Model Structure
This cost-effectiveness analysis was primarily based on the clinical data from the EAGER study, which was a randomised, open-label, active-controlled, parallel-arm study [15]. Eligible patients were randomised to TIP or TIS in a 3:2 ratio, and treatments were administered for 28 days followed by 28 days off-treatment, for a total period of 6 months [15].
The model was developed from the US healthcare perspective over a time horizon of 10 years, with a cycle length of 6 months. The 10-year time horizon was considered appropriate to account for potential variances in costs and intrinsic health effects, especially given the median predicted survival of CF patients of 39.3 years and the median age of patients in the EAGER study of 24 years [1, 15]. Other time horizons were also explored (2 and 5 years) to inform shorter term decision making.
The model employed a patient-level simulation rather than a cohort model as the former accounts for patient heterogeneity and allows obtaining unbiased estimates of mean outcomes, thus avoiding the addition of many health states. Moreover, one can reasonably expect a non-linear relationship between patient characteristics (age, baseline lung function, treatment adherence, etc.) and model outcomes [cost and quality-adjusted life-years (QALYs)]. Furthermore, the choice of model structure is consistent with the guidelines published by Brennan et al [22].
The parameters considered in this model (Fig. 1) included decline in forced expiratory volume in 1 s (expressed as FEV1 % predicted), frequency of pulmonary exacerbations, and overall survival. FEV1 is a reflection of the extent of airway obstruction [23], and the rate of FEV1 decline is a marker of progression of lung impairment in patients with CF, long-term morbidity, and mortality [24, 25]. The frequency and severity of exacerbations with FEV1 % predicted (<40, 40–70, >70%) and treatment adherence were also assessed as these tend to provide a clinically relevant endpoint [24], and are correlated with increased morbidity, compromised QoL, and mortality [8, 26]. Moreover, pulmonary exacerbations are the major drivers of increased costs [3, 27].
This model assumed high adherence to TIP and low adherence to TIS and the adherence categories were defined as low (≤2 cycles), medium (>2 to <4 cycles), and high (≥4 cycles) [16]. This assumption of higher adherence with TIP is based on adherence observed in Harrison et al [17]. The model simulated one patient at random, assigning a full set of risk factors based on the EAGER study population (e.g. age, gender, and baseline FEV1 % predicted) (Table 1) [15]. The model recorded the costs and health outcomes for each patient for the entire time horizon, and a total of 5000 patients were simulated to ensure stability of the model results. To further enhance the utility and generalisability of this analysis, two additional scenarios were evaluated: high adherence to both TIP and TIS and high adherence to TIP with medium adherence to TIS. For high adherence to TIP and medium adherence to TIS, the rate of hospitalisation from Briesacher et al [16] was used to calculate the rate of pulmonary exacerbations for TIS based on the rate of exacerbations for high adherence from Bradley et al [28].
Table 1 Baseline patient population characteristics
The mean (standard deviation) improvement in FEV1 % predicted was 3.1% (19.92) for TIP as observed in the EAGER study. Due to the lack of data for improvement of FEV1 % predicted in patients with low adherence to TIS, the mean improvement in FEV1 % predicted of 2.3%, which was associated with high adherence to TIS as observed in the EAGER study, was considered as a conservative measure. In addition, the model used the continual longitudinal decline in lung function in CF patients with chronic P. aeruginosa infection. The rate of lung function decline assessed by FEV1 % predicted was assumed to be constant, and was based on the published studies in patients aged 6–22 years [29, 30], and patients >18 years [31, 32]. The risk factors considered to contribute to lung function decline in children and adolescents in this analysis included crackles, wheezing, sputum production, sinusitis, exacerbations, elevated liver enzymes, and pancreatic insufficiency [32].
Based upon severity, pulmonary exacerbations were classified into two categories, mild (no hospitalisation) and severe (hospitalisation) exacerbations [28]. The rates of severe exacerbations per patient per year based on FEV1 % predicted (<40, 40–70, >70%) were obtained from an observational study [28]. These rates were then converted to probabilities of exacerbations (0, 1, 2, 3, and 4) per patient in 6 months for TIP. For TIS, due to low adherence, the probabilities of exacerbation per patient were calculated based on the odds ratio of 0.4 (probability of hospitalisations in a patient with high adherence compared to a patient with low adherence) [16]. Similarly, the rates of mild exacerbations per patient per year based on FEV1 % predicted (<40, 40–70, >70%) were obtained from the same observational study [28]. These rates were converted to the probability of exacerbations in 6 months for TIP. A conservative assumption was made to consider the same probability of mild exacerbations for TIS (Table 2). The assumption for exacerbation data was based on the observational study [28] rather than the EAGER study. The EAGER study did not measure exacerbations directly; instead, exacerbations were generally reported by the investigators as adverse events using the safety preferred term of lung disorders. Moreover, the EAGER study did not differentiate between minor and major exacerbations.
Table 2 Mean number of exacerbations by patient, per year
The mortality rates were estimated using the survival curve reported by Harness-Brumley et al [33], with an appropriate mortality rate assigned to each simulated patient based on their age and gender.
The utility values were derived from the EuroQoL (EQ)-5D values collected in a UK-based observational study (Table 3) [28]. The EQ-5D data were grouped based on FEV1 % predicted (<40, 40–70, >70%) and type of pulmonary exacerbation. Patients in both treatment groups started with the same base utility values, and these would change with the change in FEV1 % predicted. Disutility (decrement in the normal utility) was applied when an exacerbation occurred. The standard willingness-to-pay (WTP) threshold was considered as US$50,000–100,000 per QALY [34], and a 3% annual discount rate was applied for both costs and benefits. Clinical experts validated the model structure. After 1 year in the model, characteristics of the patient cohort such as age, gender, risk factors (e.g. sinusitis), and number of exacerbations were similar to those of clinical trial data [15]. Data collection and analysis of the results were also validated using Drummond’s checklist [35].
Table 3 Utility values for the cystic fibrosis model
Model Assumptions
The 6-month cycle length was used to match the clinical data in the EAGER study [15]. The treatment administration was considered during the time period the patient was alive or throughout the time horizon, whichever occurred first. The treatment efficacy was considered only in the first cycle due to the lack of long-term data. The safety profile was assumed to be similar for both treatments and was not considered in this model.
Resource Use and Costs
The components included under costs were medication costs, nebuliser costs, regular follow-up costs, and costs of pulmonary exacerbations. The cost-base year was 2016. The total drug costs for three treatment cycles of TIP and TIS (Novartis data on file) were taken at US$26,588 and US$22,013, respectively (package price/month US$8863 and US$7338, respectively) [36]. Additionally, costs associated with nebuliser use were considered for TIS (Novartis data on file), which included recurrent costs for air compressor and aerosol mask (US$80, replaced every 6 months), and PARI LC Plus® and air filter (US$180, replaced every year) [36]. The annual management costs for follow-up were taken at US$2043 (Table 4). Resource utilisation data for pulmonary exacerbations were obtained from a UK-based study due to the non-availability of similar US data [28]. The resource utilisation parameters comprised the average number of routine and specialist physician visits by provider type and the average number of diagnostic tests, including routine laboratory tests and radiologic investigations [28]. The US-specific unit costs were applied to the resource use data to calculate the costs in US dollar (US$). The resource use and unit costs are presented in Tables 4 and 5.
Table 4 Resource use and costs inputs used for the analysis
Table 5 Resource use and costs of pulmonary exacerbations
Model Outputs
Three categories of model outputs were estimated for each treatment arm: clinical (overall survival, QALYs, and number of pulmonary exacerbations), costs (medication costs, nebuliser costs, regular follow-up costs, exacerbation costs, and total costs), and cost-effectiveness (incremental cost-effectiveness ratio, ICER).
Sensitivity Analyses
As this economic analysis was based on heterogeneous data sources with several built-in assumptions, one-way sensitivity analyses were performed by varying different factors to understand their impact on the outcomes. To arrive at a fair point for eliciting important variables, 95% confidence interval (CI) was used for the parameters for which a standard deviation was available, and a 10% standard error was selected for other parameters. A probabilistic sensitivity analysis (PSA) was also performed to further account for uncertainty in the estimated values. The PSA was run for 1000 patients and 100 cohorts using Monte Carlo simulation. Model results were recorded for incremental QALYs gained and incremental cost for TIP versus TIS.