A Markov model was developed to simulate the therapeutic management, course of disease, and effect of complications in hypothetical cohorts of OAB patients over a 5-year period (Supplementary Figure 1). Several existing models based on a structure developed by Kobelt et al.  were already available, but were limited to a time horizon of 1 year and did not model treatment pathways after drop-out. Thus, differences in costs and outcomes related to treatments used after withdrawal could not be accurately estimated. We captured the influence of treatment discontinuation and switch on costs and OAB symptoms. This is particularly important because persistence with antimuscarinic agents affects HRQoL, health status and healthcare resource utilisation . HRQoL was assumed to be dependent on both variations in symptoms over time and the side effects of treatment.
The SCORPIO Trial
SCORPIO was a randomised, double-blind, parallel group, placebo- and active-controlled phase III trial carried out in 27 countries in Europe and Australasia to compare the efficacy and safety of mirabegron 50 and 100 mg daily with those of placebo in patients with OAB treated over 12 weeks . The co-primary endpoints were the change from baseline to final visit in the mean number of incontinence episodes and micturitions per 24 h. Assessment of safety and tolerability was a secondary endpoint. A secondary comparison of the efficacy and safety of treatment with tolterodine ER 4 mg daily and placebo was also performed.
In total, 1,987 patients were randomised. Mirabegron produced statistically significant (p < 0.05) improvements (adjusted mean change from baseline [95 % confidence intervals]) compared with placebo in terms of number of incontinence episodes per 24 h (50 mg: −1.57 [−1.79 to −1.35]; placebo –1.17 [−1.39 to –0.95]) and number of micturitions per 24 h (50 mg: −1.93 [−2.15 to −1.72]; placebo −1.34 [−1.55 to −1.12]). Statistically significant improvements were also noted for other key efficacy endpoints and HRQoL outcomes. Incidences of treatment-emergent adverse events were similar across treatment groups, but the incidence of dry mouth with mirabegron 50 mg was similar to placebo (2.8 and 2.6 %, respectively), whereas it was more than threefold higher in patients receiving tolterodine ER 4 mg (10.1 %) .
The model was programmed to run in Microsoft Excel 2007 (Microsoft Corp., Redmond, WA, USA). The treatment pathway, based on the model described above, for the analysis is shown in Fig. 1. Transitions are shown in more detail in Supplementary Figure 1. The model simulated changes in symptoms (frequency of micturitions and incontinence episodes, for which the model was run in parallel) at monthly intervals (i.e. 60 cycles over 5 years). Every month, patients could remain on treatment (mirabegron or tolterodine), switch to a treatment with efficacy and price similar to solifenacin or discontinue (i.e. go to ‘no treatment’). A small proportion received botulinum toxin (BTX) after this next line of therapy (this transition was not allowed directly after mirabegron or tolterodine). The probabilities of switch and discontinuation were dependent on adverse events. Patients with adverse events could stay on treatment, but incurred a disutility. Patients who discontinued treatment could naturally improve and thereby transition to a lower severity category after 1 month, or could worsen or stay the same. These patients could also restart their previous treatment, could move to a new treatment, or remain off treatment. In the event of success on BTX, patients were assumed to move to the lowest level of severity and remain there until the end of the simulation.
The model accounted for the differences in probability of improving or worsening between the short- and long-term. Thus, the probability of improvement was greatest in the first month after treatment initiation, after which it decreased progressively, and was then assumed constant after 3 months.
Several populations were considered. Any patient presenting with OAB formed the base case population. Other populations considered were patients dissatisfied with previous treatment, patients dissatisfied with previous treatment due to lack of efficacy, patients dissatisfied with previous treatment due to intolerance, treatment-naïve patients, elderly patients, female patients, male patients and incontinent patients.
Model Input Parameters
All model input parameters are presented in detail in Supplementary Table 1.
Symptoms (micturition frequency and incontinence) had five levels of severity (Table 1). Initial proportions of patients with symptoms of each severity were derived from SCORPIO . Transition probabilities between symptom levels for mirabegron 50 mg and tolterodine ER 4 mg were estimated by applying multinomial logistic regression models to the SCORPIO trial data (see Electronic Supplementary Material and Supplementary Tables 2 and 3).
Treatment Persistence, Switching and Restarting
Other model inputs are summarised in Table 2. Persistence on mirabegron was extrapolated from the study of Wagg et al. , in which data were extracted via medical practice software and anonymised from the records of >1,200,000 registered patients, of whom 4,833 had documented OAB. An additional 12-week observational study of OAB patients in Spain showed that 24 % of patients who changed treatment did so because of adverse events ; these results were consistent with those of a large US survey of patients with OAB . As no real-life data are available for persistence with mirabegron, the model assumed that discontinuation rates for patients without adverse events were similar for mirabegron and its comparator. The probability of switching to next-line therapy was obtained from an analysis based on the UK general practice research database . The base-case frequency (Table 2) was derived from 5,424 patients who received first-line tolterodine, of whom 68.92 % discontinued and 26.06 % switched to another medication (most frequently oxybutynin).
No data are available in the literature regarding the probability of restarting treatment after a period of no treatment. We assumed an annual probability of 50 % (monthly probability of 5.6 %). The 79 % probability of improvement of symptoms after BTX injection was based on a published cost-effectiveness analysis comparing BTX to antimuscarinic therapy .
Monthly probabilities of adverse events were derived from SCORPIO . Patients were assumed to experience dry mouth or constipation (Table 2). Other adverse events were excluded based on the results of a European cross-sectional survey that showed that these two events are the main drivers of adverse event-related treatment discontinuation with antimuscarinic therapy  and the finding that the events reported most frequently with mirabegron occur at a similar incidence with placebo .
Utility values according to symptom severity and adverse events were derived from EuroQol 5-Dimension (EQ-5D) index scores, based on the UK time trade-off tariff , collected in SCORPIO . A linear regression model of EQ-5D utility as a function of symptom severity was developed as described in detail in the Electronic Supplementary Material and Supplementary Table 4. The utility decrement for adverse events was an average, weighted according to the treatment-specific probabilities for the different adverse events. Utilities derived from the overactive bladder questionnaire (OAB-q), using the algorithm developed by Yang et al. , were also used.
Resource Utilisation and Costs
Resources and associated costs included in the model are from a UK NHS payer perspective (Table 3). Drug therapy, primary care visits, specialist (urologist) visits, BTX injections and incontinence pads were accounted for. Patients were assumed to use one tablet of mirabegron per day; wastage or partial compliance was not accounted for. Pad utilisation was tiered according to incontinence severity.
Costs are presented in 2012 British pounds (£) and are summarised together with sources in Table 3. An annual discount rate of 3.5 % was applied to costs and health benefits.
The model was designed to output (i) average annual cumulative costs by treatment strategy; (ii) quality-adjusted life-years (QALYs) gained by using a particular treatment strategy; (iii) and incremental cost-effective ratios (ICERs) expressed as incremental cost per QALY gained.
Sensitivity and Subgroup Analyses
Sensitivity analyses evaluated the impact of assumptions used in the model and variability surrounding model inputs. For deterministic sensitivity analyses, one variable or assumption was changed at a time. One-way sensitivity analyses were conducted on all model parameters associated with uncertainty: proportions of patients by severity level at baseline; transition probabilities between symptom levels; utilities by symptom level; probabilities of treatment-related events; probabilities related to BTX injections; and probabilities associated with adverse events and resource use. Outcomes were computed using confidence limits around each parameter or other fixed values.
A probabilistic sensitivity analysis was also performed. Appropriate statistical distributions were assigned to input parameters (Table 2). Values were drawn at random from statistical distributions for these variables and the process was iterated 5,000 times to provide distributions for ICERs. Cost-effectiveness acceptability curves (CEACs) were generated at various willingness-to-pay (WTP) thresholds. The probabilistic analysis was programmed in visual basic applications for Excel.
Finally, because indirect costs were not included in the model, a sensitivity analysis including indirect costs valued using the human capital approach to determine the effect on the ICER estimate was also performed. Absenteeism was derived from the percentage work time missed, assessed for each patient in the SCORPIO trial using the work productivity and activity impairment (WPAI) questionnaire .
Subgroup analyses were also performed.