparoscopic colposuspension (45.8%), single-incision sling (42.0%), PFMT (39.2%), bladder neck needle suspension (14.3%) and anterior repair (4.1%).

Adverse events: direct pairwise comparisons

The adverse event results are shown in Appendix 15. Full meta-analyses results for all included studies are available in Report Supplementary Material 1.

Overall, few studies reported adverse events. Numbers of events included in the analyses were generally small and, therefore, CIs were wide. It is worth noting that many of the adverse events meta-analyses were based on < 5 studies. This is mainly owing to the dearth of available data but also to the inconsistencies with regard to the type and definition of adverse events as well as to the time points at which these were measured across individual trials and across Cochrane systematic reviews. In particular, the lack of common definitions made it difficult to combine results and to incorporate data extracted from newly identified studies into the pool of data extracted from the relevant Cochrane reviews. For some of the included studies, follow-up time was unknown, as it could not be extracted from the relevant Cochrane systematic reviews. Below we summarise the most relevant meta-analyses results, focusing on comparisons for which most data were available (in terms of the number of studies and number of participants).

Resource utilisation: pairwise comparison

Appendix 15, Figure 64 reports the meta-analyses results for the different treatment comparisons for the assessment of resource utilisation. For some comparisons only a limited number of studies were available and no firm conclusions can be drawn.

Summary of clinical effectiveness assessment

The systematic review of clinical effectiveness was based on data from 175 studies (21,598 women in total) comparing one surgical procedure with another for the treatment of SUI in women. The number of included studies identified from the eight published Cochrane systematic reviews was 147 (84%). A further 28 studies (16%) were identified through an updated literature search conducted in May 2017. This updated search used the same search strategy and eligibility criteria as those used for the included Cochrane systematic reviews.

A further, more recent, updated search was conducted in October 2017 and identified an additional 10 articles that appeared to meet our inclusion criteria. These studies have not been incorporated into the current assessment, but are listed, for information, in Appendix 16.

The majority of included studies had a high or unclear risk of bias across all risk-of-bias parameters but most notably for allocation concealment (selection bias). As blinding of participants and personnel is not feasible in trials assessing surgical interventions, protection against performance bias and detection bias was likely to be compromised in the included studies. This is a general issue with all surgical trials.

The assessment of effectiveness focused on two primary outcomes: the number of women cured from incontinence and the number of women who experienced an improvement in their incontinence symptoms. The NMA, which combined evidence from direct head-to-head comparisons and indirect comparisons, included 120 studies that reported data on cure or improvement. The NMA results indicate that retropubic MUS, transobturator MUS and traditional sling are more effective than other surgical procedures for both primary outcomes. Open colposuspension appears to be relatively effective for both outcomes (see the SUCRA ranking in Figure 2). For both primary outcomes, cure and improvement, the overall quality of evidence was low. As we were unable to extract study characteristics, we were unable to check the transitivity assumption (studies are similar with regard to important characteristics).

Assessment of adverse events was hampered by the dearth of available data. Direct head-to-head meta-analyses were available mainly for treatment comparisons involving retropubic MUS, transobturator MUS or single-incision sling. Follow-up time was generally short, with a median of 12 months.

Compared with retropubic MUS, transobturator MUS was associated with lower rates of major vascular complications and voiding difficulties but higher rates of repeat surgery.

For other intervention comparisons, the number of studies was generally small and the CIs wide. However, there was some evidence to suggest that bladder perforation was more likely to occur after retropubic MUS than after transobturator MUS, open colposuspension or traditional sling. Rate of tape or mesh erosion or extrusion was similar between transobturator MUS and retropubic MUS (27 studies with a total of 4523 women).

Transobturator MUS was associated with a higher rate of groin pain but a lower rate of suprapubic pain than retropubic MUS. Retropubic MUS had a higher rate of post-operative pain than single-incision sling and transobturator MUS had a higher rate of unspecified pain than single-incision sling. Studies that assessed pain had usually short follow-up periods (i.e. ≤ 12 months).

The main area of uncertainty for all surgical interventions relates to the lack of long-term data for the assessment of their safety.

Step 1: identification of attributes and levels

The findings from the clinical effectiveness review (section 3) were consolidated to facilitate the creation of key attributes and associated levels related to women’s preferences for the different surgical treatment options. The chosen attributes and levels were plausible in both clinical and policy terms. 205 Once the attributes and levels were finalised a DCE design was generated (see Step 2: experimental design).

It is recognised that a reasonable number of attributes to be included within a DCE is around six, as more than this might lead to an unmanageable number of options to be meaningfully considered by the respondent. 213 The more attributes that are included in a DCE, the less likely respondents are to fully consider the pros and cons of different choices as described by differences in the attribute levels. Therefore, it may prove more difficult to identify trade-offs between attribute levels that a respondent is willing to make during the decision-making process, which is crucial when estimating marginal rates of substitution (MRS) (an estimation of the ratio between values that individuals place on different attributes, quantifying the trade-off a respondent is willing to make between two attributes). Probabilities for all adverse events included in the choice tasks were assumed to be similar and were therefore considered constant for all scenarios.

The final list of attributes and corresponding levels is reported in Table 7.

Step 2: experimental design

The DCE survey consisted of a number of possible clinical scenarios based on the attributes and levels identified in step 1. Each scenario was made up of all five attributes and the associated levels contained in each attribute. The combination of attributes and levels presented in Table 7 would generate > 4000 possible scenarios [4320 (5 × 4 × 6 × 6 × 6)]. We reduced the number of scenarios to a manageable number by using a measure known as D-efficiency, which provides an efficient choice design by minimising the overlap in attribute levels between scenarios. The D-efficient design was used to reduce the number of scenarios to the minimum required to estimate all main effects and higher order interactions while still providing sufficient data to estimate preferences from responses. 205,209 We applied one restriction that implied that an adverse event of infection could only occur with a hospital stay of ≥ 2 days to mirror clinical practice in which post-operative infections extend the hospital stay.

The design for the DCE was generated using Ngene 1.1 (ChoiceMetrics, Sydney, NSW, Australia). The best design generated by Ngene was chosen with the aim of minimising standard errors. The reliability of the model parameters to be estimated can be quantified in terms of the asymptotic standard errors and covariances; thus, improvements in reliability indicate a reduction in the asymptotic standard errors (i.e. variance around preference estimates). 215 A blocked design of DCE choice sets was used to ensure greater variance in the data. We used three blocks of 12 questions to minimise respondents’ burden, randomly allocating respondents to a block of DCE choices. Based on a pilot run of the DCE, we updated the design to improve the validity of the results.

An example of a pairwise choice set is shown in Table 8. Respondents were asked to choose their preferred scenario from each pairwise choice set. An opt-out option was given, which in this instance was defined as the choice of no treatment (i.e. ‘I would choose not to undergo surgery’).

The DCE survey was part of a full questionnaire that included (1) questions on basic sociodemographics (including age, ethnicity and employment status), (2) questions on potential incontinence history, (3) an introductory text explaining the DCE task and (4) the main DCE survey.

Step 3: pretesting and data collection

Step 4: data analysis and interpretation

Data were returned to the research team in an anonymised format. The research team did not receive contact details or any personal identifier information from survey participants.

Data were analysed using a random utility model framework and appropriate logistic regression techniques to estimate the mean change in utility that women place on different attribute levels compared with the reference level. This assumes that respondents choose the alternative in a choice set that gives them greatest utility out of the available options218 and that therefore the choices individuals make in a DCE reveal the utility they place on the alternatives presented.

Analyses were undertaken on the full study sample of the general population and separately for subgroups of women with and women without any type of UI.

Descriptive analysis

A total of 789 women completed the DCE. Each of the respondents provided responses for 12 choices, resulting in a total of 9468 choices that were included in the analysis. Table 9 describes the sample population. The number of missing data for some of the characteristics describing the women responding to the survey was ≲ 1%.

The average age was 53 years and significantly lower in the patients group than in the non-patients group. The majority of respondents lived in England (80%), followed by Scotland (12%), Wales (6%) and Northern Ireland (2%). The percentage of women with children differed significantly between both groups: 81% of women with UI reported having children, whereas only 63% of women without UI reported having children. This is intuitively sensible, as having had children is a risk factor for developing some forms of incontinence. The majority of respondents in both groups lived with their partner (with UI, 71%; without UI, 66%) and around half were in full- or part-time employment (with UI, 53%; without UI, 48%), 21–30% were retired and 3–7% were self-employed. Most respondents in either group had an annual household income of £15,000–25,999 and between 15–20% reported an annual household income of ≤ £15,000 or £26,000–34,999 and £35,000–49,999, respectively. The distribution of educational attainments was relatively even between both groups, although a higher proportion of patients had a university degree (with UI, 41%; without UI, 35%). The majority of respondents (92–93%) had a white ethnic background. Health status varied significantly between both groups. On average, women with UI reported poorer health than those without UI and around two-thirds of non-patients described their general health to be ‘excellent’ or ‘good’.

Among women with UI, the most common type of UI was SUI (36%), followed by MUI (33%) and UUI (26%). The majority of women with UI found their incontinence to be not or only slightly limiting in their daily activities (13% and 48%, respectively), 29% reported it to be moderately limiting and 10% reported it to be extremely limiting. Around 43% had received treatment previously. However, the majority of previous treatment took the form of management of incontinence (i.e. non-surgical treatment). Out of the 128 patients with SUI, nearly half reported to have treatment scheduled and 7% reported to be a member of an incontinence support group.

Among women without UI, 28% knew someone close to them with UI.

Statistical analyses

Two-thirds of women were prepared to make a choice between surgical options and one-third of women declared that they would rather choose no surgery. Responses to the three choices were evenly split: around one-third of all responses was allocated to each of the options (A, B and the opt-out option), indicating that the majority of respondents would choose surgery (options A and B) over no surgical treatment (opt-out option).

Table 10 reports the marginal effects of each attribute on utility for the conditional models (1–7) and the MXL model (8). Conditional logistic regressions were run for all respondents (models 1–2), women with UI (models 3–6) and women without UI (model 7). Marginal utility values indicate relative preferences for levels within an attribute (for example, relative preferences against mild/moderate/severe pain compared with no pain). Positive marginal utility values indicate an attribute level is preferred to the reference level and negative marginal utility values indicate that the attribute level is valued less than the reference level. For the MXL model, means and SDs are reported for the random parameter. The sign of SDs is irrelevant; significant SDs imply preference heterogeneity among respondents for that specific attribute, suggesting that the reported coefficient may disguise differences in the underlying preferences. To identify which differences are statistically significant, p-values are included.

The results were found to be very similar across all models, except for the marginal effect of the ASC capturing a preference for or against surgical treatment in general. Preferences for surgical treatment varied significantly in both the significance and direction of the effect depending on model specifications. The coefficient for the ASC was only significant in some of the models for women with UI (model 4–6) and the mixed-effects model (model 8), whereas the estimations for the full sample and women without UI suggest no significant result (i.e. no preference for or against surgical treatment in general). Significant coefficients for both the mean and SD from the mixed-effects model suggest that preference heterogeneity is present among respondents.

Preferences for the different types of adverse events were relatively robust to varying model specifications. Infections and pain during intercourse were found to be associated with a utility gain and, therefore, not disliked as much as the reference category of new urinary symptoms, and damage to organs or nerves and voiding difficulties were less preferred.

The results for different levels of chronic pain were in line with expectations, describing a gradient of increasingly negative preferences with higher levels of pain. Compared with the reference level of no pain, mild chronic pain was not found to affect utility significantly across most models, whereas moderate and severe pain were associated with a greater negative preference. However, the results were mediated in models for women with UI that included interactions between the ASC and how limiting the type of UI was perceived with regard to daily activities (models 4 and 6) and between the ASC and a positive treatment history (models 5 and 6). In these models, the negative gradient could still be observed but was no longer statistically significant. Instead, if respondents had had treatment before, and the more limiting their UI, the more likely they were to choose surgery.

As expected, women would prefer not to experience longer hospital stays or a higher risk of recurrence. However, the negative preference for a higher risk of recurrence was not significant in models for women with UI (models 3–6) in which interaction terms between the ASC (and, therefore, the overall preference for surgery) and different characteristics capturing patient experiences may have dominated the impact of different levels of pain. The sample of women without UI showed the strongest gradient of negative preferences for higher levels of pain. A longer time to return to normal activities was associated with a utility gain and, therefore, described a positive preference.

Adding interactions between ASCs and additional variables improved the goodness of fit for all patient models, and model 6 was found to be the best-fitted model. Compared with patients with a mixed form of UI, those suffering from SUI were more likely to choose surgery. Similarly, patients with a positive treatment history and a more limiting form of UI gained more utility from choosing surgical treatment options.

Comparing the log LRs between model 1 (conditional logit) and model 8 (MXL) that used the same variables and same study populations showed that the MXL model provided better model estimates when accounting for preference heterogeneity around the ASC. In the non-patient model 7, knowing someone close with UI did not significantly increase the preference for surgical treatment.

Marginal rates of substitution were calculated for all attributes included in the DCE based on average willingness to wait for a return to normal activities (i.e. how much longer an individual would be willing to wait, on average, for a change in the utility of another attribute). In comparison with new urinary symptoms, individuals would, on average, be willing to wait an additional 3.8 weeks to return to normal activities to avoid damage to their nerves or organs and up to an additional 12 weeks, on average, to avoid voiding difficulties. Individuals were not willing to wait any additional time to avoid infections or pain during intercourse in comparison with new urinary symptoms.

To avoid chronic pain, individuals would, on average, be willing to wait an additional 2.3 weeks to avoid mild pain compared with no pain. Individuals would be willing to wait an additional 8 and 9 weeks, on average, to avoid moderate and severe pain, respectively.

For a one-night reduction in their hospital stay an individual would, on average, be willing to wait 1 additional week to return to normal activities. To reduce the risk of recurrence by 1%, an individual would on average be willing to wait an additional 15 hours.

Predicted uptake rates, based on a hypothetical best-case surgery (adverse event of infection, no pain, a length of stay of 2 days, 12 weeks to return to normal activities and 0% risk of recurrence) versus a no surgery option were calculated. Given this option, it is predicted that 62% of respondents would choose the surgical option over no surgery. Similarly, for the hypothetical worst-case surgery (voiding difficulties, severe pain, length of stay of 6 days, 2 weeks to return to normal activities and 50% risk of recurrence) versus the no surgery option, the predicted uptake rates were estimated as 33% for the worst-case scenario. In this case, 67% of respondents would, on average, choose no surgery.

Summary of results

To our knowledge, this is the first DCE to investigate differential characteristics of surgical treatment options for UI and their potential impact on women’s decisions to undergo surgery. Analysis of choice set data from the DCE revealed that around two-thirds of respondents chose a treatment option over no surgery. However, we found evidence that preferences around the choice of surgical treatment may vary across individuals.

In terms of types of adverse event associated with surgical treatment, infections and pain during intercourse were preferred to the reference category of new urinary symptoms, whereas damage to organs or nerves and voiding difficulties were less preferred. These preferences were relatively robust across all models, but the results, although confirming the same trend in preferences, were not statistically significant for damage to organs or nerves and pain during intercourse in non-patient respondents. Respondents associated increasing levels of disutility (i.e. negative preference) with higher levels of chronic pain, longer durations of hospital stay and higher risk of recurrence. We also found evidence that longer durations to return to normal activities after the surgery was associated with a statistically significant, positive preference.

Interpretation of results

The heterogeneity around a general preference for or against surgical treatment may be explained by the range of surgical treatments available, their level of invasiveness and, therefore, their varying impact on some of the attributes included in the DCE. The different subgroups in our sample may have contributed to the heterogeneity around the preference for surgical treatment in general. Comparing results across all models suggests that general preferences concerning surgical treatment vary not only between patients and non-patients but also within the patient subgroup, as implied by the change in sign of the ASC coefficient between model 4 and models 5 and 6. Although the results for the full sample suggest a preference for surgery, this finding may have been driven by the majority of non-patients within the sample. We found some evidence that patients in general may associate a negative preference with surgery for UI. However, those patients with SUI, who reported their UI to be extremely limiting with regard to daily activities and who had undergone surgical treatment before were more likely to choose surgery again. This may indicate that patients who have never undergone surgery fear the surgical procedure and are more reluctant to choose surgical treatment. One reason for this finding may be that patients are more used to managing their UI and may consider non-surgical treatment as long as their condition is not experienced as limiting their daily activities too much. Interviews conducted with patients during the pretesting stage of the DCE provided further support for this finding.

Preferences around different types of adverse event following surgery were captured compared with the reference of new urinary symptoms. Our findings suggest that infections and pain during intercourse are considered to be more acceptable adverse events, whereas damage to organs or nerves and voiding difficulties were less acceptable and reduced the likelihood of respondents choosing a surgical treatment option. This is also confirmed by the MRS analysis, which showed that individuals were willing to wait longer to recover to avoid damage to organs or nerves and to avoid voiding difficulties but were not willing to wait to avoid infections or pain during intercourse.

As expected, we found evidence for higher levels of chronic pain and increased risk of recurrence to be associated with negative preferences. However, although the same trend could be observed in all models, the statistical significance varied, which may indicate the difference in importance of these attributes for patients and non-patients. When accounting for patient characteristics around their type of UI, the degree to which it was limiting them in daily activities and their treatment history, preferences for chronic pain and risk of recurrence were less significant. Patients with SUI or UUI, extremely or moderately limiting UI and a treatment history had a stronger preference for surgery regardless of the associated level of pain or risk of recurrence.

There was a strong preference for shorter hospital stays, suggesting that the less invasive may be the preferred treatment options, with reduced risk of infections that may increase the length of hospital stay following surgery.

Other than expected, we found evidence for a preference for longer durations to return to normal activities, although the pilot data had suggested a negative preference for this attribute. A reason for this finding may be that respondents may have interpreted ‘return to normal activities’ as ‘return to work’ (assumes that they would not lose any wages). Alternatively, and especially for those who had never experienced any surgical treatment, respondents may have found the attribute levels to be unrealistic, expecting the recovery after surgery to be longer than the stated maximum of 12 weeks.

Strengths and weaknesses

A major strength of our study, which elicited preferences for different characteristics of surgical treatment options of UI is the use of a DCE in combination with an online panel. It allowed us to use a relatively large sample size and control for differences in respondent characteristics accordingly. The anonymity of the online setting may have allowed the responding UI patients to be more honest and report their problems in greater detail than they may feel comfortable to report in a health-care setting. The advantage of using a DCE for preference elicitation is that it allows consideration of real-life choice situations in which multiple factors may influence the decision-making process. The combined influence of those factors and potential trade-offs between their varying levels cannot be readily elucidated or quantified by other methods.

However, DCEs have been criticised as difficult for participants to understand. We aimed to minimise this risk by following best practice guidelines for the design and development of DCEs205,211 and engaged members of the general public and patients in the development process to maximise internal and external validity.

We acknowledge that the sample recruited by the research company is a convenience (i.e. non-random) sample that may not be representative of the target population and did not allow us to determine response rates to the survey. Some of our analyses were aimed at identifying differences in preferences between subgroups of patients and non-patients; however, the characteristics of those subgroups may have differed by definition (e.g. see the lower average health status and higher proportion of respondents with children in the patient subgroup). Alternative methods of recruitment and data collection in the context of DCEs, such as postal or telephone surveys, have been found to achieve very low response rates, thereby potentially reducing the representativeness and validity of results. 220

Implications of findings for research

To our knowledge, this research provides the first insights into what women (with and without) UI want their treatment options to be like and which fundamental surgical treatment characteristics they prefer. More than half of respondents had UI or knew someone close with UI. However, given the wide range of surgical treatments and significant differences between them, capturing specific influences on and preferences for or against certain treatment characteristics is a complex process. Our patient subsample was not large enough to capture sufficient variation beyond the interaction effects for patient experience already included in our analyses. More research is needed to distinguish trade-offs between different characteristics of surgical treatment options. One study found that women suffering from UI would choose the treatment with the lowest risk of recurrence. 221 Our results provide further support but suggest that this preference may be mediated by other factors such as previous treatment experience and that this finding cannot necessarily be generalised without further research.

Although UI can have a substantial impact on a woman’s daily activities and quality of life, previous research suggests that women under-report their condition and manage their incontinence before seeking treatment. 222 Barriers to seeking professional help may be imposed by a lack of awareness of treatment options, by the perception that these symptoms are normal after childbirth or in older age, or by feelings of shame or embarrassment. 223,224 Considering women’s preferences and trying to reduce those barriers may help to reduce the levels of unmet need for incontinence services and to improve the organisation of surgical services.

Search strategy

The search strategy is outlined in Appendix 17.

Data sources

The following databases were searched during the review process:

• MEDLINE (via Ovid)

• MEDLINE In-Process & Other Non-Indexed Citations (via Ovid)

• Excerpta Medica Database (EMBASE) (via Ovid)

• NHS Economic Evaluation Database (NHS EED) (via Ovid)

• Health Management Information Consortium (via Ovid)

• Cost-effectiveness Analysis (CEA) Registry.

All databases were searched using the Ovid interface, except for the CEA Registry, which was searched through the CEA Registry website (https://cevr.tuftsmedicalcenter.org/databases/cea-registry).

Study selection

Two researchers undertook the screening of titles and abstracts obtained through the search using EndNote^® X7 [Clarivate Analytics (formerly Thomson Reuters), Philadelphia, PA, USA] reference management software. Prior to initial screening, deduplication of records was undertaken. After initial screening, all potentially relevant articles were obtained for further scrutiny against the full selection criteria, with any disagreements resolved by discussion. The inclusion criteria were as follows:

• Study design: full economic evaluations involving a decision model-based analysis. A full economic evaluation can be defined as a comparative study including both costs and effects.

• Population: women with SUI and stress-predominant MUI (adult women of any ethnic background).

• Intervention: any of the surgical treatments for UI being evaluated in our own study (as either a primary or a repeat surgery).

• Comparator: alternative treatment (surgical and non-surgical) or no treatment for UI.

• Outcome: cost-effectiveness, cost estimates, utilisation estimates and quality-of-life estimates.

• Language: studies with full text in English.

No restrictions were placed on the publication time frame or the study country. Modelling studies where data were imputed from multiple sources, without specific reference to the age and ethnic background of women, were included in the review. Studies comparing diagnostic/screening techniques or other non-surgical procedures without considering surgical interventions in the model pathway were excluded.

Data extraction

Relevant data were extracted into an Excel spreadsheet. Data on the following, where available, were extracted from included studies by one reviewer and checked by another:

• study characteristics, such as study question, form of economic analysis, population, interventions, comparators, perspective, time horizon and form of modelling

• clinical effectiveness and cost parameters, such as effectiveness data, health-state valuations (utilities), resource use data, unit cost data, cost year, discounting and key assumptions

• results and sensitivity analyses.

Summary of review data from included model-based economic evaluations

A total of 17 studies were included in the final review. An overview of the key data extracted from these studies is presented in Table 11. In the section that follows, a more comprehensive description of the data extracted from the included studies is presented. Finally, a descriptive quality assessment of the included studies is presented, based on the criteria defined in the Drummond checklist for economic evaluations. 242

Comprehensive overview of data extracted from included model-based economic evaluations

Model structure and comparators

Model inputs and analysis

Quality assessment

The completed Drummond checklist for each of the included studies is presented in Appendix 18. In this section, a descriptive summary of the quality assessment of all included studies is presented.

Limitations of identified model-based economic evaluations in informing economic analysis

A total of 17 model-based economic evaluations were identified through the systematic review. All studies were conducted among women with SUI or stress-predominant MUI and included at least one of the surgical treatments for UI being evaluated in this study (as either a primary or a repeat surgery). However, none of the models was sufficient for the analysis required for our own study.

First, the time horizons of the models identified were generally of insufficient length, with only one of the included studies assessing costs and consequences over a lifetime time horizon. 226 One study did report a 40-year time horizon, but even this would not be of sufficient length to capture the entire lifetime of all patients being modelled in our own analysis (patient starting age, 45–55 years). The majority of studies extended to a maximum time horizon of 10 years.

Second, only 11 of the identified studies were conducted as pure cost–utility analyses. 225–230,233,235,237,238,241 Of these, only three used EQ-5D as a quality of life measure in their analysis (the measure proposed for our own analysis). 229,238,241 The other eight studies that conducted a cost–utility analysis did not report a quality of life measure and this was noted as a generally poorly reported feature among the studies identified. 225–228,230,233,235,237

Furthermore, very few studies had used appropriate utility weights estimated using tools like EQ-5D and the Health Utilities Index227,229,233,235,241 for cured and incontinence health states in the economic models. 227,235 Because the intended outcome of the surgical treatments is increased quality of life, estimating effectiveness using an appropriate tool is essential for the estimation of actual treatment outcomes. Moreover, not all of the studies had included complications in the models. Complication after surgical treatment is another important factor that affects women’s quality of life, which highlights the importance of including major complications in the calculations.

Finally, only one of the identified models did not omit important alternatives from their analysis. 241 As highlighted during the quality assessment, all other studies failed to include at least one important alternative. 225–240

Given these limitations, and the information presented in this chapter, it was decided that a new economic model needed to be developed to estimate the relative cost-effectiveness of the nine surgical interventions.

Relative effectiveness of surgical treatments

The relative effectiveness of surgical treatments, in terms of subjective cure rates, were based on the results of the NMA reported in Chapter 3. Table 12 describes the mean and median ORs for different surgical treatments versus retropubic MUS, which was taken to be the reference treatment. The reported values for mean and median ORs were used to define log-normal distributions in the model.

Estimation of absolute cure rates

The absolute cure rates were calculated in the model by combining the information on relative cure rates described in Table 13 with the absolute cure rates for retropubic MUS. The absolute cure rates at 1 year for retropubic MUS were estimated to be 84%, based on a meta-analysis of retropubic MUS trials (44 studies were included).

To calculate the absolute probability of success for each surgical treatment, the OR of a given intervention compared with the baseline treatment (retropubic MUS) was derived from the NMA. This figure was then converted into a relative risk (RR) using the following formula:

where:

The RR for each treatment was then applied to the baseline risk247 to estimate the absolute probability of success for each surgical treatment.

Repeat surgery

Because data from HES for England in 2013–14 and data for hospital discharge in Scotland in 1996–2014 show that around 91% to 96% of SUI surgeries were for MUS (retropubic, 55%; transobturator, 45%), and given that, from a clinical point of view, MUS (either retropubic or transobturator) can be offered after all types of surgery, it was assumed that all subsequent surgeries were either retropubic MUS (55%) or transobturator MUS (45%). 251 Owing to data limitations, subgroup analysis was not conducted to explore success rates after repeat surgery (after failed previous continence surgery) versus primary procedures. Therefore, it was assumed in the model that repeated MUS surgeries after failed primary surgeries are almost as effective as primary MUS surgeries, based on the finding from Kociszewski et al. 252 Kociszewski et al. 252 showed that 96% of women after primary sling implantation, and 88.46% of women after repeated sling, were dry at 6 months. The difference in percentage of negative pad test after primary and repeated procedure was not statistically significant. However, some previous studies show a lower success rate for retreatment with the same procedure. For example, Black and Downs253 showed that retreatment colposuspension was 78.4% less effective on average than primary colposuspension. Therefore, we tested two scenarios: in the first scenario it was assumed that subsequent surgeries (retropubic and transobturator MUS) are 90% as effective as they would have been as a primary surgery, and in the second scenario it was assumed that the subsequent surgeries are 75% as effective as primary retropubic and transobturator MUS.

Long-term repeat surgery rates estimation

To estimate the rate at which women choose to remain incontinent rather than seek retreatment in the model, individual-level data from a previous study was obtained. 254 In brief, the study used data from Aberdeen Maternity and Neonatal Databank (AMND) to estimate the lifetime risk of undergoing pelvic floor surgery in a cohort of UK parous women. The AMND stores linked information on all obstetric-related events occurring in women living in Aberdeen city and district since 1950 and currently contains data for approximately 200,000 women. Within the cohort of women, 762 (2.2%) had an operation for UI. We utilised this subset of data to estimate the risk of reoperation over time after different types of SUI surgeries. We excluded 11 women who had urogenital fistula repair surgery as it is not a SUI procedure. We analysed 751 cases that had SUI surgery. The number of women with one, two, three and four surgeries for each type of surgery are reported in Appendix 20. As some of the surgical treatments are no longer popular, this dataset was used because it contains data for those surgical treatments.

These data were analysed to estimate the rate at which women chose to remain incontinent rather than seek retreatment in the model. Parametric methods were used to extrapolate survival times beyond the follow-up duration in the data. We analysed the data using graphical exploration and statistical criteria of goodness of fit. First, Kaplan–Meier curves and survival, hazard and other related functions were examined graphically. The exponential, Gompertz, log-logistic, Weibull and log-normal distributions were fitted to the data and their fit was assessed visually and compared using statistical criteria based on log-likelihood, the Akaike information criterion (AIC) and the Bayesian information criterion (BIC); the model with the lowest AIC and BIC is preferred. 255 A log-normal distribution was applied to estimate the long-term rates of repeat surgery based on the AIC and BIC criteria. All fitted and observed survival data for different types of SUI surgeries are presented in Appendix 20.

Finally, the rate at which women choose to remain incontinent rather than seek retreatment was estimated based on the estimated cure rates and estimated repeat surgery rates at each time point. For example, if 1-year cure and repeat surgery rates after a particular surgery were 80% and 10%, respectively, the following formula was used to estimate the proportion of women who seek a second treatment following failure of the first treatment:

where P_rt is the probability of having repeat surgery after a given surgery at time t and p_cure is the probability of a surgery still being successful at time t.

The observed repeat surgery rates were compared with data from the included RCTs and pairwise meta-analysis results for repeat surgery rates after different surgical treatments. We felt that the observed repeat surgery rates were underestimating the actual rates. For example, results from the analysis show that the rate of repeat surgery after anterior vaginal repair (colporrhaphy) in 2 years is about 3%, whereas data from two trials show that the rate is ranging from 26.0% to 37.5%. 194,196 Therefore, in the base-case analysis it was assumed that (based on the data from previous assessment of surgical treatments for SUI) 75% of women whose first treatment was not successful would seek retreatment and 30% of women whose first retreatment failed would seek second retreatment in the base-case analysis. 256 We used the results from the analysis of AMND data in the SA to inform the model in terms of the proportion of women who will seek retreatment after failure of first surgery.

Complication rates after surgical treatments

There are different types of complications that can happen after each surgical treatment. The model incorporates only severe complications/adverse events that are important in terms of their effect on women’s quality of life, the cost and duration of treatments and data availability. To estimate complication incidence rates, random-effect meta-analysis models were fitted using WinBUGS software. 33 A Bayesian approach was used to combine existing knowledge with prior information based on established rules of probability. We checked the convergence of the model in different number of iterations; 50,000 Markov chain Monte Carlo iterations after a burn-in period of 20,000 iterations was chosen to get posterior distributions. The convergence was checked graphically using trace and autocorrelation plots. All the complications included in the model, and estimated incidence rates, are presented in Table 16. The estimated incidence rates were defined as distributions in the economic model (beta distribution). Lastly, the rates were used to estimate health-care costs and utility decrements associated with each treatment (see Resource use and unit costs and Health utility for more details).

All-cause mortality rates

In the model, women were at risk of death (from any cause) during any given cycle period. The cycle-specific risk of mortality was dependent on the simulated woman’s age. The risk of mortality will therefore increase with each cycle period (the ‘all-cause ageing effect’). Age-specific all-cause mortality rates were derived from general population mortality statistics reported in national life tables (Office for National Statistics). 232 The model incorporated the risk of operation-related death also. A rate of 0.0005 has been applied for open surgeries in previous studies241,256 based on the results from a report by Leach et al. 257 In addition, a recent study showed that urethral injection therapy had a higher mortality rate than open colposuspension [adjusted incidence rate ratio (IRR) 1.98, 95% CI 1.36 to 2.91]. They also showed retropubic MUS had a lower mortality rate than open colposuspension251 (adjusted IRR 0.65, 95% CI 0.44 to 0.96).

Clinical inputs for effectiveness of urge incontinence treatments

For those women with MUI who have UUI following surgery for SUI, or those SUI women who develop de novo UUI after surgery for SUI, three lines of treatment were assumed. Although it is possible to get more than three types of treatment for UUI, to simplify the model only three types of treatment were incorporated:

1. first-line treatment: bladder training

2. second-line treatment: anticholinergics drugs (oxybutynin hydrochloride)

3. third-line treatment: botulinum toxin type A.

The results from a systematic review undertaken by Lee et al. 258 showed that the efficacy of bladder training on symptoms of urgency ranges from 12% to 90%. In the base-case analysis, a 32% cure rate after bladder training was assumed.

The results from another NMA, which was conducted recently, showed that the absolute probability of being continent at 12 weeks after oxybutynin hydrochloride treatment is 21%. 259 Oxybutynin hydrochloride (5 mg twice daily) was used as an index second-line treatment. After 3 months, probabilities were assumed to be constant, which is consistent with the evidence from a long-term study suggesting that the treatment effect of antimuscarinic therapy at 4 months is maintained to 24 months. 260 Failure rates after botulinum toxin type A treatment at different time points were obtained from a previous economic evaluation that was conducted within the development of clinical guidelines for UI. 259 It was assumed that the cure rate at 6 months using botulinum toxin type A is 36%. 259

Surgery costs

Surgeries for SUI and stress-predominant MUI vary in terms of the complexity of the procedure and the setting in which surgery would be conducted. For women undergoing anterior vaginal repair, bladder neck needle suspension, open colposuspension, laparoscopic colposuspension and traditional sling procedures, surgery would typically be conducted in an inpatient setting, based on expert clinical input. Women undergoing retropubic MUS and transobturator MUS procedures would typically be treated in a day case setting, based on expert clinical input. Costs for each of these individual surgeries were derived from NHS Reference Costs 2015/16. 261 Finally, both single-incision sling procedures and urethral injection therapy would be conducted in a day case setting, on the basis of clinical advice. No appropriate NHS reference costs were identified for these procedures and, therefore, costs for both of these surgeries were derived from previous literature, with costs inflated to a 2015/16 price year using the hospital and community health services (HCHS) index listed in the Unit Costs of Health and Social Care 2016. 262 The cost of each surgery is presented in Table 17.

Additional costs associated with surgery

In addition to the cost of individual surgeries, costs associated with complementary tests, treatments and consultations that would typically be carried out in advance of, and following, each surgery were also considered. Clinical experts advised that women undergoing surgery for SUI and stress-predominant MUI would typically use incontinence pads until their condition had been resolved. The weekly cost of pads was estimated from a NICE guidance document214 with costs based on a guidance development group opinion on estimated pad usage per week. Costs reported were inflated to the current price year and converted to a 3-monthly cost to fit the model time cycle. Clinical experts advised that a urodynamic test would be conducted in a separate consultation before each surgery, and that the woman would also have a separate 20-minute consultation before undergoing surgery with the clinician carrying out the surgery. Costs for test and consultation were derived from NHS Reference Costs 2015/16261 and PSSRU 2016262 respectively.

Clinical experts involved in the study advised that a urine dipstick analysis and full blood count would need to be taken prior to each surgery, excluding injectable agent procedures, for which only a urine dipstick analysis would be required. Costs for both were obtained from NICE clinical guidelines for routine pre-operative tests for elective surgery. 214 In addition, it was advised that a cystoscopy on the day of the procedure would be mandatory for women undergoing a retropubic MUS or transobturator MUS procedure. The cost of a diagnostic flexible cystoscopy in an outpatient setting was obtained from NHS Reference Costs 2015/16. 261 Finally, the cost of medication needed for pain relief post surgery (paracetamol) was considered, with the duration of pain relief required greater for inpatient procedures (7 days) than for the less invasive day case procedures (3 days). It was also advised that lactulose would be taken as pain relief for 1 day following anterior vaginal repair only. Costs of medication were obtained from the BNF 2016. 263 All additional costs are reported in Table 17. Omitting the cost of incontinence pads, which are used by all women with incontinence regardless of the type of initial surgery, the total additional costs associated with anterior vaginal repair, bladder neck needle suspension, open colposuspension, laparoscopic colposuspension, traditional sling, retropubic MUS, transobturator MUS, single-incision sling and urethral injection therapy were £212.87, £212.37, £212.37, £212.37, £212.37, £363.73, £363.73, £211.73 and £205.42, respectively.

Urge urinary incontinence costs

As well as the cost of surgical treatments for SUI and stress-predominant MUI, the costs associated with treatment for UUI were also considered. The NICE clinical guidelines on the management of urge incontinence recommend three broad lines of treatment for women with this type of incontinence:214

1. conservative measures such as bladder training

2. antimuscarinic drugs, most typically oxybutynin hydrochloride

3. invasive therapy such as botulinum toxin type A.

This treatment pathway was confirmed by clinical experts in the study team. The cost of bladder training was obtained from the appendices of a NICE draft for consultation on guidelines for UI in women. 214 Costs reported in this document were based on women being seen 5 times over a 4-month period (i.e. initial consultation and once in each of the 4 months). To estimate 3-monthly bladder training costs to fit the model time cycle, it was assumed that the woman would have an initial consultation with a band 6 hospital-based physiotherapist and would be seen for training once in each of the 3 months. The cost of a consultation was based on data obtained from PSSRU 2016. 262 The cost of consumables required during training were obtained directly from the guidelines214 and were also included in the estimate. These costs were inflated to the current price year using the HCHS index listed in PSSRU 2016. 262

Antimuscarinic medication taken was assumed to be oxybutynin hydrochloride, on the basis of recommendations from the clinical guidelines214 and expert clinical advice. Appropriate costs were obtained from the BNF 2016. 263 Additionally, it was assumed that the woman would require a consultation with a specialist before beginning medication, and this cost was based on the average unit cost of a consultant-led non-admitted face-to-face follow-up attendance in urology, obtained from NHS Reference Costs 2015/16. 261 Cost of medication was estimated for 3 months. Finally, the cost of a botulinum toxin type A procedures was based on micro-costing details presented in a NICE draft for consultation on guidelines for UI in women. 214 Included in this overall procedure cost was the cost of an intermediate endoscopic bladder procedure (day case) and the cost of botulinum toxin type A 100-unit powder for solution for injection vials. These costs were obtained from NHS Reference Costs 2015/16261 and the BNF 2016,263 respectively. Additionally, it was assumed that the woman would require a consultation with a specialist before undergoing a botulinum toxin type A procedure, and this cost was based on the average unit cost of a consultant-led non-admitted face-to-face follow-up attendance in urology, obtained from NHS Reference Costs 2015/16. 261 Costs of urge incontinence treatments are reported in Table 17.

Adverse events treatment costs

The cost of treating complications associated with surgical treatments was also considered in the model. Typical treatments were informed by expert clinical advice and costs were obtained through routine sources. The cost of treating an infection was based on the cost (obtained from the BNF 2016263) of co-amoxiclav antibiotic medication over a 2-week period (advised duration of infection). It was advised that women experiencing voiding difficulties (including urinary retention), or bladder or urethral perforation, would require catheterisation. This cost was based on the cost, obtained from NHS Reference Costs 2015/16,261 of a bladder catheter procedure in an out-patient setting. As voiding difficulties may impact the woman for up to 3 months, the cost of self-catheterisation over this period was also estimated. For a bladder injury the cost of a suprapubic bladder catheter procedure in an out-patient setting was applied, as well as the cost of self-catheterisation over a 12-day period (based on expert clinical advice). The cost of a single catheter was obtained from the NICE draft for consultation on guidelines for UI in women214 and inflated to the 2016 price. It was assumed that the woman would require four catheters per day, based on data presented in the NICE guidelines. 214

The cost of a mesh excision or repair to treat mesh erosion was based on the cost of a partial removal of TVT [HRG (Healthcare Resource Group) code MA04: intermediate open lower genital tract procedures – day case], obtained from NHS Reference Costs 2015/16. 261 Based on the reference cost schedule, a relatively small percentage (16%) of these procedures will be major procedures. The intermediate option was chosen, reflecting that the greatest percentage of procedures are intermediate. The cost difference between the major and intermediate procedure is in the region of £200. This was assumed to be a one-off cost. It was advised that persistent pain would typically be treated through the use of non-steroidal anti-inflammatory drugs. For the purpose of this costing, medication taken for pain was assumed to be aspirin (based on expert clinical advice) and relevant costs were obtained from the BNF 2016. 263 The estimated incidence rates for each complication were used to estimate cost associated with complications after each surgical treatment by multiplying respective incidence rates by the cost of each complication. All costs associated with treating adverse events are reported in Table 17.

Applying higher incidence rates of mesh complications after mid-urethral sling procedures

As mentioned in the methods section, data from the trials were used to estimate the complications incidence rates after each surgical treatment. As some of the complications may happen over the longer term, and given that the trials are short in terms of the follow-up time, results from a very recent study on MUS were used to estimate the mesh complication incidence rates after retropubic MUS and transobturator MUS in the model. Keltie et al. 266 conducted a study to investigate the rate of adverse events after MUS procedures using data from the HES database. The mean follow-up time was 4.2 years (unconfounded cohort). They showed that the readmissions for further surgery after initial retropubic MUS and transobturator MUS were 3.7% [removal (2.7%), repair (1%), renewal (0%)] and 2.8% [removal (1.9%), repair (0.9%), renewal (0%)], respectively. These values were used within the SA for retropubic MUS and transobturator MUS. In addition, we explored the effect of assuming both a 10% and a 20% incidence rate of mesh complications both after retropubic MUS and after transobturator MUS on the results.

Incorporating longer duration for persistent pain complication

In the base-case analysis, it was assumed that persistent pain complication will last for, on average, 6 months. In this SA, the effect of longer-term suffering from persistent pain was explored: we tested scenarios in which women suffer for 3 years and women suffer for 5 years.

Applying higher incidence rates of persistent pain after mid-urethral sling procedures

In the base-case analysis, it was assumed that the incidence rate for persistent pain complication after retropubic MUS and transobturator MUS are 5.09% and 4.93%, respectively, based on the results from meta-analysis. In these sensitivity analyses, the effect of higher incidence rates of persistent pain (10% and 20%) after MUS procedures was explored.

Applying higher incidence rates and longer duration for persistent pain after mid-urethral sling procedures

A scenario analysis was also conducted by incorporating higher incidence rates of persistent pain after MUS procedures and longer duration. It was assumed that the incidence rates for persistent pain complication after retropubic MUS and transobturator MUS are 20% and that, on average, women with persistent pain will suffer for 5 years.

Assuming that all the women in the model have stress urinary incontinence

A SA was performed to investigate the disease type of women in the model. Owing to data limitations, it was not possible to do a full subgroup analysis and estimate different success rates for two subpopulations (i.e. women with SUI and women with MUI). In this SA it was assumed that all the women in the model have SUI and the success rates are the same as the base-case analysis.

Assuming alternative values for the proportion of women who will seek retreatment after failure of each surgical treatment

In the base-case analysis, it was assumed that 75% of women whose first treatment was not successful would seek retreatment and 30% of women whose first retreatment failed would seek second retreatment. This SA was conducted to explore results using data from the linked database (AMND) to estimate the proportion of women who will seek retreatment after failure of each surgical treatment.

As mentioned in the methods section, primary data from linked databases were analysed to estimate repeat surgery incidence rates over time and were then used to estimate the proportion of women who will seek retreatment after failure of the previous surgery. Because the linked database may not be complete, the health economists felt that the rates of repeat surgery were underestimated. In this SA, the method described in Repeat surgery was used to estimate the proportion of women who will seek retreatment. In addition, in this SA lower probabilities of repeat surgeries (50% and 10% for first retreatment and 20% and 0% for second retreatment) were explored.

Applying different values for short- and long-term success rates after retropubic mid-urethral sling

As indicated in the estimation of model probabilities, the short-term and long-term cure rates after retropubic MUS were estimated using a random-effects meta-analysis. There are some uncertainties associated with these estimates and it is probable that these values are either an overestimate or an underestimate because of small sample size. We have explored the effect of using two different sources of data (Ward et al. 248 and Song et al. 249) for extrapolation of long-term success rate after retropubic MUS in this SA. The short- and long-term cure rates were used to estimate an appropriate shape and scale parameter for a Weibull hazard function, and these parameters were used to estimate failure rates at each time point after retropubic MUS, as described in Transition probabilities.

Applying lower success rates when repeating same surgeries

This SA was conducted to explore the effect of assuming lower cure rates when the same surgeries are being conducted for a second or third time. In the base-case analysis it was assumed that all surgical treatments have the same effectiveness as a primary procedure, whereas in this SA the subsequent treatments are assumed to have less effectiveness: 90% and 75%, respectively.

Incorporating different health utility values and effect of natural decline in health utility over time

As mentioned in Health utility, the same values were used for SUI and MUI women using data from Manca et al. 244 and Haywood et al. 245 in the base-case analysis. Tincello et al. 267 showed that MUI (EQ-5D 0.75) may have a greater impact on the EQ-5D score than SUI (EQ-5D 0.85). SA was performed by adjusting the EQ-5D scores used in the base-case analysis with lower values applied to MUI health states. EQ-5D scores used for the MUI pretreatment and MUI after failure of first/second treatment states were 0.69 and 0.65, respectively. The impact of the natural decline in health utility over time was also considered in further SA. The values for the age-related reduction were derived based on published values for age-related health utility. 268

Applying higher incidence rates for mesh complications after mid-urethral sling procedures

In this SA, results from the study by Keltie et al. 266 were used to inform the mesh complication incidence rate after retropubic MUS and transobturator MUS. The results from this SA show that retropubic MUS remains the least costly strategy compared with the alternative surgical interventions and is the most cost-effective option, at a WTP threshold value of £20,000 (Table 20). However, when 10% and 20% incidence rates are used, traditional sling is the most cost-effective option, with an ICER of < £20,000 per each QALY gained.

Incorporating longer duration for persistent pain complication

In the base-case analysis, it was assumed that persistent pain complication will last for, on average, 6 months. In these sensitivity analyses, the effect of longer-term suffering from persistent pain was explored and it was assumed that, on average, women with persistent pain will suffer for 3 and 5 years. The results from these sensitivity analyses show that with longer duration of persistent pain, the probability of retropubic MUS being the most cost-effective option decreases and the respective probability for open colposuspension increases (Table 21).

Applying higher incidence rates of persistent pain after mid-urethral sling procedures

Results from these sensitivity analyses show that when a higher incidence rate (10% and 20%) of persistent pain after retropubic MUS and transobturator MUS was assumed, the probability of retropubic MUS being the most cost-effective option decreases and the respective probability for traditional sling increases (Table 22). The probability of retropubic MUS and traditional sling being the most cost-effective are 23–24% and 28–29%, respectively.

Applying higher incidence rates and longer duration for persistent pain after mid-urethral sling procedures

In this scenario analysis, the impact of incorporating higher incidence rates of persistent pain after MUS procedures and longer mean duration for persistent pain was tested on the base-case results. As would be expected, the probability of retropubic MUS being the most cost-effective with a £20,000 WTP decreases from 26% to 4.2% and the corresponding probability for traditional sling increases from 27% to 30.8%. The estimated ICER for traditional sling is £619 per QALY gained. Open colposuspension is not a dominated strategy and has a 25.8% probability of being cost-effective (Table 23).

Applying different values for short- and long-term cure rates after retropubic mid-urethral sling

When data on long-term cure rates derived from the UK trial248 comparing TVT with Burch colposuspension were applied, the results show that the effectiveness of most of the strategies increased, as women spend more time continent over the modelling time horizon and have lower costs. The probability of retropubic MUS being cost-effective also increased, compared with the base-case analysis. When the data from Song et al. 249 were applied, the findings are broadly similar (Table 24).