COPD is a chronic degenerative disease with recurring exacerbations. A Markov model provides a suitable structure to model such a disease. A Markov model comprises discrete health states that describe the status of a patient. Patients stay in one health state for the duration of a model cycle, accumulating the costs and benefits associated with that health state. At the end of a cycle, patients may remain in a state or move to another state, according to defined transition probabilities.
The model was developed as a global model allowing local country adaptations to inform the cost-effectiveness of indacaterol compared with current treatments. A previous publication on the cost-effectiveness of indacaterol in the German setting presents in detail the methodology used to develop the model . This paper presents the specific methods and results for the UK setting.
This analysis used a Markov model constructed in Microsoft Excel 2007 (Microsoft, Redmond, Washington) with four main health states describing the different COPD severities plus a state for death. For each disease severity state, two further health states were created for patients who experienced a severe (requiring hospitalisation) or non-severe exacerbation (requiring medical management in the community), to give a total of 12 health states plus death, as shown in Fig. 1. Mirror states were created for the 12 health states to describe the disease progression of patients who discontinued therapy.
Cycle lengths were set to 3 months in order to capture initial lung function improvement. Forced Expiratory Volume in 1 second (FEV1) was the primary endpoint of the indacaterol studies and most clinical studies show that improvement in trough FEV1 is recorded over the first 8–16 weeks of treatment. Trough FEV1 versus placebo after 12 weeks of treatment was the primary outcome of the indacaterol studies.
Discounting was set to 3.5 % per annum for both costs and outcomes in line with the NICE guidelines for health technology assessments . The time horizon in an economic evaluation should cover the full period over which the benefits and costs of an intervention are expected to be accumulated . Although a lifetime time horizon is recommended in the UK ; in the base case, the analysis was run over a 3-year period. The rationale for conducting the analysis over a shorter period was that the relevant clinical trials for indacaterol extended to 6 months and although it is reasonable to expect patients to take indacaterol for longer time periods given that COPD is a life-long, progressive illness; extrapolation of 6-month data to very long time periods would lead to unacceptable levels of uncertainty in the analysis. Several published economic evaluations in COPD have also used shorter than lifetime time horizons [10–13]. A 3-year duration was considered sufficient to capture costs and benefits, given that many patients are likely to step up or discontinue their maintenance therapy after 3 years . The results for a 5-year and lifetime time horizon are also presented.
The model population was based on the patients enrolled in the clinical trials who had moderate to severe COPD (classified by post-bronchodilator FEV1 between 30 % and 80 % of the normal population), a mean age of 63.6 years, and 67 % were male. The starting distribution in the model reflected the baseline distribution of each trial between health states and is presented in Table 1.
COPD guidelines recommend post-bronchodilator measurement of FEV1 for assessment of disease severity , but apart from the screening visit, these were not available from the indacaterol trials. As a consequence, and in line with previous economic analyses [11, 15], pre-bronchodilator values were used to describe efficacy, utilities and medical resource use for the disease severity health states. The available literature all followed the convention of using pre-bronchodilator values to define these data and since pre-bronchodilator values were used for all comparators the impact on incremental results was minimal. The cut-off points adopted in the clinical trials to define COPD severity were the same as the Global Initiative for Chronic Obstructive Lung Disease (GOLD) , which compares patient FEV1 with FEV1 for the normal population to ascertain percentage of normal lung function.
A cohort analysis was conducted with the Markov model, based on a population of 1,000 patients. Expected costs and outcomes were used to estimate a total cost, total life-years gained, total severe and non-severe exacerbations and total quality adjusted life-years (QALYs) gained per patient. Incremental cost-effectiveness ratios were reported as cost per QALY gained.
Both one-way and probabilistic sensitivity analyses (PSA) were conducted to test the assumptions of the model and the overall robustness of the results. Tornado diagrams give an indication of the parameters/assumptions which are particularly important contributors to the model’s results. Parameters from clinical trial data were varied within the 95 % CIs and parameters for which data on uncertainty were not available were varied by ± 30 %. For the PSA, costs and resource use were varied according to a gamma distribution, utilities by a beta distribution and rate ratios by a log-normal distribution. These distributions were stochastically sampled for 1,000 iterations and the ICER output was recorded. The results of the PSA are shown on a cost-effectiveness plane, where the incremental costs and incremental QALYs for each iteration of the PSA are plotted. The quadrant in which the ICER falls is informative of the cost-effectiveness of the treatment under evaluation.
A threshold analysis was also conducted to assess the impact of using lower prices for tiotropium on the cost-effectiveness results. This is aimed to reflect the situation when branded tiotropium loses exclusivity in the UK.
Clinical Model Inputs
Two key, 6-month trials for indacaterol informed the clinical inputs for the model. The trials were both multi-centre, placebo-controlled, randomised studies with active controls [4, 5]. The INLIGHT-2 trial was a double-blind study comparing indacaterol (Onbrez Breezhaler®, Novartis) 150 μg daily, salmeterol (Serevent®, A&H) 50 μg twice daily and placebo ; the INHANCE trial compared indacaterol 150 or 300 μg daily with open-label tiotropium (Spiriva®, Boehringer Ingelheim) 18 μg daily . Each trial provided data on the rate of exacerbations and the improvement in lung function (trough FEV1 at week 12 was the primary endpoint). Detailed patient characteristics for these trials have been previously published .
Data from the trials using indacaterol 150 μg dose is presented as the base case as this is the most commonly recommended dose licensed in the UK. Patients are allowed to increase their dose to the 300 μg dose on medical advice, which has been shown to provide additional clinical benefit for patients with severe COPD and therefore the results of the comparison of indacaterol 300 μg versus tiotropium 18 μg have also been presented. Indacaterol 300 μg was not included in the trial versus salmeterol and therefore there are no efficacy results available to inform this comparison in the model.
Transition probabilities used in the model were based on transition matrices constructed from individual patient level data, which tracked the movement of patients over the first 12 weeks across different disease states. This allowed the model to fully capture change in disease states rather than using aggregated average improvement in FEV1 levels. As discussed above, the primary endpoint of the indacaterol trials was FEV1 improvement at week 12 which is a commonly used regulatory time frame cut-off for registration clinical studies. The transition matrices describing the movement of patients over the first 12 weeks in the INHANCE and INLIGHT-2 trials are presented in Table 2. This transition matrix was applied to the first cycle (12 weeks) of the model only, to describe the initial improvement in FEV1 after starting therapy. After the first cycle, all patients experience a uniform lung function decline over the model duration (Table 3).
The observed exacerbation rates from the placebo control arms of the indacaterol trials were used in the model. Annual rates in the placebo arm were similar across the trials ranging from 0.60 to 0.74 exacerbations per patient per year and these rates set the overall number of exacerbations that occurred in the model. The definition of an exacerbation in the model was based on resource use and differs from that of the clinical trial; severe exacerbations were those requiring hospitalisation and non-severe were those requiring a change in medication and/or contact with a healthcare provider. The probability that an exacerbation would be severe or non-severe in each of the severity stages was based on the clinical trial data. Exacerbations from the placebo arm of the indacaterol trials were pooled and the most recent pre-bronchodilator FEV1 value recorded before an exacerbation was used to classify the disease severity status in which the exacerbation took place.
The treatment effect on exacerbation reduction was described as rate ratios and based on the respective studies. Annual rates of exacerbations for the active arms were low in the clinical trials (typically around 0.5 per patient for all treatments) since the study was not enriched to recruit an exacerbating population. All rate ratios were applied to the baseline rate of exacerbations, and therefore rate ratios versus placebo were used. Annual exacerbation rates could not be derived for each treatment by disease severity from the trial data due to the small number of observed events in each study (Table 3).
Other Clinical Inputs
A differential rate of lung function decline was not demonstrated in these two studies and the rate of lung function decline for COPD patients was therefore derived from the UPLIFT  trial, which gave the annual rate of decline in pre-bronchodilator FEV1 measurements as 30 ml per year with a standard error of 1 ml; the UPLIFT trial is one of the longest COPD trials conducted recently (Table 3).
Two different mortality rates were utilised in the model in order to fully describe the death rate of a COPD population. The first was a COPD-related mortality and the second was an all-cause mortality rate used to account for deaths from competing causes in the study population. COPD-related mortality was obtained from a published study which gave mortality rates for a COPD patient cohort grouped by disease severity . While this source accounts for all-cause mortality within this cohort, we have made the assumption that all deaths recorded within this study were COPD-related, since separate measures of COPD-attributable death by disease severity (required to separate the types of death) were not available (Table 3). All-cause mortality was obtained from UK interim life tables for 2008–10 issued by the Office for National Statistics . In the model, all-cause mortality was not adjusted for COPD-specific deaths due to lack of data.
Economic Model Inputs
Unit costs were obtained from a combination of UK cost sources and costs from the literature. Table 4 shows the unit costs for all healthcare products and services used in the model. All costs are expressed as Pounds Sterling (£) for the cost year 2011 and have been inflated to 2011 prices using the Hospital and Community Health Services Index  where necessary. Resource use from the OPCRD  and the literature was validated with a UK clinician with expertise in COPD management.
The cost of the different disease states and exacerbations consist of costs accrued in the community and the hospital setting. The cost of community-based care providers were taken from the Personal Social Services Research Unit (PSSRU)  and all hospital services including ambulance, admission costs, professional and lab services were taken from the NHS reference costs 2009–2010 .
Drug prices were obtained from the British National Formulary (BNF) . For COPD-specific drugs, the recommended dose and package size was used to calculate the cost per day. For the influenza and pneumococcal vaccine, the unit cost to the Department of Health was used, which includes the cost of the vaccine and the GP administration fee . The OPCRD  reports the number of prescriptions for a specific drug class over a 12-month period. Therefore for these drugs (short-acting beta-2 agonists (SABA), short-acting antimuscarinic bronchodilator (SAMA), inhaled corticosteroids (ICS), leukotriene receptor antagonists (LTRA), theophylline, mucolytics and oral corticosteroids), the Prescription Cost Analysis dataset 2011  was used to ascertain the average cost per prescription. The weighted average cost per prescription of the five most common products in each of the drug classes was used.
Home oxygen therapy in the UK is supplied by the Home Oxygen Service which uses specific contractors to supply oxygen in patients’ homes . The cost of this service was not readily available and therefore the cost per day of oxygen therapy reported in a published cost-effectiveness analysis  has been inflated and converted from local currency and used in the model. NICE clinical guidelines recommend that all very severe COPD patients are assessed for home oxygen . It is reported that approximately 40 % of very severe COPD patients should be on home oxygen, however currently it is believed that only half of those who require home oxygen receive it . In the base case, it is assumed that all patients who require home oxygen receive it (i.e. 40 % of very severe COPD patients  and 5 % of severe COPD patients).
There was limited evidence on the cost of delivering pulmonary rehabilitation to COPD patients in the UK. The cost estimate used in the model was taken from a report by the Department of Health on the development of a national strategy for COPD services .
This analysis used real-life resource use from the OPCRD . Of the over 28,000 COPD subjects in the database, 20,001 patients had a confirmed diagnosis with an FEV1/Forced Vital Capacity (FVC) ratio < 0.7. Using the 20,001 subjects, the average use of concomitant medications (SABA, SAMA, ICS, LTRA, theophylline, mucolytics and oral corticosteroids), pulmonary rehabilitation and GP consultations related to COPD were calculated for each of the COPD GOLD stages. Resource use data not available in the OPCRD  (respiratory specialist visits, and spirometry) were taken from the literature . Table 5 shows the resource utilization rates for the health states that were applied in the model.
Resource use associated with exacerbations was unavailable from the OPCRD  and therefore are based on assumptions that have been validated by a clinician. For a non-severe exacerbation, it was assumed that all patients visited the GP once, and 50 % took antibiotics and oral corticosteroids for 7 days and 50 % took these therapies for 14 days (an average of 10.5 days for all patients). For severe exacerbations it was assumed that 70 % of patients arrived at hospital in an ambulance before they were admitted. In addition it was assumed that patients took antibiotics and oral corticosteroids for 10.5 days (Table 6).
European Quality of Life – five dimensions (EQ-5D) is a standardised questionnaire used to measure HRQoL and allows the calculation of a utility value for a specific health state . EQ-5D data were collected in the three indacaterol phase III clinical trials at the start of the studies, at week 12 (primary endpoint) and at week 26 (end of the studies) . In the indacaterol trials, whenever an EQ-5D questionnaire was completed at a time for which a pre-bronchodilator FEV1 value was available, the EQ-5D score was labelled as describing the corresponding disease severity health state. Over 11,000 EQ-5D questionnaires were pooled and analysed to describe the HRQoL for the COPD disease severity health states. Although the EQ-5D data were collected from subjects across 21 different countries, the UK National Health Survey preference weights was used to value the utility score [24, 25].
No utility data were available from the trials to describe an exacerbation and therefore values from a published study which measured the disutility associated with an exacerbation was used . This study recruited participants from the general population in the Netherlands and aimed to include subjects who represented the Dutch population in terms of age, sex and education level . Participants were asked to value several COPD health profiles using both the visual analogue scale and the time trade-off method . The utility decrements for COPD exacerbations reported using the time trade-off methodology was used in the model (Table 7).