This appendix contains additional details of the methodology, results and sensitivity analysis for the health economic modelling undertaken on drugs to treat overactive bladder with incontinence (OAB wet).
N.1. Structure of the health economic model for OAB drugs
The model is structured around two main model parameters: continence status (the percentage of women who are completely dry), and discontinuation rate (the percentage of women who discontinue treatment for any reason). Two separate models were developed. The first model considered continence status and discontinuation separately as this is how the data were reported in the clinical review presented in chapter 6. A woman with OAB can switch between health states (continence/incontinence) and treatment states (continue on treatment/ discontinue treatment) from week to week up until one year.
In a second model, it was assumed that all women who continue on treatment are continent and all women who discontinue are incontinent. In this version, data on discontinuation rates only was included after 4 weeks. The reason the second model was also developed is twofold: first, it was suggested by stakeholders that continuation and continence should not be modelled separately. Second, the 12-week incontinence derived from the network meta-analysis was based on the 4-week data for all drugs except oxybutynin (immediate release). The 12 week data were calculated by using the same relative effects as 4-weeks anchored to the actual 12-week continence rate for oxybutynin IR. Since the 12-week data for all other drugs were not based on published trial data, a different approach using only discontinuation data after 4 weeks (which was based on trial data) was also suggested to assess whether this changed the overall conclusions of the analysis.
In the first (base case) model, a hypothetical cohort of 1000 women with OAB, start on treatment and can either continue treatment week by week or discontinue treatment. It was not assumed that all women who continue are on successful treatment; some women who continue on treatment remained incontinent. The model was structured so that the probability of being continent and the probability of being on treatment were independently calculated for each cycle.
The pathway was as follows: a woman could be either on treatment and incontinent (treatment failure) or on treatment and continent (treatment success) or discontinuation and be incontinent (treatment failure). As the data from the network meta-analysis reported rates of discontinuation separately from rates of continence, the model was structured so that in each week, a woman could be both in a continence state and in a treatment state. When a woman discontinued treatment, she was assumed to receive no further active treatment for the duration of the model. The model further assumed that a woman who discontinued treatment was incontinent for remainder of the time period of the model (up to one year). This is a simplification of reality in order to compare first line antimuscarinic treatments only.
The structure of this model is not a state transition model as an individual woman is in two states independently in each cycle. It was not possible to calculate exclusive health states for this model because the data on continence status and discontinuation were reported separately for each time point in the network meta-analysis (NMA) (at 4 and 12 weeks).
Cycle length was one month to reflect the usual prescriptions for OAB drugs in the NHS.
The alternative model structure was developed as a Markov chain state transition model. A woman could be in one of three exclusive health states in each cycle from 4 to 52 weeks (thirteen cycles in all). It was assumed that all women who discontinued were incontinent and all women who continued with treatment after 4 weeks were continent.
It was assumed for both models that women started off the model incontinent with a specific probability of being continent at the midpoint of a cycle (calculated as a half cycle adjustment of the 4 and 12 week probabilities).
The monthly rate of discontinuation between 4 and 12 weeks was assumed to be constant and calculated so that the cumulative rate was that reported in the network meta-analysis for each OAB drug. The GDG view was that this is likely to be an underestimate of the rate of discontinuation which is usually higher in the first 4 weeks due to lack of immediate efficacy. As the data available for this period were limited it was not possible to make a different assumption.
The rate of continence and discontinuation at one year was assumed to be constant for all drugs in the first instance with sensitivity analysis using recently published data considered in sensitivity analyses. The monthly rate of continence between 12 and 52 weeks was assumed to be constant.
The following schemas in and shows the structure of the two health economic models developed for this analysis. and report the equations for deriving the state transitions.
Treatment status and continence status are modelled independently in 4 week cycles up to one year (13 cycles). Data derived from network meta-analysis of trial data at 4 and 12 weeks,
Treatment status and continent status are combined into discrete health status. After the first 4-week cycle, continence is determined by continuation rate only.
Equations for transition probabilities for model structure 1.
Equations for transition probabilities for model structure 2.
N.3. Inputs into the health economic model
Parameters used in the PSA
All model parameters are reported in tables N.3 to N.6
Parameters for the baseline distributions (oxybutynin immediate release) from the NMA used in the probabilistic sensitivity analysis.
Parameters for the comparator distributions (all OAB drugs other than oxybutynin immediate release) from the NMA used in the probabilistic sensitivity analysis.
Utility distributions used in the probabilistic sensitivity analysis.
Calculation of four-week costs for drugs to treat OAB wet used in the model.
Effectiveness and discontinuation parameters were derived from the network meta-analysis. A normal distribution was assumed for the log-odds of the baseline comparator drug which was oxybutynin (immediate release) at 4 and 12 weeks.
Relative treatment effects (continence status) were sampled from the posterior distributions of the log-odds ratios of every alternative drug compared with oxybutynin (immediate release). These log-odds ratios were then applied to the 20,000 simulated log-odds of oxybutynin (immediate release). It was then assumed that the log-odds were normally distributed with the mean and variance estimated from the meta-analysis. In the main network meta-analysis program, the estimated log-odds ratios were applied to the baseline log-odds of oxybutynin (immediate release) to simulate the absolute probabilities for the health economic model at 4 weeks. This was repeated for 12 week data for continuation. For continence status at 12 weeks, the log-odds of continence were calculated from a meta-analysis of the oxybutynin (immediate release) arms of the selected trials. The same relative effects as 4 weeks were then assumed for each OAB drug, anchored to oxybutynin (immediate release) data for 12 weeks. Because the 12-week probabilities for continence status for all other drugs were not derived from trial data in the NMA (as was the case for discontinuation probabilities), a second model was assumed that did not use 12-week continence status probabilities derived from the NMA.
The mean probabilities for each of the OAB drugs are reported in table 6.9 in the chapter 6.
Four-weekly drug costs were calculated from the NHS drug tariff (June 2013). http://www.ppa.org.uk/ppa/edt_intro.htm. Prices were cross-checked with the BNF for the same time period. There was good agreement on prices from these two sources. The NHS tariff was used for all drugs except trospium extended release which had no published price in the NHS Drug Tariff and the BNF price was used. The trials were selected on the basis that they used the starting dose reported in the BNF. Where a range of doses was reported as starting dose in the BNF, the dose used in the trials was assumed (that is, for propiverine immediate release). Oxybutynin topical gel is not marketed in the UK, and therefore not reported in either the BNF or NHS Tariff. In that case, the price of the closest alternative was assumed (oxybutynin transdermal).
First line versus second- and third-line treatment
The network meta-analysis reported continence status for OAB drugs as first-line treatment. Efficacy data were not reported for OAB drugs as second line therapy once a first-line drug had failed. After reviewing the evidence and receiving stakeholder comments, it was the view of the GDG that it could not be assumed that the efficacy of an OAB drug would be the same for a woman for whom OAB drug therapy had not achieved continence using a different drug. For that reason, a cost-effectiveness analysis of drug treatments as second line therapy that had been presented in the draft guideline was removed from the final version.
Calculating the average cost of drugs where two doses were reported
The only drug for which dose titration was reported and doses had different published prices was solifenacin. To calculate the weighted average cost of solifenacin, data on patient use of different doses was extracted from the studies included in the clinical review and the proportion of women on the different doses calculated (see below).
Calculation of the weighted average cost of solifenacin based on doses reported in the clinical trials included in the NMA.
In Chapple et al., 2005, 276/578 (48%) increased dose of solifenacin from 5mg to 10mg with an average dose of 7.4mg at endpoint. In Chapple et al., 2005, 276/578 (48%) increased dose of solifenacin from 5mg to 10mg with an average dose of 7.4mg at endpoint. In Vardy et al., 2009, 211/385 (55%) increased solifenacin dose from 5mg to 10mg (10 decreased) − average dose at endpoint = 8.9mg In Karram et al., 2009, 225/372 (60%) dose to increased dose of solifenacin from 5mg to 10 mg after two weeks average dose = 8mg
Estimates of quality of life
Quality of life weightings (or ‘utilities’) for continence and incontinence have been published in the health economic literature ( below). There have also been data published on quality of life weightings of women experiencing varying degrees of severity of OAB symptoms. Severity was measured in the published studies as episodes of micturition (voluntary) and leakage (involuntary) per day. Where a health economic model is based on an individual patient level data in a clinical trial that reports episodes of micturition and leakage this approach may be feasible. The advantage of using levels of severity in a model is that it can capture improvements in quality of life that do not lead to complete recovery. The GDG had two concerns about this approach: First that episodes of micturition and leakage reported together may underestimate the severity of the condition (if most reported episodes were involuntary leakage then the woman would have a more severe condition than a woman whose episodes were mainly voluntary micturition). Second, an improvement in episodes of urgency or incontinence that did not lead to continence may be overstated as many of the problems women experience with carrying out their lives (leaving the home, working, socialising) would remain with any level of incontinence regardless of severity.
Quality of life weightings (“utilities”) for urinary incontinence identified in the literature.
Therefore, the GDG chose to focus on continence status as the primary outcome in the health economic modelling. The GDG view was that drugs that lead to the most improved continence status (“absolutely dry”) would also be the drugs leading to the most improved symptoms without achieving continence. Therefore, the cost-effectiveness of drugs is likely to be underestimated in this analysis.
A recent health technology appraisal reported values based on a systematic review of quality of life studies (Imamura et al., 2010). This study reported a review of the literature on values and reported the quality of life of weightings for “the success of treatment” and “the failure of treatment”. These weightings were used in this model as the estimates for the health states ‘continent’ and ‘incontinent’. Other published studies reported values within this range for women with different levels of severity of incontinence which supports the robustness of these estimates. Quality of life estimates reported in other studies are presented below.
Health service costs apart from drugs
The economic analysis was conducted from the perspective of the UK NHS and includes the pharmacological costs and incontinence pad use. All costs were valued at 2011/12 prices and all dosages are based on the starting dose in the British National Formulary (April 2013). Pads use is included in the model as pads can be supplied by the NHS, but this is not always the case as the source of funding for continence pads varies across the country. They constitute a large part of the cost of incontinence which is saved by effective treatment.
Costs were derived from the number of women who are in treatment in any week (drug costs) and the number of women who are incontinent in any week (pad use) A woman who continues to experience incontinence incurs weekly costs for both her drug treatment and continued use of pads. Additional GP visits were included in the sensitivity analysis for women who discontinue treatment.
A GDG member obtained figures for the cost of continence pads from one English county. The weekly cost of supplying disposable continence products was £13,748, or £4.93 per woman. However, as this estimate was NHS costs only and did not include on-going lifestyle advice (for example from physiotherapists and continence advisors or in primary care), or physiotherapy sessions, the GDG considered this to be an underestimate of the total average cost per woman per week of incontinence as it was unlikely to include all overheads. In the first ‘base case’ analysis (before assessing the impact of changing key variables on cost-effectiveness), the cost of incontinence was estimated to be £8 per week which was the first estimate used in the economic model. This value was varied in the model to see if it changed the relative cost-effectiveness of the drugs. If it was found to be an important driver of the relative cost-effectiveness of drugs, then it would be important to ensure that the cost was accurately estimated. If a wide margin of error does not change the relative cost-effectiveness of drugs then an accurate estimate is less important.
The value of £8 per week is similar to that reported in the US study by Subak and colleagues who reported an annual cost of incontinence of $750 per person (Subak et al., 2006). This cost was varied in sensitivity analysis.
The cost of GP and specialist consultations were not included in the base case as it was not considered that these costs would differ by use of different drugs. This assumes that all women who start OAB drugs require follow-up in primary care whether they continue with drug treatment or not. The cost of treating adverse conditions was not included in the model (see ‘Limitations of the analysis’ for further discussion).
The original (base case) model assumed no additional costs associated with the side-effects of treatment. This was based on the GDG view that adverse events lasting longer than the period of treatment were very rare and the consequences of the most common adverse events (such as dry mouth and constipation) would not have a lasting effect on quality of life. However, women who discontinue treatment due to adverse events may require additional primary care support and this was not included in the base case analysis. Also, in very rare cases, there has been concern raised that oxybutynin (immediate release) may be associated with acute delirium requiring hospitalisation. The cost of incontinence pads was provided by a GDG member. The cost was varied to assess its impact on the results.
To take these additional costs into account, a further sensitivity analysis was undertaken with the following variations in the model:
1 additional GP visit every eight weeks for women who discontinue antimuscarinic treatment and are no longer on active treatment for the remainder of the model. Cost per GP surgery consultation, £40 (PSSRU 2012)
A hospitalisation episode for delirium in all age groups, calculated for 1 in 5000 women on oxybutynin (immediate release) was assumed. This was a GDG estimate based on the published literature on incidence of acute delirium with OAB drugs. The cost of hospitalisation was based on HRG code WD11Z (All Patient over 69 years with a mental health primary diagnosis treated by a non-specialist mental health service provider), local tariff £2029. Data provided by a GDG member based on local hospital data.
Cost of incontinence (continence pads) was varied from £2 to £10 per week in sensitivity analyses
Longer term data in the health economic model
The systematic review of antimuscarinic therapy presented in chapter 6 reported data on:
discontinuation rates for any reason at 4 and 12 weeks,
discontinuation rates due to adverse effects alone at 4 and 12 weeks, and
rate of adverse effects at 4 and 12 weeks.
A systematic review was not undertaken to identify studies reporting discontinuation rates at 52 weeks since the GDG members already knew that data trials did not routinely report follow-up at one year. A separate literature review was undertaken for the health economic analysis to identify non RCT evidence that reported longer term follow-up. below reports the data identified in the review on discontinuation rates reported in the literature for women with OAB treated with OAB drugs.
Long-term (up to one year) discontinuation rates reported in recent health economics studies.
The studies report high discontinuation rates for all OAB drugs at one year with discontinuation for all drugs reported to be around 73%.
Before the recent UK study had been published, the GDG agreed to assume a discontinuation rate of 80% for all OAB drugs at 12 months. Once the new study was published that demonstrated varying discontinuation rates at 12 months for some of the drugs, this was included in the health economic model as a sensitivity analysis to assess whether discontinuation at 12 months was an important variable in the analysis, that is, changing the order of cost-effectiveness. If this turned out to be the case, then a more careful scrutiny of 12 month data on the quality of the published studies would be required to come up with a more robust estimate.
For preparations that were not included in the published study, three scenarios were explored:
The first sensitivity analysis assumed that the rates for the OAB drugs with missing data were the same as the lowest discontinuation rates reported (for solifeancin 65%).
The second sensitivity analysis assumed that the rates for the OAB drugs with missing data were the same as the highest discontinuation rates reported (for darifenacin 83%)
The third sensitivity analysis assumed that the rates for the OAB drugs with missing data were the midpoint between the highest and the lowest discontinuation rates (74%)
N.4. Description of probabilistic sensitivity analysis
Probabilistic sensitivity analysis is an approach to evaluating the robustness of cost-effectiveness results. When there are many input parameters, NICE recommends using probabilistic sensitivity analysis to characterise uncertainty. This allows several parameters to be varied simultaneously, rather than one at time as in one-way sensitivity analysis. In a probabilistic sensitivity analysis each input is assigned a probability distribution which is defined by measures of variability (such as standard deviations). A simulation is then set up to sample inputs at random from their assumed distributions. The simulation is run a large number of times (20,000 times for this probabilistic sensitivity analysis). If one option is consistently more cost-effective than the others then there is high probability that that option will always the most cost-effective option and decision-makers can have a higher level of confidence in the results of the analysis.
The mean four-week and 12-week continence rates were calculated from 20,000 probabilities generated by the NMAs. For one year discontinuation and continence rates were assumed to be constant across all OAB drugs due to the lack of long term data. Recent data were inputted in one of the sensitivity analyses (see below). No distribution could be calculated so is not included in the probabilistic sensitivity analysis.
Net benefit is another way of reporting and ordering the cost-effectiveness of alternatives by making costs and benefits into equal units. For each simulation, it uses a given willingness to pay threshold to value QALYs in monetary terms and then offsets the costs of alternatives against that valuation to arrive at a net monetary value for each intervention.
The network meta-analysis (NMA) provided baseline probabilities of continence at 4 weeks and baseline probabilities of continence at 12 weeks on oxybutynin (immediate release) which was the reference case in the NMA (since it was recommended in the previous version of the guideline). The same relative effects were then applied to obtain the absolute probabilities at 4 and 12 weeks for the other treatments.
N.5. Results
The data reported below are the base case results in more disaggregated form and the results of the sensitivity analyses
Base case results
The first table presented below is the costs and QALYs for each OAB drug presented in a disaggregated form for the base case analysis. The total values differ slightly from that reported in chapter 6 because the results in the chapter show the mean results of the PSA and the data reported here are from the deterministic model from which disaggregated values could be obtained.
Mean costs and QALYs reported in base case PSA model with a constant rate of discontinuation at 52 weeks of 80%, disaggregated (granulated) results.
Cost-effectiveness acceptability curve for first-line antimuscarinic therapy to treat OAB wet, 20,000 simulations.
Consequences of changes in the model structure
Given that the probabilities generated by the NMA data after four weeks were not based on actual 12 week trial data for continence, a sensitivity analysis was undertaken that assumed that all women who discontinued treatment were incontinent and all women who continued were continent (using the model described as structure 2 above). This sensitivity analysis was undertaken assuming a constant 80% discontinuation at 52 weeks across all OAB drugs. .
In the second model structure, the continence rate was only used to calculate health states for the first cycle. In the subsequent cycles, discontinuation data only were used to determine health state after the initial four-week cycle. Results are presented below.
Mean cost per QALY using base case PSA model of 20,000 simulations with a constant 80% discontinuation rate for all drugs at one year with discontinuation data only after 4 weeks.
Cost-effectiveness acceptability curve for first-line antimuscarinic therapy to treat OAB wet using only discontinuation probabilities after four weeks, 20,000 simulations.
The probability that oxybutynin immediate release is the most cost-effective is lower (57% versus 65% in the base case) but the two most cost-effective treatments are still oxybutynin immediate release and tolterodine immediate release. Drugs with better continuation rates do not have a higher than 6% change of being cost-effective and solifenacin which has a better continuation rate has zero chance of being cost-effective.
To illustrate the impact of different discontinuation rates over 13 cycles, a Markov trace showing the proportion of women in each health state by cycle is reported below for two of the OAB drugs that had different discontinuation rates at 4 and 12 weeks, oxybutynin immediate release and solifenacin.
Markov trace for oxybutynin (immediate release) with model structure 2 (discontinuation probabilities only after 4 weeks).
Markov trace for solifenacin with model structure 2 (discontinuation probabilities only after 4 weeks).
PSA changes to model inputs - results
Sensitivity analysis 1
Changes in the assumption about longer-term use of drug therapy to treat OAB wet
Mean cost per QALY using base case PSA model (assuming lowest values for drugs with missing data (best case scenario).
Cost-effectiveness acceptability curve for first-line antimuscarinic therapy using recently published one-year discontinuation of treatment (assuming lowest values for drugs with missing data (best case scenario), 20,000 simulations.
Sensitivity analysis 2
The data from a recent UK study was used for one year discontinuation (Wagg et al., 2012). The base case results reported in the chapter assumed that discontinuation rates for drugs with missing data was the median value of all reported drugs. In the first sensitivity analysis, the missing data was assumed to be the same as the highest discontinuation rates reported; the second assumption was that missing data were the same as the worst reported discontinuation rates reported. The results of using published discontinuation with the different assumptions for missing data are reported below. It shows that the two most cost-effective drugs did not change using this dataset, regardless of which assumptions were adopted about missing data. The probability of any other drug being the most cost-effective remained below 10%.
Cost per QALY using base case PSA model assuming highest values for drugs with missing data (worst case scenario).
Cost-effectiveness acceptability curve for first-line antimuscarinic therapy using recently published one-year discontinuation of treatment (assuming highest values for drugs with missing data (worst case scenario), 20,000 simulations.
One way sensitivity analysis – results
The results below show that oxybutynin immediate release and tolterodine immediate release remained the most cost-effective OAB drug options in all scenarios explored in one-way sensitivity analysis. Taking into account additional costs for regular health service consultations for women who discontinued treatment or were on no treatment did not make drugs with better discontinuation rates relatively more cost-effective. The tables below report the results and further discussion of the results is presented in the evidence to recommendations section in chapter 6.
Sensitivity analysis 3 – GP costs
Cost per QALY with base case PSA model assumptions, with a constant 80% discontinuation rate for all drugs at one year with additional health service consultations for women who discontinue treatment.
Sensitivity analysis 4 – increased risk of hospitalisation associated with oxybutynin IR
Cost per QALY with base case model assumptions with a constant 80% discontinuation rate for all drugs at one year with additional hospitalisation costs for oxybutynin IR.
Sensitivity analysis 5 – higher and lower estimate of weekly incontinence pads costs
Cost per QALY with base case model assumptions with a constant 80% discontinuation rate for all drugs at one year with lower costs for incontinence pads (£2 per week).
Cost per QALY with base case model assumptions with lower costs for incontinence pads (£10 per week).
Sensitivity analysis 6 – Extending the model time horizon using model structure 1 (continence status and discontinuation modelled independently)
For the following models, it was assumed that women do not change their treatment status or continence status after 12 months.
Cost per QALY at 2 years with base case model (model structure 1), using Wagg data for discontinuation at 12 months, assuming no change in health status or treatment status after 12 months.
Two-year cost-effectiveness acceptability curve for first-line antimuscarinic therapy in model structure 1 using recently published one-year discontinuation of treatment (assuming midpoint values for drugs with missing data), 20,000 simulations.
Cost per QALY at 5 years with base case model (model structure 1), using Wagg data for discontinuation at 12 months, assuming no change in health status or treatment status after 12 months.
Five-year cost-effectiveness acceptability curve for first-line antimuscarinic therapy in model structure 1 using recently published one-year discontinuation of treatment (assuming midpoint values for drugs with missing data), 20,000 simulations.
Sensitivity analysis 7 – Extending the model time horizon using model structure 2 (continence status dependent on treatment status after the first 4-week cycle)
Cost per QALY at 2 years using model structure 2, using Wagg data for discontinuation at 12 months, assuming no change in health status or treatment status after 12 months.
Two-year cost-effectiveness acceptability curve for first-line antimuscarinic therapy in model structure 2 using recently published one-year discontinuation of treatment (assuming midpoint values for drugs with missing data) , 20,000 simulations.
Cost per QALY at 5 years using model structure 2, using Wagg data for discontinuation at 12 months, assuming no change in health status or treatment status after 12 months.
Five-year cost-effectiveness acceptability curve for first-line antimuscarinic therapy in model structure 2 using recently published one-year discontinuation of treatment (assuming midpoint values for drugs with missing data), 20,000 simulations.
