Introduction

COPD management is aimed at reducing the symptoms of the disease, preventing exacerbations and slowing disease progression. It consists of a number of components that may include a self-management strategy, vaccinations, smoking cessation treatment and support, pulmonary rehabilitation, oxygen therapy and non-invasive ventilation, and the use of inhaled medicines. Inhaled drugs can be grouped into short-acting bronchodilators, that aim to provide rapid relief of acute symptoms, long-acting bronchodilators that are taken by people with moderate to very severe COPD as a maintenance therapy, and inhaled corticosteroids (ICS).

The long-acting bronchodilators can be taken as single or fixed-dose combined inhalers. The possible combinations of drugs include: long-acting muscarinic antagonist (LAMA); long-acting beta-adrenoceptor agonist (LABA); LABA/inhaled corticosteroid (LABA/ICS) and LAMA/LABA. Treatment with ICS aims to reduce inflammation and ICS may act synergistically when combined with a LABA. LAMA and LABA combinations may also lead to synergistic effects.

This review aims to determine the comparative effectiveness of different drug classes for managing stable COPD. The evidence presented in this review was provided by the Cochrane Airways Group as part of a collaboration between the NICE Guideline Updates Team and the Cochrane group. We thank the Cochrane Airways Group for their assistance in providing the literature searches and data for this review question. The full details and results are provided in the published Cochrane review (Oba 2018). The protocol used by the Cochrane Group is summarised in Table 1 and detailed in appendix A, with any additions noted in the methods section below. The review does not consider the comparative effectiveness of different drugs within a given class, or the comparative effectiveness of different inhaler devices.

Table 1

PICO for the comparative effectiveness of combinations of inhaled therapies.

Methods and process

This review was carried out as a collaboration with the Cochrane Airways Group. The published review protocol (Oba et al 2017) contains details of the methodology the Cochrane group planned to use to carry out their review and network meta-analysis (NMA).

The evidence presented here is the work of the Cochrane group, with the exception of any alterations made to reflect the methodology used by the NICE Guideline Updates Team, that are stated in the relevant sections. Any errors introduced by these changes are the responsibility of the NICE Guideline Updates Team alone. The sections of the review carried out by the NICE Guideline Updates Team were developed using the methods and process described in Developing NICE guidelines: the manual. Methods specific to this review question are described in the review protocol in appendix A, and the methods section in appendix B. The search strategies used in this review are detailed in appendix C.

In particular, the following definitions, key outcomes and methods have been adopted:

1. The Cochrane review divided exacerbations into moderate to severe and severe categories. A moderate exacerbation is defined as worsening of respiratory status that requires treatment with systemic corticosteroids and/or antibiotics; a severe exacerbation is defined as a rapid deterioration that requires hospitalisation.
2. Data for the St George’s Respiratory Questionnaire (SGRQ) were presented in 2 ways, depending on the format of data in the included studies: as changes in SGRQ total score and as the number of responders (decrease in SGRQ score of ≥4 units).
3. End of study data was reported for dichotomous outcomes, while continuous outcomes were reported for the end of the study and at 3, 6 and 12 months where possible. Data that did not fit into these categories was assigned to the closest category.
4. The Cochrane group reported change in trough FEV1 in litres (L). This was not converted to millilitres (ml) as used in the other reviews carried out by the NICE Guideline Updates Team for the COPD guideline update to prevent the introduction of rounding errors in the data.
5. Resource use and costs were not included in the Cochrane review, but were addressed by the economic searches carried out by the NICE reviewers.
6. This review only includes drugs and doses licensed in the USA and EU.
7. The following inhaled bronchodilators were included in the review:

   • LAMA monotherapy (aclidinium, glycopyrronium, tiotropium and umeclidinium).
   • LABA monotherapy (formoterol, olodaterol, salmeterol, vilanterol).
   • LABA/ICS (formoterol/beclomethasone, formoterol/budesonide, formoterol/ciclesonide, formoterol/fluticasone, formoterol/mometasone, indacaterol/mometasone, salmeterol/fluticasone, vilanterol/fluticasone).
   • LABA/LAMA (formoterol/aclidinium, indacaterol/glycopyrronium, indacaterol/tiotropium, olodaterol/tiotropium, vilanterol/umeclidinium).
8. The Cochrane group NMA models allowed analysis of the drugs at the class level and at the individual drug level within and between classes. However, this review was limited to comparisons between drug classes. Please refer to the Cochrane review for additional information.
9. For data analysis, the Cochrane group divided the studies into low and high risk groups, based on the previous exacerbation history of the participants. Studies that specifically recruited people with a history of hospital admission due to COPD exacerbation within 12 months of study entry (or contained subgroup data on these people) were classed as high risk and those that didn’t mention this as an entry criteria or actively recruited people without an exacerbation requiring hospitalisation in this time frame were classed as low risk. Data was presented for both low and high risk groups in the forest plots. Only the pooled effects from combining both groups was presented in the GRADE tables for the pair-wise comparisons because the use of these subgroups was not prespecified by the committee.
10. PINNACLE 3 (Hanania 2017) is an extension of the PINNACLE 1 and 2 (Martinez 2017 a and b) trials. Data were extracted for PINNACLE 3 in preference to PINNACLE 1 and 2 where possible. If data were included for all 3 studies, the PINNACLE 3 data were for the period of the extension trial only to prevent double counting.
11. The minimally important differences (MIDs) used in this review are summarised in Table 15 in appendix B. These were selected based on the literature with input from the committee.
12. Evidence tables, individual domain risk of bias judgements and reasons for study exclusion were extracted directly from the Cochrane review. However, overall study risk of bias and applicability assessments were carried out by the NICE Guideline Updates Team based on the information provided in the Cochrane review.
13. Publication bias was assessed using the funnel plots shown in appendix F, but in the absence of a clear risk of bias, was not incorporated into the GRADE tables.
14. The planned subgroup analyses were not carried out for this review because the included studies did not report data for the categories of interest in an accessible format.
15. The NMA models and data were provided by the Cochrane review authors. The models included fixed and random effect models with/without fixed or random class effects. These models were run according to the Cochrane group methods and choice of burn in, with priors specified by them. However, the NICE Guideline Updates Team used a larger burn in of 100,000 iterations to allow convergence of chains for the Cardiac SAEs low and high risk models.
16. Cochrane group did not write and test all possible models for each outcome. They started with the simplest model (fixed effect and fixed class) and then moved to more complex models as needed to achieve a good model fit to the data. If a simpler model was a good fit, then more complex models were not always tested. The Guideline updates team chose which of these models to use based on the rules in appendix B.
17. In cases where the data contained a large number of zero events, the Cochrane group used a continuity correction. This involved adding 0.5 to the zero event arm and its matching comparator arm.
18. Data were extracted for the mean effect and 95% credible intervals from the NMA model with the best fit to the data based on the NICE Guideline Updates team criteria for model choice detailed in appendix B. Pooled results were reported as mean differences (MD) or Relative Risks (RR).
19. The Cochrane group presented dichotomous outcomes, apart from exacerbations, as odds ratios (OR). These were converted to RR by the NICE Guideline Updates Team using the event rate in the reference or control arm for each outcome from sources used in the health economics model or, if this was not available, based on LABA arm data for the largest trial for a particular outcome.
20. The Cochrane group used hazard ratio (HR) models to look at exacerbations in their NMAs. The HR data obtained from these models cannot be compared to the pair wise RR data and, as a result, the pairwise data section of the tables for exacerbations are left blank (Table 27, Table 28, Table 29, Table 30).
21. Although there were studies at high risk of bias included in the NMA, a sensitivity analysis excluding these studies was not carried out because the sensitivity analysis carried out on the pair wise data did not alter the interpretation of the effects of the treatments.

Declarations of interest were recorded according to NICE’s 2014 conflicts of interest policy.

Clinical evidence

Summary of clinical studies included in the evidence review

The evidence tables for the included studies are presented in appendix E and the studies referenced in full in appendix M.

Quality assessment of clinical studies included in the evidence review

The included studies were assessed for risk of individual biases and applicability by the Cochrane group. Overall study level risk of bias and applicability was judged by the Guideline Updates Team and both sets of information are presented in appendix E.

Please refer to appendix F for forest plots, appendix G for the NMA data and appendix H for full GRADE tables.

Economic evidence

Summary of studies included in the economic evidence review

Gani et al. (2010) conducted a cost–utility analysis with a 1-year time horizon comparing tiotropium (LAMA) with salmeterol (LABA) and with ipratropium (SAMA) in UK COPD patients with FEV1 of < 80% predicted. This study was funded by 2 manufacturers of tiotropium. The evaluation used a Markov structure based on GOLD stages 2, 3 and 4 (50%–80% FEV1 predicted, 30%–49% FEV1 predicted, and < 30% FEV1 predicted, respectively). In each cycle of the model patients could remain the same GOLD state or progress to a different GOLD state. In each cycle patients were also at risk of either a severe or non-severe exacerbation.

Treatment effects were implemented as a relative risk of exacerbations and treatment-specific probabilities of moving between GOLD stages in each cycle (determined by patients’ change in FEV1 over time). These data were taken from RCTs comparing tiotropium 18 micrograms once-daily with either salmeterol 50 micrograms twice-daily (described in Brusasco 2003), ipratropium 40 micrograms four-times daily (not included in the clinical review), or placebo (described in Casaburi 2002).

The model included 3 categories of cost: (1) maintenance costs, which were estimated based on disease severity by a Delphi Panel of GPs and secondary care consultants; (2) exacerbation costs, which were calculated by estimating the proportion of patients managed in primary or secondary care for each type of exacerbation and weighting the appropriate NHS reference costs by these proportions; and (3) drug costs, which were calculated based on the list prices and recommended dosage of each treatment.

Baseline utility scores stratified by GOLD stage were taken from a study which measured EQ-5D scores of a sample of 1,235 COPD patients, with a utility reduction of 50% or 15% applied over the course of a month for severe or non-severe exacerbations, respectively.

Base-case results showed that, compared with salmeterol, tiotropium is associated with a cost saving of £126 and generates an additional 0.014 QALYs, and therefore dominates salmeterol. Probabilistic sensitivity analysis indicated that tiotropium was the cost-effective option in 97% of iterations. A subgroup analysis showed that tiotropium continues to dominate salmeterol when patients are stratified by baseline GOLD stage.

This study was classified as being partially applicable as it only considered 2 of the interventions of interest. It was categorised as having potentially serious limitations as it uses a short time horizon, does not include treatment-related adverse events, estimates costs via a Delphi Panel rather than using empirical data, and is subject to a potential conflict of interest.

Hertel et al. (2012) conducted a cost–utility analysis with a lifetime horizon of various combinations of LAMA, LABA, ICS and roflumilast in UK COPD patients with severe and very severe COPD, with ICS-tolerant and ICS-intolerant patients analysed as 2 separate cohorts. This study was funded by a manufacturer of roflumilast. The evaluation used a Markov structure based on GOLD stages 3 and 4 (30%–50% predicted FEV1 and < 30% predicted FEV1 respectively). In each cycle of the model, patients could remain in the same GOLD state, progress to a more severe GOLD state or die. In each cycle patients were also at risk of exacerbation, which could be community- or hospital-treated. The model also allowed treatment switching to a second line regimen: LAMA + LABA/ICS for ICS-tolerant patients and LAMA + LABA for ICS intolerant patients.

Patients’ probability of progressing to a more severe GOLD stage was modelled based on the mean rate of FEV1 decline in COPD patients. Mortality was incorporated by applying the standardised mortality ratio for COPD to the background mortality rate for the UK population, and also by including a probability of death associated with hospital-treated exacerbations. Treatment effects were incorporated as relative differences in exacerbation rates derived from a network meta-analysis.

The analysis included three categories of cost: (1) maintenance costs, which were estimated using resource use data from a tiotropium and unit cost data from NHS reference costs; (2) exacerbation costs, which were estimated using resource usage data from the GOLD strategy group, and unit costs from NHS reference costs; and (3) drug costs, which were sourced from the BNF. Baseline utility scores according to GOLD stage were obtained from clinical trials of roflumilast, and utility decrements associated with exacerbations were obtained from a previous study evaluating holistic preferences of a variety of COPD health states.

Relevant base-case results of the evaluation are shown in Table 2 and Table 3, which excludes interventions not relevant to the review question (ICERs have been manually calculated as were not reported by the authors). These results show that LAMA+LABA produces the greatest number of QALYs and is associated with an ICER of less than £20,000 per QALY, and is therefore the most cost-effective option at this threshold.. The authors’ sensitivity analyses addressed a comparison which is not relevant to the review question.

Table 2

Incremental results for treatments of interest in Hertel et al. (2012) in ICS-tolerant patients.

Table 3

Incremental results for treatments of interest in Hertel et al. (2012) for ICS-intolerant patients.

This analysis was categorised as being partially applicable as it is conducted in a population of patients with severe or very severe COPD. It was classified as having potentially serious limitations as it relies on assumed exacerbation rates with no empirical basis, does not conduct a probabilistic sensitivity analysis for the comparisons of interest, does not include treatment-related adverse events, and is subject to a potential conflict of interest.

Price et al. (2013) conducted a cost–utility analysis with a 3-year time horizon comparing indacaterol (LABA) with tiotropium (LAMA), and indacaterol (LABA) with salmeterol (LABA) in patients with COPD in the UK. This study was funded by a manufacturer of indacaterol. The evaluation used a Markov structure with states based on GOLD stages 1, 2, 3 and 4 (FEV1 ≥ 80% predicted, 50%–80% predicted, 30%–50% predicted, and <30% predicted, respectively). In each cycle of the model, patients could remain in the same GOLD stage, change GOLD stage, or die. Patients could also experience a mild or severe exacerbation in each cycle.

Effects of treatment on FEV1 and exacerbation rates were incorporated using data from the INLIGHT-2 and INHANCE trials (reported in Donohue 2010 and Kornmann 2011). Improvement in patients’ FEV1 was implemented via empirical transition probabilities in the first 12-week cycle of the model. After this initial period the assumption was made that all patients experienced a uniform decline in FEV1 regardless of treatment received. Differences in exacerbation rates were implemented by applying rate ratios for each treatment versus placebo to the number of exacerbations experienced in the placebo arms of the trials.

Resource use data were obtained from the Optimum Patient Care Research Database and were validated with ‘a UK clinician with expertise in COPD management’. Unit costs were taken from standard NHS sources. Baseline utility scores for each GOLD state were taken from indacaterol clinical trials, and utility decrements associated with exacerbations were obtained from a previous study evaluating holistic preferences of a variety of COPD health states.

Results were presented as pairwise comparisons, rather than as a fully incremental analysis. Base-case results indicate that, compared with tiotropium 18 micrograms daily, indacaterol 150 micrograms daily produces a cost saving of £248 and generates 0.008 additional QALYs and therefore dominates tiotropium. Similarly, indacaterol 300 micrograms produces a saving of £259 and generates 0.008 additional QALYs compared with tiotropium 18 micrograms daily, and therefore also dominates tiotropium. The authors report that this result is primarily due to a substantially larger 12-week improvement in FEV1 produced by indacaterol compared with tiotropium.

One-way sensitivity analyses showed that indacaterol (at both dosages) dominates tiotropium regardless of the time horizon. Probabilistic sensitivity analysis showed that, at a threshold of £20,000 per QALY, indacaterol is cost effective compared with tiotropium 18 micrograms in 84% of iterations (although the authors do not state which dosage of indacaterol this comparison relates to).

This study was classified as being partially applicable, as it only considers 2 of the interventions of interest. It was categorised as having potentially serious limitations, as it uses a short time horizon in the base case, and does not include treatmentrelated adverse events, and is subject to a potential conflict of interest.

Punekar et al. (2015) conducted a cost–utility analysis with a lifetime horizon comparing umeclidinium/vilanterol combination therapy (LAMA + LABA) with tiotropium monotherapy (LAMA) in patients with COPD in the UK. The study was funded by a manufacturer of umeclidinium/vilanterol. The evaluation used a linked-equation model of COPD, which consisted of a series of regression equations to describe how patients’ baseline variables and disease characteristics (cough/sputum, exacerbations, and FEV1) affected their disease progression and final outcomes (resource usage, HRQoL and mortality) over time. These equations were estimated from the Evaluation of COPD Longitudinally to Identify Predictive Surrogate Endpoints (ECLIPSE) study.

Treatment effect was implemented in the model through the difference in change from baseline in FEV1 at 24 weeks between umeclidinium/vilanterol and tiotropium in four umeclidinium/vilanterol phase 3 clinical trials. Three of these trials are described in the clinical evidence review (Decramer 2014a, Decramer 2014b, and Donohue 2013), and one (Celli 2014) was excluded due to using a umeclidinium dose not licensed in the UK.

Resource use was predicted from a linked equation, based on patients’ intermediate outcomes. Unit costs were taken from standard NHS sources (National Schedule of Reference Costs and PSSRU Unit Costs of Health and Social Care). Cost of treatment with tiotropium was obtained from the BNF (£33.50 for a 30 day supply), and the assumption was made in the base case that the cost of umeclidinium/vilanterol was equivalent to this (although the BNF reports its cost as £32.50 for a 30 day supply). HRQoL was predicted from a regression equation in the form of a Saint George’s Respiratory Questionnaire (SGRQ) score, which was converted to an EQ-5D score via a mapping algorithm.

Base-case results showed that umeclidinium/vilanterol produces an ICER of £2,088 per QALY compared with tiotropium monotherapy. Umeclidinium/vilanterol remained cost effective at a threshold of £20,000 per QALY in scenario analyses using 1- and 5-year time horizons, and in which the benefit of treatment was assumed to only persist for 12 months. Probabilistic sensitivity analysis showed that umeclidininum/vilanterol was cost effective in 85% of iterations.

This study was classified as being partially applicable, as it only assesses 2 of the interventions of interest, and is partly informed by clinical data on a dose of umeclidinium not licensed in the UK. It was categorised as having potentially serious limitations, as it only implements treatment effect via improvement in FEV1, implicitly makes the assumptions that all intermediate and final outcomes of treatment can be explained by change in FEV1, and is subject to a potential conflict of interest.

Ramos et al. (2016) conducted a cost–utility analysis with a 5-year time horizon comparing aclidinium bromide/formoterol (LAMA + LABA) with aclidinium bromide alone in patients with COPD in the UK. This study was funded by a manufacturer of aclidinium bromide. The evaluation used a Markov model with states based on GOLD stages 1, 2, 3, and 4 (FEV1 ≥ 80% predicted, 50%–80% predicted, 30%–50% predicted, and <30% predicted, respectively). In each cycle of the model, patients could remain in the same GOLD stage, change GOLD stage or die. Patients could also experience a hospitalised or non-hospitalised exacerbation or a pneumonia adverse event in each cycle.

Treatment effect was implemented via improvement in FEV1 at 24 months from the ACLIFORM and AUGMENT studies (described in Singh 2014 and D’Urzo 2014), which was incorporated in the model via probabilities of changing GOLD state. After this initial period the assumption was made that all patients experienced a uniform decline in FEV1 regardless of treatment received. Exacerbation rates stratified by disease severity were taken from previous trials of tiotropium, ipratropium, and salmeterol, but were assumed not to be directly affected by treatment.

The analysis included four categories of cost: (1) maintenance costs, for which resource use data were taken from a trial of tiotropium conducted in the Netherlands, stratified by disease severity, with unit costs taken from standard NHS sources; (2) exacerbation costs, which were taken from a previous economic analysis; (3) drug costs, which were taken from the BNF; and (4) cost of a pneumonia adverse event, which was based on HRG data. Baseline utility scores according to severity were taken from a previous quality of life study of COPD patients from the UPLIFT trial, with utility reductions of 15% and 50% for moderate and severe exacerbations respectively, as per the methods of previous economic analyses. A disutility of 50% was also assumed for a pneumonia event.

Results showed that aclidinium bromide/formoterol produces an ICER of £2,976 per QALY compared with aclidinium bromide alone. Aclidinium bromide/formoterol remained cost effective at a threshold of £20,000 per QALY in scenario analyses in which alternative lower values were used to inform patients’ baseline FEV1, and in which 1- and 15-year time horizons were used. Probabilistic sensitivity analysis showed that aclidinium bromide/formoterol was cost effective in 79% of iterations.

This study was classified as being partially applicable, as it only includes 2 of the interventions of interest. It was categorised as having potentially serious limitations, as it did not incorporate the effect of treatment on exacerbations in the analysis (only the effect of treatment on FEV1), did not incorporate treatment-related adverse events other than pneumonia, and is subject to a potential conflict of interest.

Economic model

This section summarises the de novo economic modelling conducted for this review question. For a full, comprehensive description of methods, results and conclusions please refer to the model report in Chapter H.

Methods

Model structure

In order to represent the natural history of COPD over time, the model uses a Markov structure, with states based on GOLD severity stages defined by FEV1 percent predicted (shown in Figure 1). In each cycle of the model, patients have a probability of moving to a more severe GOLD stage (defined by the natural rate of FEV1 decline over time), and a probability of death (defined by stage-specific mortality rates). In the first cycle of the model, patients may move to a less severe GOLD stage, in order to reflect the initial FEV1 benefit from initiating long-acting bronchodilator therapy.

In each cycle, patients can also experience a hospitalised or non-hospitalised exacerbation, or an adverse event. The model uses a 3-month cycle length, which was deemed an appropriate period of time to capture progression between states, as well as interfacing well with clinical trial data on long-acting bronchodilators, which typically use 3-, 6-, or 12-month endpoints.

Figure 1

overall structure of the model.

The model also simulates patients’ treatment progression over time. In each cycle, patients have a probability of either stepping up their treatment (adding in another drug) or switching their treatment (changing to a regimen of the same number of drugs). The pathway for treatment progression is shown in Figure 2. While triple therapy (LAMA+LABA+ICS) was outside of scope of the guideline update, this regimen is typically provided for patients whose symptoms are not adequately controlled by dual therapy (as per the recommendations in the 2010 update of this guideline), and is therefore included as a final step in the modelled pathway.

Figure 2

treatment progression pathway in the model.

Incorporating treatment effects

Treatment benefits

The network meta-analysis (NMA) conducted for this review question provided a number of outcomes which could be used to model treatment benefit: exacerbations, SGRQ, FEV1, and TDI. However, independently incorporating all of these outcomes simultaneously in the model would introduce double-counting of benefits. Therefore, a number of scenarios were modelled, using the following combinations of outcomes from the NMA:

• Scenario 1: Exacerbations alone
• Scenario 2: SGRQ and exacerbations
• Scenario 3: FEV1 and exacerbations – this scenario was modelled by allowing differences in transition probabilities in the first cycle of the model, with more effective treatments associated with a greater probability of moving to a less severe GOLD stage
• Scenario 4: TDI and exacerbations – this scenario was modelled using coefficients from a regression analysis in order to predict the effect of breathlessness on SGRQ score
• Scenario 5: FEV1, TDI and exacerbations – as above, this scenario used coefficients from a multiple regression analysis in order to predict the independent effect of FEV1, breathlessness and exacerbations in the previous year on SGRQ

Effect on treatment progression

Differences in the probability of stepping up treatment were implemented by assuming an inverse relationship with treatment effect on TDI, since breathlessness provides a reasonable indication of how well patients’ disease symptoms are managed. Differences in the probability of treatment switching were implemented using the discontinuation due to adverse events outcome from the NMA.

Treatment effect on mortality and adverse events

Treatment effect on mortality was applied directly to the baseline mortality rate for each GOLD stage.

Adverse events were categorised as either cardiac, pneumonia, or ‘other’ events. Treatment effects from the NMA for the appropriate adverse event category were applied to these, using total serious adverse events as a proxy for the ‘other’ events category.

Since the mortality and adverse event outcomes from the NMA were generally associated with a high degree of uncertainty, results were presented both with and without treatment-specific differences in these outcomes in 3 scenarios:

• Option A: Treatment-specific differences in adverse events and mortality excluded
• Option B: Treatment-specific differences in adverse events, but not mortality, included
• Option C: Treatment-specific differences in adverse events and mortality included

Costs

Five categories of cost were used in the model

1. Drug costs – acquisition costs of long-acting bronchodilators
2. Maintenance costs – routine healthcare resource use for each GOLD severity stage
3. Exacerbation costs – resource use associated with a hospitalised or non-hospitalised exacerbation
4. Adverse event costs – costs associated with treating acute and chronic adverse events
5. Treatment progression costs – healthcare costs associated with switching or stepping up treatment

Health-related quality of life

Patients’ stable quality of life (QoL) initially depended upon their GOLD stage, with disutilities applied depending on whether patients experienced an exacerbation or adverse event within each cycle.

SGRQ values were used to inform patients’ baseline QoL. These were converted to EQ-5D scores via a mapping algorithm in line with the NICE Reference Case.

Subgroups

As well as modelling the overall population, results were also produced for patient subgroups stratified by high and low risk of exacerbations. These subgroups differed from the overall population in two ways:

1. NMA outcomes for high- and low-risk subgroups were used to model treatment effect, rather than combined outcomes for the overall population
2. Baseline exacerbation rate was stratified according to patients who had experienced one or more exacerbations in the previous year, versus patients who had experienced no exacerbations, for the high- and low-risk subgroups respectively

Results

Results presented in this section are means of 5,000 probabilistic iterations. Structural uncertainty in the model is also addressed stochastically, by randomly selecting 1 of the 5 scenarios for implementing treatment benefit in each iteration. Individual results for these scenarios are presented in Chapter H.

Overall population

Table 4 shows results for the overall population, when treatment effects on adverse events and mortality are excluded. These results indicate that starting treatment on LAMA+LABA is the most cost-effective option (ICER of £3,652 per QALY), with a relatively high degree of certainty (cost effective in 86.4% of iterations at a threshold of £20,000 per QALY).

Table 4

Mean probabilistic results for the overall population. Option A: treatment-specific differences in adverse events and mortality excluded.

Table 5 shows results when the effect of treatment on adverse events is included. These results show that LAMA+LABA still has the highest probability of being cost effective (55.1% at a threshold of £20,000 per QALY), but this result is somewhat less certain than in the previous scenario.

Table 5

Mean probabilistic results for the overall population. Option B: treatment-specific differences in adverse events but not mortality included.

Table 6 shows results when treatment effects on both adverse events and mortality are included. These results show that LABA+ICS is now the strategy which generates the highest number of QALYs, but is associated with a mean ICER in excess of £20,000 per QALY (£21,308 per QALY). Probabilistic sensitivity analysis also shows that there is now a high degree of uncertainty surrounding results (LABA+ICS is cost effective in 37.2% of iterations, and LAMA+LABA in 34.7% of iterations at a threshold of £20,000 per QALY).

Table 6

Mean probabilistic results for the overall population. Option C: treatment-specific differences in adverse events and mortality included.

High-risk population

Table 7 shows results for the high-risk population, when treatment effects on mortality and adverse events are not included. These results show that LAMA+LABA produces a lower mean ICER for the higher risk population than in the overall population (£463 per QALY), and has a high probability of being the most cost-effective treatment (94.7% at a threshold of £20,000 per QALY).

Table 7

Mean probabilistic results for the high-risk subgroup. Option A: treatment-specific differences in adverse events and mortality excluded.

Table 8 shows results for the high-risk population when the effect of treatment on adverse events is included. These results show that, despite slightly higher uncertainty, there is still a high probability that LAMA+LABA is the most cost-effective treatment (75.6% at a threshold of £20,000 per QALY).

Table 8

Mean probabilistic results for the high-risk subgroup. Option B: treatment-specific differences in adverse events but not mortality included.

Table 9 shows results for the high-risk population when treatment effects on mortality and adverse events are included. Results show that uncertainty increases substantially when mortality effects are included, but LAMA+LABA still shows a considerably higher probability of being the most cost-effective treatment than any other strategy (60.6% at a threshold of £20,000 per QALY).

Table 9

Mean probabilistic results for the high-risk subgroup. Option C: treatment-specific differences in adverse events and mortality included.

Low-risk subgroup

Table 10 shows results for the low-risk population, when treatment effects on mortality and adverse events are not included. LAMA+LABA is associated with the highest probability of being the most cost-effective treatment (58.9% at a threshold of £20,000 per QALY), although there is substantially less certainty in the probabilistic results than in the equivalent scenario for the overall population and high risk subgroup.

Table 10

Mean probabilistic results for the low-risk subgroup. Option A: treatment-specific differences in adverse events and mortality excluded.

Table 11 shows the results for the low-risk population, when treatment effect on adverse events is included. In this scenario, the mean ICER for LAMA+LABA exceeds £20,000 per QALY (£24,495), and LABA -to- LAMA+LABA shows the highest probability of being cost effective (32.0% at a threshold of £20,000 per QALY), but no one strategy is clearly the optimal choice.

Table 11

Mean probabilistic results for the low-risk subgroup. Option B: treatment-specific differences in adverse events but not mortality included.

Table 12 shows the results for the low-risk population when treatment effects on mortality and adverse events are included. Results show that, in this scenario, strategies containing LABA and LABA+ICS have a higher probability of being cost effective than other strategies (38.6% at a threshold of £20,000 per QALY), although no one strategy is clearly the optimal choice.

Table 12

Mean probabilistic results for the low-risk subgroup. Option C: treatment-specific differences in adverse events and mortality included.

Evidence statements

The committee’s discussion of the evidence

The committee used the evidence for this question, the new economic model and the evidence from the LAMA monotherapy review below to make a number of related recommendations for the use of inhaled therapies in people with COPD. Their discussions for both reviews are contained in the section on LAMA monotherapy.