NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

National Clinical Guideline Centre (UK). Chronic Obstructive Pulmonary Disease: Management of Chronic Obstructive Pulmonary Disease in Adults in Primary and Secondary Care [Internet]. London: Royal College of Physicians (UK); 2010 Jun. (NICE Clinical Guidelines, No. 101.)

Cover of Chronic Obstructive Pulmonary Disease

Chronic Obstructive Pulmonary Disease: Management of Chronic Obstructive Pulmonary Disease in Adults in Primary and Secondary Care [Internet].

Show details

Appendix MNEW 2010 update cost effectiveness modelling

A cost-effectiveness model comparing LAMA, LABA+ICS, and LAMA+LABA+ICS (triple therapy) in people with severe/very severe COPD requiring initial maintenance therapy

Model overview

The GDG were interested in the following question: Is LAMA, LABA+ICS or triple therapy more cost-effective as initial therapy in COPD patients with an FEV1 <50% predicted (severe to very severe COPD)?

A cost-utility analysis was undertaken where costs and quality-adjusted life-years (QALYs) were considered from a UK NHS perspective. Both costs and QALYs were discounted at a rate of 3.5% per annum in line with NICE methodological guidance37.

Topic selection for modelling

Areas were prioritised for new analysis by the GDG. The GDG was interested in assessing the cost-effectiveness of alternative regular maintenance therapies (or combinations of such therapies) for people with stable COPD. Due to complexities in the clinical data it was judged unfeasible to adequately conduct an analysis looking at all possible inhaled interventions in all treatment scenarios within the scope of the guideline update. This included the following issues:

  • There were inconsistencies in the clinical evidence network i.e. seemingly contradictory relative risks
  • The maintenance therapy decision is not a one off decision – there is the initial decision and then subsequent decisions about adding in additional therapy. Clinical trials generally do not match a particular scenario, i.e. initial maintenance treatment or patients on a specific treatment who are experiencing symptoms, but instead recruit COPD patients meeting variable criteria and randomise to therapy – this makes explicit consideration of the initial decision and subsequent decisions muddied (for example we have information about using triple therapy but not separately for using it straight away and using it after using other therapies but still experiencing symptoms).

The aim was to therefore undertake a focussed analysis that would be useful to the guideline and inform decision making. Following review of the clinical evidence and published economic literature it was considered that examining the following question was the highest priority: is LAMA, LABA+ICS or triple therapy more cost-effective as initial therapy in COPD patients with an FEV1 <50% predicted (severe to very severe COPD)?

These treatment options were selected as those that represent the most appropriate possible clinical options for people with COPD and an FEV1 <50% predicted. The GDG felt that the clinical and cost-effectiveness literature suggested that LAMA or LABA+ICS were probably the appropriate options for initial maintenance therapy for patients with an FEV1 <50% predicted. However, it was felt that if triple therapy could be justified on cost-effectiveness terms that it might be considered as an initial therapy. Therefore these options were incorporated into the model. It was felt unnecessary to include LABA as there was good existing evidence that use of LABA+ICS over LABA alone was more effective and cost-effective in this patient group. No data was available for LAMA+ICS as a treatment option and so it was considered inappropriate to include in the model. Clinical effectiveness data for LAMA+LABA was considered insufficient for it to be considered a primary treatment option and it was felt that it would only be appropriate to consider in patients in whom ICS was declined or not tolerated. On this basis, it was felt that inclusion of LAMA+LABA was also not a priority for inclusion in the model.

It was felt that in less severe patients (FEV1 ≥50% predicted) the key issue was whether to use LAMA or LABA in initial therapy but that issues with the available clinical data would mean that new health economic modelling would be unlikely to reduce uncertainty around this decision and so was considered less of a priority for modelling.

The analysis aimed to consider initial maintenance treatment. It did not incorporate changes to therapy over time. This was judged to be a pragmatic approach given the available data.

Approach to modelling

A Markov model was constructed describing how a population with COPD changes over time. Specifically, this represents an increase in mortality and exacerbations over time, and a reduction in quality of life, as patients' lung function declines. The Markov model consisted of three mutually exclusive health states: severe COPD (FEV1 30 to <50% predicted), very severe COPD (FEV1 <30% predicted) and dead. Patients can progress from severe to very severe COPD; they cannot regress in COPD severity. A cycle length of one year was used. Different exacerbation and hospitalisation rates, mortality rates, utilities and maintenance costs are assigned to each COPD severity stage.

Table 1. Markov model depiction.

Table 1

Markov model depiction.

For the baseline, we populated the model with data relating to the LABA+ICS treatment group. Running the model estimates outcomes over a specified time period. By applying cost and utility weights we estimated mean costs and QALYs over the whole time period.

To compare the impact of treating the same population with a different treatment option we applied relative treatment effects from RCTs for each treatment option to the baseline estimates in the model, reran the model and then recalculated mean costs and mean QALYs.

Comparing these mean results for the three different treatment options allowed us to identify which was the most cost-effective.

Analyses undertaken

Outcomes incorporated into the model were based on the systematic review of the clinical effectiveness data and GDG discussion. The aim was to incorporate key outcomes that are differentially impacted by treatment across the treatment options being considered by the model and that result in differences in costs and/or QALYs.

The basecase analysis incorporates a differential treatment effect in terms of exacerbations. Exacerbations in the model are attributed a cost and a utility loss (quality of life impact) and so impact costs and QALYs. This was considered the most robust assessment that could be made based on the available data. Some EQ-5D utility data was available from the literature to inform the estimate of the impact of exacerbations.

  • Basecase analysis (exacerbation effect only):
    • Outcomes impacted by treatment:
      • exacerbations (non-hospitalised)
      • exacerbations (hospitalised)
    • Costs will vary between treatment options due to differences in drug costs and exacerbations between treatment options.
    • QALYs will vary between treatment options due to differences in exacerbations between treatment options – each exacerbation is associated with a QALY loss; so if the number of exacerbations varies between treatments then so will the QALYs.

An alternative analysis was undertaken that incorporated a differential treatment effect in terms of stable utility (quality of life) as well as exacerbations. This was not included in the basecase due to concerns regarding estimating this effect. Model inputs are discussed in detail in subsequent sections.

  • Alternative analysis 1 (exacerbation and stable utility effect):
    • Outcomes impacted by treatment:
      • exacerbations (non-hospitalised)
      • exacerbations (hospitalised)
      • quality of life during stable COPD (due to improved symptoms with treatment)
    • Costs will vary between treatments as in the basecase analysis.
    • QALYs will vary as in the basecase analysis but also due to the difference in utility between treatment arms whilst patients are stable.

Careful consideration was given to whether or not it was appropriate to incorporate a differential treatment effect in terms of mortality. It was generally considered that there was not currently strong evidence to support a differential mortality effect between the treatments being considered in the model but that it was plausible given the effect of treatments on exacerbations. Many studies were also not powered to detect a mortality effect. It was concluded that it would be most appropriate to run the analysis both excluding and including mortality. As such, a second sensitivity analysis was undertaken where mortality was differentially impacted between the treatments in the model, in addition to exacerbations.

  • Alternative analysis 2 (exacerbations and mortality effect):
    • Outcomes impacted by treatment:
      • exacerbations (non-hospitalised)
      • exacerbations (hospitalised)
      • mortality
    • Costs will vary due as in primary analysis but COPD maintenance costs will also vary between treatment options as there will be different numbers of people alive with each treatment option due to differences in mortality.
    • QALYs will vary as in primary analysis but there will also be a difference in life years between treatment options due to the different mortality with the treatment options.

Note that progression was assumed not to be impacted differentially between the treatments being compared.

Time horizon

In all the above analyses, a treatment duration of four years was examined. This matches the longest follow-up of the clinical trials that inform the comparisons in this model.

As sensitivity analyses, we also examined the effect of using a shorter time horizon of 1 year (matching the shortest follow-up of the clinical trials that inform the comparisons in this model) and a longer time horizon of a lifetime (35 cycles).

In the basecase and first alternative analysis, where a differential treatment effect on mortality was not incorporated, it was expected that the time horizon would not have a large impact on results. In the analysis that included mortality however it was considered that it may have a greater impact. When mortality is impacted differentially between treatments there are a different numbers of people alive at the end of the four year treatment period. Due to this, even assuming no further differential treatment impact, costs and QALYs therefore vary between treatment options beyond 4-years.

Uncertainty

The model was built probabilistically in order to take account of the uncertainty around input parameter point estimates. A probability distribution is defined for each model input parameter. When the model is run a value for each input is randomly selected from its respective probability distribution simultaneously and costs and QALYs are calculated using these values. The model is run repeatedly – in this case 5000 times – and results are summarised. Probability distributions in the analysis were based on error estimates from data sources, for example confidence intervals around relative risk estimates.

In addition to the sensitivity analyses already described above around the outcomes incorporated in the model and the time horizon, various additional sensitivity analyses, where one or more inputs were varied, were undertaken to test the robustness of model assumptions and data sources.

Model inputs

Inputs summary table

Model inputs were selected following a review of the literature and validated with the GDG. Note that healthcare utilisation defined exacerbations were used in the model. Point estimates and the 95% confidence interval for inputs are shown in the table; the latter to illustrate the range of values taken in the probabilistic analysis. Confidence intervals are as reported from the data where available (for COPD utility and relative treatment effects for exacerbations, hospitalisations and mortality), where not reported or where the input value in the table below is the result of a calculation the confidence interval shown is generated from 10,000 simulations of the probabilistic analysis. Where no confidence interval is presented the input was not varied in the probabilistic analysis. More details about sources and any calculations can be found in the sections following this summary table. Details of the probability distributions used for the probabilistic analysis are also included in subsequent sections.

Table 2. Summary of model inputs – point estimates and 95% confidence intervals.

Table 2

Summary of model inputs – point estimates and 95% confidence intervals.

Initial cohort setting

The cohort is assumed to have a starting age of 66 years and be 46% female. The former is based on the average age in the three trials utilised in the model for treatment effects197,200,219. The latter is based on a published analysis of UK GP records15.

The analysis considers a population of people with COPD and an FEV1 less than 50% predicted (that is people with more severe disease). On entering the model the cohort is distributed as 67% severe (FEV1 30 to <50% predicted) and 33% very severe (FEV1 <30% predicted). This was based on the estimated distribution of severity stages in people diagnosed with COPD in England from an analysis undertaken by the Department of Health609.

Progression

The annual transition probability for progression from severe (FEV1 30% to <50% predicted) to very severe (FEV1 <30%) in the model was derived based on a mean decline in FEV1 of 39ml/year (SE 0.003) as reported in the TORCH study in the LABA+ICS arm610. The mean annual decline was incorporated into the probabilistic analysis using a gamma distribution. Details of calculations and data selection are provided below.

Note that no differential effect between the three treatment options in the model was applied to disease progression as the GDG felt that current evidence did not support this. This means that the time spent in the severe and very severe severity states only varied between treatment options in the secondary analysis where mortality was impacted.

A non-systematic review of the literature identified a variety of potential sources of data for the annual decline in lung function, including cohort studies and randomised controlled trials. Data from a selection of key studies are summarised in Table 3. There is some evidence of a significant difference in decline in FEV with pharmacological treatment compared to no treatment (notably in the TORCH study)610. On this basis it was considered that an ‘on-treatment’ rate of decline was most appropriate to use in the model as all comparators were active treatments. Given that TORCH was a large study with 3-years of follow-up this was considered an appropriate source of data.

Table 3. Selected studies of COPD lung function decline.

Table 3

Selected studies of COPD lung function decline.

The probability of transitioning from severe (FEV1 30 to <50% predicted) to very severe (FEV1 <30% predicted) was calculated as follows.

A typical patient in the severe (FEV1 30 to <50% predicted) was attributed the following characteristics:

  • male – based on UK GP records15
  • aged 66 years – the average in the trials used in this analysis for treatment effects197,200,219
  • 1.75m tall – the average male height in the UK630
  • an FEV1 40% of predicted – the midpoint of the range in this group and the mean in this group in the TORCH study207.

A male, aged 66 years, of height 1.75m and with an FEV1 of 40% his predicted FEV1 must have an FEV1 of 1.27 according to the European Respiratory Society 1993 reference equations45. Assuming a decline of 39ml/year in FEV1 we calculated his FEV1 for subsequent years. His predicted FEV1 in corresponding years was also calculated using the same reference equations as above. His resulting FEV1 % predicted was then calculated for each year by dividing his FEV1 by his predicted FEV1. The resulting figures are displayed in Table 4. On this basis, he would reach the very severe stage (FEV1<30%) in 10.4 years.

Table 4. Modelled FEV1 decline for male aged 66, height 1.76m, FEV1 40% predicted and a decline of 39ml/year.

Table 4

Modelled FEV1 decline for male aged 66, height 1.76m, FEV1 40% predicted and a decline of 39ml/year.

It was then assumed he represents the median patient and that on average 50% of the population would have progressed by 10.4 years. Therefore in the population there would be a 50% probability of progressing in 10.4 years. Assuming a constant hazard the instantaneous rate was calculated as:

Annual rate=ln(1p)t=ln(10.5)10.4=0.0664

Where: p = the proportion of patients that progress over time period t.

This was then converted from an annual rate to an annual transition probability using the standard formula:

Probability of progressing(moderateto severe)=1e-rt=1e-0.0664×1=0.0642

Where: r = rate; t = time period

Baseline event rates with LABA+ICS

The model must be populated with appropriate event rates for one of the comparators in the model (baseline events). Event rates for the other comparators are then calculated in the model by applying relative effect figures from randomised controlled trials. The model was populated with baseline event rates for LABA+ICS.

Exacerbations

Overall average annual exacerbation rates of 0.91 (SE 0.023) per person per year for severe (FEV1 30 to <50% predicted) and 1.54 (SE 0.051) per person per year for very severe (FEV1 <30% predicted) were applied in the model for people treated with LABA+ICS. This was based on rates observed in the TORCH study LABA+ICS arm in these FEV1 groups and imputed error estimates (see below)207. Hospitalisation rates for exacerbations were not reported by GOLD stage and it was assumed that 19% of all exacerbations required hospitalisation as observed in the TORCH LABA+ICS arm197. This equated to an average of 0.17 per patient per year and 0.29 per patient per year for severe and very severe respectively. Note that healthcare utilisation defined exacerbations were used in the model. Exacerbation rates were incorporated into the probabilistic analysis using log normal distributions.

Error estimates were not reported for the exacerbations rates by FEV1 severity stage. In order to incorporate uncertainty around the exacerbation rate into the model a standard error was imputed based on the reported mean rate for each severity stage and the estimated total patient years. Total patient years were estimated using the number of patients for each severity stage (GOLD 3 = 728; GOLD 4 = 243) multiplied by the average patient follow-up for the TORCH study as a whole (2.4 years). The following formula for the standard error of a rate was then used:

SE rate=ratetotal patient years

Baseline exacerbation rate data stratified by FEV1 was sought through a non-systematic review of the literature. The TORCH study data was selected as it provided stratified rates from a large cohort for people treated with LABA+ICS207. Rates were also similar to those observed in the clinical trials being used in the model for relative treatment effect. It included 728 LABA+ICS patients FEV1 30-49% predicted and 243 FEV1 <30% predicted. Donaldson et al. also reported stratified rates from a UK cohort however the population was smaller and rates were not specific to any one treatment631. A Spanish and a Swedish cohort study were also identified496,632.

Mortality

Age-dependant mortality was incorporated into the model using life tables for England and Wales and severity specific COPD mortality data611,612. A relative risk for mortality with COPD was applied of 3.1 and 5.0 for severe (FEV1 30 to <50% predicted) and very severe (FEV1 <30%) stage respectively611.

COPD severity specific mortality data was reported by Ekberg et al. based on a Swedish population study with 22,044 people611. Relative risks were presented for smokers, former smokers and never smokers stratified by GOLD COPD severity stage and gender compared to the general population without symptoms of chronic bronchitis and with normal pulmonary function (see Table 5). A weighted average of the reported GOLD stage 3 (FEV1 30 to <50% predicted) and GOLD stage 4 (FEV1 <30% predicted) figures were used in the model. These inputs were incorporated into the probabilistic analysis using log normal distributions.

Table 5. COPD mortality risks in GOLD stages 3 and 4 compared with the general population.

Table 5

COPD mortality risks in GOLD stages 3 and 4 compared with the general population.

COPD mortality rates stratified by FEV1 were sought through a non-systematic review of the literature. The Ekberg et al. data was selected as it provided relative estimates of COPD mortality by FEV1 group compared with a general population611. Soriano et al. also reported stratified COPD mortality rates compared to a matched control group from a UK cohort15. COPD severity was however classified as mild, severe and very severe by prescribed drugs and the Ekberg FEV1 stratified data was considered more appropriate for the model. Both studies found higher COPD mortality in the more severe groups. Other COPD mortality data was identified but was either not stratified, not compared with a control non-COPD group or the source of estimates was unclear.

Utilities (health-related quality of life)

QALY loss per exacerbation

Exacerbations drive the differences in QALYs between treatment options in the basecase analysis. Each hospitalised exacerbation was attributed a QALY loss of 0.020 and each non-hospitalised exacerbation was attributed a QALY loss of 0.011. The basis for this is described below.

In order to estimate the impact of COPD exacerbations on QALYs, information was required on the magnitude of effect on utility during an exacerbation and the duration of effect.

COPD utility data was sought by searching Medline using COPD and EQ-5D specific terms and reviewing previous cost-utility analyses. As limited data was identified further ad hoc searches looked more broadly for information about the impact of COPD exacerbations on quality of life. A review of health-related quality of life data (including utility and non-utility measures) in COPD was identified and checked for useful papers633.

Two studies were identified that looked at utility change during exacerbations of COPD:

  • Paterson and colleagues evaluated utility using EQ-5D in patients with an established diagnosis of chronic bronchitis and who presented at a general practice clinic with an acute exacerbationkkk615. The study enrolled 81 patients at a single centre in Glasgow, UK. The UK tariff for the EQ-5D was used. They reported a mean increase in EQ-5D of 0.17 (SD 0.24) from initial presentation for an acute exacerbation and at a second visit within one week of treatment completion. Average treatment duration is not reported but typically treatment with antibiotics/oral corticosteroids would be for 7-14 days.
  • O'Reilly and colleagues evaluated utility using EQ-5D in patients hospitalised for an acute exacerbation of COPD614. The study enrolled 222 patients at a single hospital in Blackpool, UK. Patients had a diagnosis of COPD and were admitted for an acute exacerbationlll. The UK tariff for the EQ-5D was used. Patients were assessed at admission, then every other day during their hospital stay. A group that entered the study following a protocol amendment were also assessed at 3 months after discharge (n = 40). They reported a mean increase in EQ-5D of 0.653 (SD 0.434) between admission and discharge, and a decrease of 0.240 (SD 0.373) between discharge and 3-month follow-up. Average length of stay in hospital was eleven days.

Limited information was identified regarding the duration of impact on utility. As described above, O'Reilly and colleagues reported a reduction in utility between discharge and 3-month follow-up, however this result is difficult to interpret and may reflect new exacerbations that occur during the 3-month follow-up. Spencer and Jones used the SGRQ (a disease specific measure of health-related quality of life) to examine the time course of recovery of health status following an acute exacerbation616. They reported the biggest improvement between presentation and 4 weeks. But SGRQ score continued to improve beyond this. In patients that did not experience another exacerbation SGRQ continued to improve (although at a slower rate) 4 to 12 weeks and even up to 26 weeks. In patients that did experience another exacerbation, SGRQ showed a minor improvement beyond 4 weeks. This suggests that the impact of COPD exacerbation on patients extends beyond the treatment phase.

QALY loss due to an exacerbation was modelled in two parts – the first 2 weeks following an exacerbation and then following this up to 12 weeks (3 months). For non-hospitalised exacerbations, the change in utility from the start of an exacerbation to 2 weeks is based on that reported by Paterson and colleagues (0.17) as this was from exacerbations presenting in general practice615. For hospitalised exacerbations the figure reported by O'Reilly and colleagues is used (0.653) for the corresponding period614. These decrements were incorporated into the probabilistic analysis using a gamma distribution. The utility change over the period 2-12 weeks was estimated based on the rate of change in SGRQ between week 4 and 12 for people not experiencing a new exacerbation reported by Spencer and Jones. SQRG values at week 4 and 12 (42.5 and 37.8 – mean difference 4.7) were mapped to EQ-5D using a published algorithm617. The average change in EQ-5D per week was then calculated. This rate of utility change was then applied for the 2-12 week period resulting in a change in utility of 0.057 over the latter 10 week period of the 12 week period modelled. This parameter was incorporated into the probabilistic analysis using a gamma distribution for the mean SGRQ difference. QALY loss was then calculated for a non-hospitalised and hospitalised exacerbation using the EQ5D decrements and the durations stated. Figure 2 illustrates this graphically. Using this approach the QALY loss is the same irrespective of starting utility and so does not vary with COPD severity. Note that more detail regarding the mapping of SGRQ to EQ5D is given later in this report.

Figure 2. QALY loss during an exacerbation.

Figure 2

QALY loss during an exacerbation.

Previous approaches to modelling the impact of exacerbations on utility

Previous cost-utility analyses in COPD were also reviewed for methods employed for estimating the impact of exacerbations in terms of utility as part of the model development. These are summarised in Table 6.

Table 6. Approaches to exacerbations in cost-utility analyses in the literature.

Table 6

Approaches to exacerbations in cost-utility analyses in the literature.

Utility by COPD severity

In the model, utilities of 0.750 (CI: 0.731-0.768) and 0.647 (CI: 0.598-0.695) are used for severe (FEV1 30 to <50% predicted) and very severe (FEV1 <30%) stages respectively based on data collected pre-randomisation in the UPLIFT study613. These inputs are incorporated into the probabilistic analysis with a beta distribution.

COPD EQ-5D utility data was sought by searching Medline using COPD and EQ-5D specific terms and reviewing previous cost-utility analyses. A review of the use of EQ-5D in COPD was identified and checked for additional papers641. A number of studies were identified that reported EQ-5D estimates of COPD utility – nine reported overall COPD utility and four reported utility by severity stratification. These are summarised in Table 7. Two studies reported COPD utilities stratified into FEV1 30 to <50% predicted and FEV1 <50% predicted. Rutten-van Molken et al. reports EQ-5D data using the UK tariff collected in the multinational UPLIFT trial613. Questionnaires were administered at randomisation and patients therefore weren't on LAMA but could be on other drugs. At baseline 65% were on LABA and 62% were on ICS. Stahl et al. reports EQ-5D data using the UK tariff from a Swedish population 642. Data from the Rutten-van Molken study was selected for use in the model as the population was larger.

Table 7. COPD EQ-5D data.

Table 7

COPD EQ-5D data.

Costs

Drug costs

The annual costs applied for the treatment options in the model were £395.18 for LAMA alone, £488.76 for LABA+ICS and £883.94 for triple therapy.

Treatment costs were estimated based on recommended licensed dosing from summaries of product characteristics, costs from the NHS Drug Tariff and relative usage of different drugs and preparations within each class of therapy (that is: LAMA, LABA+ICS) based on the Prescription Cost Analysis for England 2007221,222,618-623. Table 8 presents a summary of included drug preparations, costs and usage used to calculate costs.

Table 8. Drug unit costs for LAMA and LABA+ICS.

Table 8

Drug unit costs for LAMA and LABA+ICS.

Note the following for costing purposes:

  • LABA+ICS are assumed to be administered only as a combination inhaler product (rather than separate inhalers for each mono-component) as all clinical evidence reviewed used the combination products and the GDG felt it was therefore only appropriate to recommend use of combination products.
  • LAMA and LABA+ICS products are available in a number of different inhalers. As the different inhalers have slightly different prices, an average cost was used in the model based on the relative usage of the different available inhalers from the Prescription Cost Analysis623.
  • Two LABA+ICS combination products are available that are licensed for use in COPDsalmeterol/fluticasone and formoterol/budenoside. The cost of LABA+ICS used in the model was therefore based on a weighted average of the two drug costs.
  • Salmeterol/fluticasone and formoterol/budenoside are also licensed in asthma. A range of different preparations (that is different inhalers/doses) are available, some have a COPD and asthma indication and some only asthma. Inhalers without a COPD indication will generally not be suitable to fulfil the recommended COPD dose. Information was not available in the Prescription Cost Analysis regarding what a prescription was used for and so asthma and COPD usage could not be separated. The average cost of salmeterol/fluticasone and formoterol/budenoside for a patient with COPD was based on the usage of preparations with a COPD indication only.
  • Taking the usage only from preparations of salmeterol/fluticasone and formoterol/budenoside with a COPD indication gave a relative usage between the two products of 74% and 26% respectively. However, GDG members considered this likely to be unrepresentative of true usage, probably due to misprescribing. On this basis a relative usage between the agents was calculated based on overall usage of the drugs which results in 26% salmeterol/fluticasone and 74% formoterol/budenoside. This relative split between the agents was used for costing purposes.
Acute COPD exacerbation costs

Costs of £2403 per hospitalised COPD exacerbation and £34 per non-hospitalised COPD exacerbation were applied in the model. The cost per hospitalised exacerbation was based primarily based on 2007/8 NHS reference costs625. The cost per non-hospitalised exacerbation was based on the results of a UK costing study inflated using UK healthcare inflation indices to 2007/8 prices (latest indices available at time of analysis)624,627. Further details are provided below. Cost parameters were incorporated into the probabilistic analysis using gamma distributions.

Hospitalised exacerbation cost

A cost of £2403 per hospitalised exacerbation of COPD was estimated as follows.

The NHS reference costs provide average UK costs per hospitalisation by HRG code. A weighted average of the costs for all categories of COPD hospitalisation (HRG DZ21A-K) from the 2007/2008 NHS reference costs (latest available at time of analysis) were used to estimate the cost of a hospitalisation for a COPD exacerbation625.

Costs for accident and emergency (A&E) services, paramedic services and critical care are reported unbundled from hospital costs by HRG code in the NHS reference costs and so needed to be added to the above basic hospitalisation cost625. Resource use for these services for a COPD admission was not available from the NHS reference costs and so was sought elsewhere.

It was estimated that 67% of patients would come to hospital by ambulance. This was based on data from the 2008 National COPD audit that reported data regarding admission route for a group of patients hospitalised for COPD an exacerbation626. This reported that 34% of patients saw their GP and were sent to hospital, 12% went to A&E via their own steam and 41% didn't see their GP but called an ambulance (16% had an ‘other’ route and 1% did not state a route). Information was not given about what proportion of patients who saw a GP and were sent to hospital used an ambulance. Based on discussion with a GP representative from the GDG it was judged reasonable to assume that ambulance use would be the same as among those who did not see a GP (that is of the 53% of people who did not see a GP 12% went to A&E via own stream and 41% called an ambulance). The estimate of 67% ambulance use for the model was therefore based on the 41% of patients who didn't see a GP but called an ambulance plus 26% who saw their GP and were sent to hospital by ambulance. The cost of coming to hospital by ambulance was based on a weighted average of the costs for all categories of ‘Paramedic services’ for breathing difficulties (HRG PS06A-C) from the NHS reference costs625.

It was assumed that all patients attended A&E. The cost of A&E was based on the weighted average of the costs for all categories of ‘A&E services leading to admitted’ from the NHS reference costs625.

UK data regarding the use of critical care services per hospitalisation for a COPD exacerbation was not identified. Two studies (one from Italy and one from Spain) were identified from the literature that provided estimates of time spent in ICU per COPD hospitalisation and so an average of these estimates was used; 0.6 days174,628. The cost per day in ICU was based on a weighted average of the costs per day for all categories of ‘Critical care services – Adult: intensive therapy unit’ (HRG XC01Z-XC07Z) from the NHS reference costs625.

The 2008 National COPD Audit indicated that 34% of patients would see their GP prior to coming to hospital and so this cost was also incorporated626. The cost of a GP visit was based on the 2008 average UK cost (latest available at time of analysis)627.

Non-hospitalised exacerbation cost

A cost of £34 per non-hospitalised exacerbation was based on the results of a UK costing study inflated using UK healthcare inflation indices to 2007/8 prices (latest indices available at time of analysis)(before inflated £30.69, SD 111.4)624,627. Details of the selection of the data source are provided below.

The literature was reviewed for estimates of resource use and/or the costs of non-hospitalised COPD exacerbations. Studies that were identified are summarised in Table 9. Original reports of resource or costing studies are included in this table, including those reported within cost-effectiveness study reports. Cost-effectiveness studies that utilise data reported elsewhere are not included in the table (as this would be duplication) nor are those that use estimates based on assumptions or expert opinion. Note that studies that only reported in-hospital costs for patients with COPD exacerbations are also not included in the table.

Table 9. COPD exacerbation costing studies.

Table 9

COPD exacerbation costing studies.

Estimates of cost for a non-hospitalised exacerbation from the studies varied considerably. A number of considerations were relevant in selecting a source for the model. The definition of exacerbations varied between studies and did not necessarily match up with the categorisation being used in this analysis; we were looking for an estimate where hospitalised exacerbations were not included. Most studies were not in a UK setting and management may vary between countries. For example, in the UK access to healthcare is generally via a GP but in other countries this may not be the case.

Only one study was identified that was conducted in a UK setting and the exacerbation definition in this study also matched that being used in the model624. On this basis this source was utilised. It was noted that this cost estimate was quite low compared with the overseas estimates. However, it was difficult to judge if it was inaccurate or if it represented a genuine difference in management between countries. This issue was discussed with the GDG and consideration was given to the cost of drugs used to treat an exacerbation and the average cost of typical healthcare contacts. It was concluded that while it did appear possibly too low it was not unfeasible and, in the absence of other data, should be used in the model. Sensitivity analysis was planned to explore the impact of this cost on results.

COPD maintenance costs

Annual maintenance costs for COPD of £273 (SE 35.0) and £896 (SE 79.5) for severe (FEV1 30 to <50% predicted) and very severe (FEV1 <30% predicted) stages respectively were applied in the model. Mean estimates were derived from a UK COPD costing study33; error estimates were imputed (see below for details). Details of derivation and data selection are provided below.

Note that in the model, maintenance costs only vary between treatment arms in the secondary analyses where a mortality impact of treatment is incorporated.

The literature was reviewed for estimates of per patient annual maintenance costs for stable COPD stratified by severity. Studies that were identified are summarised in Table 10. Original reports of resource or costing studies were included. This included estimates reported within a cost-effectiveness analysis. Cost-effectiveness studies that utilise data reported elsewhere are not included in the table (as this would be duplication) nor are those that use estimates based on assumptions or expert opinion. Only estimates stratified by severity are included. If this did not include stratification of the <50% group they are also not included in the table.

Table 10. COPD maintenance costing studies.

Table 10

COPD maintenance costing studies.

Estimates of annual costs excluding those associated with exacerbations were required for the model as exacerbations are costed separately. This would therefore cover healthcare contact such as regular follow-up visits and additional medications and therapies, such as oxygen. Ideally resource use would have been collected in a UK setting.

Only one study reported costs from a UK setting33. Severity classification was by self-designation or dyspnoea scale (into mild, moderate and severe) rather than FEV1 cut-offs as used in the model. Exacerbations costs were included in the estimates however the study also reported that 60% of costs in the overall population are due to unscheduled care. Some data were available that reported by FEV1 based severity groups and excluded exacerbation costs but from non-UK settings173,174. The UK data was prioritised. The figures for moderate and severe COPD defined by dyspnoea score with 60% of costs subtracted to remove unscheduled care (i.e. treatment of exacerbations) were used for severe and very severe COPD in the model respectively.

These parameters were incoporated into the probabilistic analysis using the cost for severe COPD (£723) and the difference in cost between severe and very severe COPD (£623). Gamma distributions were assigned. No error esimates were reported for the cost estimates and so a standard error was imputed that would generate a confidence interval half that of the mean cost estimate.

Relative treatment effects

As described above, baseline event rates for the LABA+ICS arm of the model were obtained from the literature. The impact of alternative treatment combinations were then modelled by applying relevant relative treatment effects from randomised controlled trials to these baseline event rates.

In the base case analysis only exacerbations are impacted differentially by treatment in the model. Two alternative analyses also incorporate: a) a difference in utility when stable; b) mortality.

Relative treatment effect data were sought from the randomised controlled trials identified in the systematic evidence reviews undertaken for the guideline. Three studies were identified that each compared two of the three treatment options that are incorporated into the model:

All three studies provide direct comparisons of two treatment options in the model. However, the studies form an evidence loop and cannot all be used at the same time to inform the model. For example, if we know the relative number of exacerbations with LAMA compared to LABA+ICS from one study, and the relative number of exacerbations with triple therapy compared to LABA+ICS from another study, the relative number of exacerbations with triple therapy compared in LAMA is therefore implicit without the use of the study that compares triple and LAMA.

There are three possible pairs of trials that can therefore be used in provide the estimates of relative treatment effect for the model (see also Figure 3 below):

Figure 3. Trials data combinations for estimates of relative effect.

Figure 3

Trials data combinations for estimates of relative effect.

  1. INSPIRE and UPLIFT subgroup
  2. INSPIRE and OPTIMAL
  3. UPLIFT subgroup and OPTIMAL

Table 11 below summarises the resulting treatment effect estimates using each of the three pairs of trials. Rate ratios are used for exacerbations, and exacerbations requiring hospitalisation. Risk ratios are used for mortality. Mean difference is used for EQ-5D – this is obtained by mapping mean SQRQ data to EQ5D and calculating the difference. Note that more detail regarding the mapping of SGRQ to EQ5D is given later in this report.

Table 11. Relative effect estimates used in model for each three pairs of trials.

Table 11

Relative effect estimates used in model for each three pairs of trials.

The model was run using each of the three pairs of trials so that the impact on results and conclusions could be examined. As LABA+ICS data had been used to populate the model, relative treatment effects were calculated and applied in the model for LAMA and triple therapy compared to LABA+ICS using the above data. In the probabilistic analysis log normal distributions were used for rate ratios and risk ratios. Normal distributions were used for the mean SGRQ differences that were used calculate the mean EQ5D differences.

Mapping SGRQ to EQ-5D

Due to a lack of utility data, SGRQ data were mapped to EQ-5D where required. This was done as part of the estimation of QALY loss with an exacerbation (direct utility data was available for the initial impact but not over the longer term) and also to estimate the impact of treatment on stable utility as described in the relevant sections above.

The SGRQ (St Georges Respiratory Questionnaire) is a widely used measure of health impairment in COPD and asthma. SGRQ is not a utility measure and so cannot be used directly to calculate QALYs. There have however been some reports of mapping of SGRQ to EQ-5D. Two algorithms were identified that mapped total SGRQ score to EQ-5D utility175,617. These were compared and the Starkie method was selected in preference to the Oba method as the latter resulted in impossible values at the extreme ends175,617. However, it is noted that both approaches yielded similar values in the middle. The Starkie formula is displayed below.

Predicted utility score=10.0335+0.0017T+0.0001T20.0279G

Where: T = total SGRQ score; G = gender (0=female, 1=male)

The GDG highlighted that they were aware of some issues with mapping SGRQ to EQ-5D when examined at a patient level and it was judged inferior to direct utility data. However, in the absence of alternatives this was considered a reasonable approach to fill in gaps in the data.

In addition, the SGRQ reflects exacerbations as well as stable symptoms. This is likely to more of an issue when used as an approximation of the difference in stable utility between treatment options than when estimating the rate of recovery following an exacerbation. In particular because the data used for the rate of recovery is in patients who do not have a new exacerbation and is also non-comparative.

Computations

The model was constructed in Microsoft Excel and was evaluated by cohort simulation.

Patients start in cycle 0 distributed amongst the model health states (severe, very severe, dead) as described above. Patients were redistributed amongst the model health states over time as follows. Each cycle, the age-dependant COPD-severity specific death rates were applied to alive patients and the probability of progressing from severe to very severe was then applied to the remaining alive patients in the severe severity group in order to recalculate the number of people in each state. Life years in severe and very severe COPD states for the cohort are computed each cycle. A half-cycle correction is applied.

Each cycle, the number of exacerbations the cohort experienced was calculated by applying the severity-specific exacerbation rates to the number of life years in each severity state. The number of hospitalised exacerbations experienced was calculated by applying the severity-specific hospitalisation rates to the number of life years in each severity state. The number of non-hospitalised exacerbations was calculated by subtracting the number of hospitalised exacerbations from the total exacerbations.

Total QALYs were calculated from the above information as follows. Each cycle, the time spent (i.e. 1 year) in each state of the model was weighted by the utility for that state. This gives the QALYs for each state for the cycle. The number of non-hospitalised and hospitalised exacerbations that occurred was multiplied by the relevant QALY loss due to an exacerbation. These were combined to give the QALYs per cycle, Q(t), and discounted to reflect time preference (discount rate = r). QALYs during year 1 were not discounted. The total discounted QALYs was the sum of the discounted QALYs per cycle.

Total discounted QALYs=t=1iQ(t)(1+r)t1

Where: t = cycle number; i = maximum cycle number; Q(t) = QALYs in cycle t; r = discount rate

Total costs were calculated from the above information as follows. Each cycle, the time spent (i.e. 1 year) in each state of the model was multiplied by the maintenance costs for that state and the relevant drug cost. The number of non-hospitalised and hospitalised exacerbations that occurred was multiplied by the respective costs. These were combined to give the costs per cycle, C(t), and discounted to reflect time preference (discount rate = r). Costs during year 1 were not discounted. The total discounted costs was the sum of the discounted costs per cycle.

Total discounted costs=t=1iC(t)(1+r)t1

Where: t = cycle number; i = maximum cycle number; C(t) = Costs in cycle t; r = discount rate

The widely used cost-effectiveness metric is the incremental cost-effectiveness ratio (ICER). This is calculated by dividing the difference in costs associated with two alternatives by the difference in QALYs. The decision rule then applied is that if the ICER falls below a given cost per QALY threshold the result is considered to be cost effective. If both costs are lower and QALYs are higher the option is said to dominate and an ICER is not calculated.

ICER=Costs(B)Costs(A)QALYs(B)QALYs(A)

Where: Costs/QALYs(X) = total discounted costs/QALYs for option X

  • Cost-effective if: ICER < Threshold

When there are more than two comparators, as in this analysis, options must be ranked in order of increasing cost then options ruled out by dominance or extended dominance before calculating ICERs excluding these options.

It is also possible, for a particular cost-effectiveness threshold, to re-express cost-effectiveness results in term of net benefit (NB). This is calculated by multiplying the total QALYs for a comparator by the threshold cost per QALY value (for example, £20,000) and then subtracting the total costs. The decision rule then applied is that the comparator with the highest NB is the most cost-effective option at the specified threshold. That is the option that provides the highest number of QALYs at an acceptable cost. For ease of computation NB is used to identify the optimal strategy in the probabilistic analysis simulations.

Net Benefit(X)=[QALYs(X)×D]Costs(X)

Where: Costs/QALYs(X) = total discounted costs/QALYs for option X; D = threshold

The probabilistic analysis was run for 5000 simulations. Each simulation, mean discounted costs and mean discounted QALYs were calculated for each treatment option. The net benefit was also calculated and the most cost-effective option identified (that is, the one with the highest net benefit), at a threshold of £20,000 and £30,000 per QALY gained. The results of the probabilistic analysis are summarised in terms of mean costs, mean QALYs and mean net benefit for each treatment option, where each is the average of the 5000 simulated estimates. The option with the highest mean net benefit (averaged across the 5000 simulations) is the most cost-effective at the specified threshold. The percentage of simulations where each strategy was the most cost-effective gives an indication of the strength of evidence in favour of that strategy being cost-effective.

Results are also presented on the cost-effectiveness plane where the difference in mean costs and the difference in mean QALYs between treatment options are plotted. All differences are calculated relative to LABA+ICS and so LABA+ICS is always at the origin of the cost-effectiveness plane. Results could have equally been presented with differences calculated relative to LAMA or triple therapy. This would make no difference to the cost effectiveness results it would simply mean that the axis would move so that a different treatment option is at zero. Comparisons not ruled out by dominance or extended dominance are joined by a line on the graph where the slope represents the incremental cost-effectiveness ratio, the magnitude of which is labelled.

Results

Detailed results are presented over the next few pages for the basecase scenario and various sensitivity analyses including the alternative treatment effect analyses. All results are means from the probabilistic analysis unless otherwise specified.

Basecase analysis – exacerbation effect only

In the basecase analysis only exacerbations (non-hospitalised and hospitalised) varied between treatment options. A four-year treatment period was considered. Three analyses were undertaken using different pairs of clinical trials to calculate relative treatment effects.

The results of these analyses are presented in Table 12 and Figure 4. A break down of costs is presented in Table 13. LAMA or LABA+ICS was found to be the most cost-effective strategy depending on the clinical trial data used to calculate relative treatment effects.

Table 12. Basecase results (exacerbation effect only; 4 years).

Table 12

Basecase results (exacerbation effect only; 4 years).

Figure 4. Basecase results on the cost-effectiveness plane (exacerbation effect only; 4 years).

Figure 4

Basecase results on the cost-effectiveness plane (exacerbation effect only; 4 years).

Table 13. Basecase cost breakdown (exacerbations effect only; 4 years) – totals for a cohort of 1000 people (deterministic analysis).

Table 13

Basecase cost breakdown (exacerbations effect only; 4 years) – totals for a cohort of 1000 people (deterministic analysis).

When INSPIRE and UPLIFT subgroup data were used, LAMA was found to be the most cost-effective option. Triple therapy was the most effective (that is it had the highest number of QALYs) but had a high ICER when compared with LAMA at £187,697 per QALY gained. LABA+ICS was more effective than LAMA (higher QALYs) but also with higher costs and was ruled out by extended dominance. LAMA was the optimal strategy at a threshold of £20,000 per QALY gained in 84% of simulations, LABA+ICS in 16% and triple therapy in 0%. When INSPIRE and OPTIMAL data were used instead results were similar although the ICER for triple therapy compared to LABA+ICS was lower at £93,737 per QALY gained.

When UPLIFT subgroup and OPTIMAL data were used LABA+ICS was found to be the most cost-effective option. LAMA was ruled out by dominance – it was more expensive with lower QALYs than LABA+ICS. Triple therapy was the most effective (that is, it had the highest number of QALYs) but had a high ICER when compared with LABA+ICS at £159,353 per QALY gained. LABA+ICS was the optimal strategy at a threshold of £20,000/QALY in 92% of simulations, LAMA in 8% and triple therapy in 0%.

The results indicate fairly low uncertainty within individual analyses. However, the fact that between analyses there is a disagreement about the most cost-effective option indicates considerable uncertainty based on the available clinical evidence.

Sensitivity analyses

Alternative analysis one – exacerbation and stable quality of life effects

In this alternative analysis stable utility is differentially impacted between comparators as well as exacerbations. As in the basecase a four-year treatment period was considered and three analyses were undertaken using different pairs of clinical trials to calculate relative treatment effects.

Results of these analyses are presented in Table 14 and Figure 5. Triple therapy was found to be the most effective (highest number of QALYs) and most cost-effective strategy irrespective of the clinical trial data used to calculate relative treatment effects. LABA+ICS was found to be the next most effective and cost-effective option also irrespective of clinical data used. LAMA was less effective but also less expensive than LABA+ICS, except for when the data pair of UPLIFT and OPTIMAL was used and it was dominated. The ICER for triple therapy compared to LABA+ICS was in the range £7000 to £15,000 depending on the clinical trial data pair used. At a threshold of £20,000 per QALY gained, triple therapy was optimal in 71% to 76% of simulations, LABA+ICS was optimal in the majority of the remaining simulations and LAMA was very rarely optimal.

Table 14. Alternative analysis 1 results (exacerbation and stable quality of life effects; 4 years).

Table 14

Alternative analysis 1 results (exacerbation and stable quality of life effects; 4 years).

Figure 5. Alternative analysis 1 results on the cost-effectiveness plane (exacerbation and stable quality of life effects; 4 years).

Figure 5

Alternative analysis 1 results on the cost-effectiveness plane (exacerbation and stable quality of life effects; 4 years).

In this sensitivity analysis there was fairly low uncertainty within and between analyses that triple therapy is the optimal strategy. That is it provided the greatest health gain at an acceptable cost.

Alternative analysis two – exacerbations and mortality effects

In this second alternative analysis mortality is differentially impacted between comparators as well as exacerbations. As in the basecase a four-year treatment period was considered and three analyses were undertaken using different pairs of clinical trials to calculate relative treatment effects.

Results of these analyses are presented in Table 15 and Figure 6.

Table 15. Alternative analysis 2 results (exacerbation and mortality effects; 4 years).

Table 15

Alternative analysis 2 results (exacerbation and mortality effects; 4 years).

Figure 6. Alternative analysis 2 results on the cost-effectiveness plane (exacerbation and mortality effects; 4 years).

Figure 6

Alternative analysis 2 results on the cost-effectiveness plane (exacerbation and mortality effects; 4 years).

When INSPIRE and UPLIFT subgroup data were used LABA+ICS was the the most cost-effective option. LAMA was less effective but also with lower costs. The ICER for LABA+ICS versus LAMA was low at £4302. Triple therapy was the most effective (that is it had the highest number of QALYs) but had an ICER of £40,722 when compared to the next most effective strategy, LABA+ICS, and so was not considered cost-effective. LABA+ICS was the optimal strategy at a threshold of £20,000 per QALY gained in 89% of simulations, LAMA in 4% and triple therapy in 7%.

When INSPIRE and OPTIMAL data were used instead results were quite different. LABA+ICS was still the most cost-effective option but was now also the most effective option (highest QALYs). LAMA was again less effective and with lower costs than LABA+ICS, and the ICER for LABA+ICS vs LAMA was low. Triple therapy was however now dominated by LAMA as it was less effective (lower QALYs) with higher costs. LABA+ICS was the optimal strategy at a threshold of £20,000 per QALY gained in 92% of simulations, LAMA in 3% and triple therapy in 5%.

When UPLIFT subgroup and OPTIMAL data were used results were again different. LAMA was now the most effective (that is it had the highest number of QALYs) and cost-effective option. LABA+ICS was less effective and less costly than LAMA and triple therapy was ruled out by extended dominance. The ICER for LAMA versus LABA+ICS was £15,566. LAMA was the optimal strategy in 64% of simulations, LABA+ICS in 34% and triple therapy in 2%.

Results indicate fairly low uncertainty within individual analyses. However, there are considerable differences between results based on difference clinical data indicating high uncertainty in this sensitivity analysis.

Time horizon

Sensitivity analysis explored the impact of the time horizon on results. The time horizon did not greatly impact results for the base case analysis or the first alternative analysis described above and conclusions remained the same. There was a small decrease in the magnitude of the ICERs as the time horizon increased.

The time horizon had a greater impact in the second alternative analysis where a treatment effect on mortality was incorporated. Results for this analysis for a 1 year, 4 year and lifetime analysis are summarised in Table 16.

Table 16. Time horizon sensitivity analysis: alternative analysis 2 results (exacerbation and mortality effects).

Table 16

Time horizon sensitivity analysis: alternative analysis 2 results (exacerbation and mortality effects).

In the 4-year analysis of option 1, LABA+ICS was the most cost-effective option; triple therapy had the highest QALY but was not cost-effective. However when this 4-year treatment period was extrapolated to a lifetime impact triple became a cost-effective option.

In the 4-year analysis of option 3, LAMA was the most effective option (hightest QALYs) and the most cost-effective option. When the time horizon was reduced to 1 year LAMA was still the most effective but was no longer the most cost-effective and LABA+ICS was.

Exacerbation rate

A sensitivity analysis was undertaken to look at the impact of varying the baseline exacerbation rate on the basecase analysis. Rates were varied by a factor of -50% to +300% – the resulting baseline exacerbation rates used in the sensitivity analysis are presented in Table 17. Results are presented in Figure 7. We found that as the exacerbation rate increases so the percentage of simultations where triple therapy was optimal increased.

Table 17. Exacerbation rates used in sensitivity analysis.

Table 17

Exacerbation rates used in sensitivity analysis.

Figure 7. Exacerbation rate sensitivity analysis: basecase analysis (exacerbation effect only; 4 years).

Figure 7

Exacerbation rate sensitivity analysis: basecase analysis (exacerbation effect only; 4 years).

Cost of non-hospitalised exacerbations

Sensitivity analysis around the cost of a non-hospitalised exacerbation was undertaken due to uncertainty about the cost being too low. In one analysis the cost was doubled from £34 to £68. This had very little impact on the basecase analysis results.

A threshold analysis was also undertaken (using the deterministic analysis) to see at what cost of a non-hospitalised exacerbation would triple therapy become the favoured option (i.e. with an ICER of under £20,000/QALY) in the basecase analysis. The result was that triple therapy was cost-effective only when the cost of treating a non-hospitalised exacerbation was assumed to be around £2000 or higher. The exact theshold varied depending on the clinical trial data pair used.

Discussion

Summary and GDG interpretation

The aim of this analysis was to evaluate which was the most cost-effective option from LABA+ICS, LAMA and triple therapy for initial management of people with COPD and an FEV1 <50%.

The base case analysis, which is driven by differences in exacerbations between treatments, found that LABA+ICS or LAMA was the most cost-effective option depending on which clinical data was used to inform the differences between treatments. Triple therapy was the most effective option (highest QALYs) but was not cost-effective. The GDG considered this analysis to be the most robust in terms of the available data. However, it was also considered likely to be conservative in terms of the benefits of treatment and may underestimate the value of triple therapy. The fact that either LABA+ICS or LAMA was the favoured option depending on the clinical data used in the analysis highlights an inconsistency in the clinical data but one that could not be resolved and so therefore was considered to represent an uncertainty over the preferred option.

In the sensitivity analysis which also incorporates a difference between treatments in terms of stable utility (quality of life), triple therapy was found to be the most effective (highest QALYs) and the most cost-effective option, irrespective of which clinical data was used to inform the differences between treatments. The GDG considered that a scenario where treatment impacted utility due to stable symptom improvement as well as exacerbations to be a realistic one but given the limitations of the estimate of treatment effect on stable utility they interpreted the results with caution.

A sensitivity analysis that looked at the impact of exacerbation rates found that as the baseline exacerbation rate increased so did the probability that triple therapy was cost-effective.

In the sensitivity analysis where a treatment effect in terms of mortality was incorporated, results varied greatly depending on the clinical data used and were sensitive to the time horizon taken. This reflected considerable inconsistency in the clinical data for this outcome.

The GDG concluded that this result was difficult to interpret and it was not used to inform decision making.

Limitations

The availability of utility data to inform the estimation of QALYs was somewhat limited. EQ-5D utility data was identified for the initial impact of hospitalised and non-hospitalised exacerbations. Mapping of SGRQ data to EQ-5D utility was used to supplement this where necessary. GDG members indicated that they were aware of problems with mapping SGRQ to EQ-5D and were generally not in favour of an approach that primarily based QALY impact on this. For this reason, in the base case analysis we attributed a QALY loss to hospitalised and non-hospitalised exacerbations, which minimised the reliance on mapped data. This lack of direct utility data impacts most analyses in the area of COPD. A notable exception being a cost-utility analysis using patient level TORCH data where EQ5D utility data was collected at various time points throughout the trial and so could be used as a basis for QALY calculations.

In the model we assumed that an exacerbation impacted a patient (to a diminishing extent) for 3 months but then stable utility will return to the same level as prior to the exacerbation. The GDG noted that there is evidence that exacerbations may permanently impact quality of life and this assumption is likely to be somewhat conservative. It was however accepted as a reasonable simplification for modelling purposes.

As described in the model input section, there was discussion regarding whether the cost of a non-hospitalised exacerbation identified in the literature was too low. Sensitivity analysis showed however that the model was not especially sensitive to the cost of a non-hospitalised exacerbation and this uncertainty was therefore not considered a major limitation.

Note that other more minor data limitations were discussed throughout the model inputs section.

Conclusions

Based on the limitations of the clinical evidence for triple therapy and the results of the cost-effective model, the GDG concluded that patients with an FEV1 <50% should be offered LAMA or LABA+ICS as initial maintenance therapy. The GDG considered that while triple therapy was potentially effective and cost-effective, the evidence was not strong enough to warrant a recommendation that all patients with an FEV1 <50% be routinely started on triple therapy. Triple therapy was instead recommended if symptoms or exacerbations persisted. They noted that triple therapy was most likely to be cost-effective in patients who will obtain a benefit in terms of exacerbation reduction and symptom relief.

Footnotes

kkk

An increase in at least two of the following: increased frequency and/or severity of cough; increase in sputum volume, dyspnoea or increased dyspnoea; increase in chest congestion as indicated by adventitious sounds, and chills and/or fever. Patients also had to be able to produce mucopurulent or purulent sputum and had to be able to provide a suitable sample for laboratory analysis and microbiological confirmation.

lll

No specific definition of an exacerbation was used; it was based on the physician and respiratory nurse's determination.

mmm

minor = requiring oral corticosteroids and/or antibacterials; major = hospitalisation

nnn

mild = patient manages in normal environment including telephone call to doctor and possibly antibiotics or oral steroids; moderate = patient must make an unscheduled visit to DR; severe = requires hospitalisation or ER visit

ooo

GOLD 1 = FEV1 >80% predicted; GOLD 2a = FEV1 50-80% predicted; GOLD 2b = FEV1 30 to <50% predicted; GOLD 3 = FEV1 <30% predicted

ppp

Non-severe = awareness of sign or symptom AND discomfort that interferes with usual activities; severe = inability to do work or usual activities

qqq

Mild = worsening of symptoms requiring outpatient physician services and institution of systemic corticosteroids or antimicrobial agents; moderate = requiring emergency department utilisation or urgent physician office visits; severe = requiring inpatient care

rrr

All costs are converts to UK £ using PPP for the appropriate year, and then inflated to 2007/8 costs using the PSSRU healthcare inflation indices177,627. Reported to nearest whole £.

sss

Direct cost only presented here; calculated by dividing exacerbation costs/year by exacerbations/year

ttt

Direct medical costs only included here; NHS sick leave benefit and other excluded.

uuu

52% of severe exacerbations required hospital admission.

vvv

Classification of exacerbations based on ratings by the physician-investigator.

www

Hospitalisation was 16% and 78% in moderate and severe exacerbations respectively.

xxx

This CEA also reports Netherlands estimates using different data but as this is based on Oostenbrink 2004 detailed above this is not included here.

yyy

All costs are converts to UK £ using PPP for the appropriate year, and then inflated to 2007/8 costs using the PSSRU healthcare inflation indices627. Reported to nearest whole £. FEV1 % predicted: GOLD 1/2/3/4 = ≥80/79-50/49-30/<30; ATS 1/2/4 = 80-50/50-35/<35.

zzz

Direct cost only presented here; calculated by dividing exacerbation costs/year by exacerbations/year.

aaaa

Other country reports of same study available but not reported as same format as for UK.

bbbb

Direct costs only presented here

Copyright © 2010, National Clinical Guideline Centre - Acute and Chronic Conditions.

Apart from any fair dealing for the purposes of research or private study, criticism or review, as permitted under the Copyright, Designs and Patents Act, 1988, no part of this publication may be reproduced, stored or transmitted in any form or by any means, without the prior written permission of the publisher or, in the case of reprographic reproduction, in accordance with the terms of licences issued by the Copyright Licensing Agency in the UK. Enquiries concerning reproduction outside the terms stated here should be sent to the publisher at the UK address printed on this page. The use of registered names, trademarks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant laws and regulations and therefore for general use. The rights of National Clinical Guideline Centre to be identified as Author of this work have been asserted by them in accordance with the Copyright, Designs and Patents Act, 1988.

Bookshelf ID: NBK65024

Views

  • PubReader
  • Print View
  • Cite this Page
  • PDF version of this title (6.1M)
  • Disable Glossary Links

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...