Model structure
A partitioned survival model was developed in Microsoft Excel 2016 (Microsoft Corp., Redmond, VA) to track the disease course and survival of adults with advanced ALK+ NSCLC who had not previously received treatment with systemic anti-cancer therapy. Partitioned survival analysis is a common approach in modeling advanced or metastatic cancers and has previously been used in many cost-effectiveness analyses (CEA) in a NSCLC setting.10-15 The approach uses clinical endpoints in trials to estimate the distribution of patients across various health states over time using the area under the survival curves. The model comprised three mutually exclusive health states: (1) stable disease; (2) progressed disease; and (3) death (Figure 1). The proportion of patients in the progressed state at each cycle was calculated as the difference between the proportion of patients surviving (i.e., OS) and the proportion of patients remaining in the stable state (i.e., PFS).
At model entry, all patients were in the stable disease (starting age: 54 years), which reflected the median age of the patient population in ASCEND-4.6 In any of the following cycles, patients either stayed in the same health state or transitioned to a progressive disease state or death, which was the absorbing state.
The model compared ceritinib 450mg orally once daily to crizotinib 250mg orally twice daily, both are currently licensed dosages for the treatment of patients with ALK+ NSCLC in HK. Patients were modeled with 1-month cycles over a 20-year time horizon which comprehensively captures the expected costs and health outcomes of patients over their remaining lifetime from first-line treatment initiation. In the base case analysis, both costs and effectiveness were discounted at a rate of 3.5% per year, as recommended by the National Institute of Health and Care Excellence (NICE) in the UK. The model was developed from a perspective of healthcare service providers or the government, thus only direct healthcare costs were considered.
Model assumptions
Due to the limited follow up and difference in primary endpoint of ASCEND-8,7 the efficacy and safety data for the ceritinib arm were based on ASCEND-4,6 which studied the efficacy of ceritinib 750mg orally. The ASCEND-8 trial showed that ceritinib 450mg, taken once daily with food demonstrated comparable pharmacokinetics as daily dose of 750mg in a fasted state. Therefore, the efficacy of ceritinib 750mg and 450mg were assumed the same based on the ASCEND-8 data, regulatory agencies (e.g. Food and Drug Administration, European Medicines Agency, HK Department of Health) have replaced the 750mg (in a fasted state) with 450mg (taken with food) as the standard dose for ceritinib. Patients in the stable disease state incurred drug and drug administration costs, a one-off treatment-associated adverse event (AE) cost, and medical costs associated with resource use. Patients in the progressed disease state incurred medical costs and additional costs of second-line treatments, which reflected those observed in the respective trials.6,16 All patients incurred one-off terminal care costs before death, irrespective of whether the death was related to NSCLC. Health utilities were dependent on health state, and additionally, the utility value for stable disease depended on the first-line treatment received. Thus, disutilities associated with AEs were not considered to avoid double counting. Finally, ALK testing was assumed to be a routine diagnostic test in HK and was therefore not considered as a cost component in the analysis.
Model inputs
Efficacy inputs
Time spent in the stable disease, progressed disease, and death states were estimated for both ceritinib and crizotinib arms based on respective areas under the PFS and overall survival (OS) curves during the model’s 20-year time horizon.
For ceritinib, survival curves were estimated and extrapolated beyond the trial period by fitting parametric models to individual patient level data for PFS and OS from the ASCEND-4 trial.5 Proportional hazards models based on Weibull, exponential, and Gompertz distributions were applied based on the methods recommended by the NICE decision support unit.17 Selection of base-case models was informed by comparisons of goodness-of-fit criteria, evaluation of log-cumulative hazard plots, and expert opinion on the clinical plausibility of long-term outcome predictions. For PFS, while the Gompertz function demonstrated the best fit based on AIC/BIC, the long-term extrapolation using the exponential function were considered to be more clinically plausible. For OS, exponential curve demonstrated the best fit to match the observed OS data and the log-cumulative hazard plot was also linear in shape. Based on that, the exponential function was selected as the base-case parametric model for both PFS and OS and other parametric proportional hazards models (e.g., Weibull, Gompertz) were explored in the sensitivity analysis.
For crizotinib, PFS and OS were extrapolated by applying the corresponding hazard ratios (HRs) for crizotinib vs. ceritinib to the PFS and OS parametric models of the ceritinib arm. In the absence of head-to-head randomized trials comparing ceritinib to crizotinib in previously untreated advanced or metastatic ALK+ NSCLC, an indirect comparison needed to be considered. Although the patient population enrolled in ASCEND-4 and PROFILE 1014 trials and both trials included chemotherapy as a comparator arm, the chemotherapy arms were significantly different to be used as a common comparator. Therefore, ‘anchor-based’ analysis was not feasible and a matching-adjusted indirect comparison (MAIC) in a non-anchor based setting was used to compare efficacy outcomes between crizotinib and ceritinib.18 The MAIC indirectly compared the ceritinib arm from ASCEND-45 to the crizotinib arm from PROFILE 1014,16 while adjusting for cross-trial differences in patient characteristics between the two trials. Compared to crizotinib, ceritinib was associated with a significantly longer PFS (HR=0.64; 95% confidence interval [CI]=0.47-0.87) and median PFS of 25.2 vs 10.8 months; the two treatments were similar with respect to OS 0.82; 95% CI = 0.54, 1.27) (Table 1).18 Predicted PFS and OS for both treatments are shown in Figure 2.
Resource use and cost inputs
All costs were converted to 2018 US Dollars (USD) from Hong Kong Dollars (HKD) using a prevailing exchange rate obtained from the Linked Exchange Rate System in Hong Kong (1 USD = 7.8 HKD).19
Cost of first-line treatment
Monthly drug costs for ceritinib and crizotinib were estimated based on local unit drug costs, the licensed dosage, relative dose intensity and duration of treatment. Costs per capsule for ceritinib and crizotinib were obtained from the Hong Kong Hospital Authority databases.20 Mean relative dose intensity for ceritinib (450mg) were based on the ASCEND-8 trial.7,16 Because the relative dose intensity for crizotinib was not reported in PROFILE 1014, it was obtained from PROFILE 1007, a Phase III open-label trial for crizotinib among previously treated patients with ALK+ NSCLC.21 No administration costs were assumed for oral drugs.
Patients were assumed to be on first-line treatment until discontinuation or progression, whichever occurs first. The proportion of patients on treatment was estimated based on PFS multiplied by the ratio of the proportion of patients remaining on treatment and the proportion of patients in PFS, which was estimated for each cycle using patient-level data from the ASCEND-4 trial.6 For crizotinib, the ratio was assumed to be the same as that of ceritinib.
Cost of subsequent therapy
Post-progression treatment costs with active agents (e.g., ceritinib, crizotinib, docetaxel, and platinum based chemotherapy22) were applied to both treatment arms for patients who newly entered the progressed disease state in each model cycle. The model assumed that 20% of post progression patients did not receive any further systemic therapy due to rapid performance deterioration or death. Among patients who initiated systemic second-line therapy after progression, the distribution of patients across second-line treatment options was estimated based on the relative frequency of different second-line treatments observed in ASCEND-4 and PROFILE 1014 trials.6,16 Drug unit cost, drug administration cost, and dosing schedule for chemotherapy were obtained from the Hong Kong Hospital Authority databases.20 Mean relative dose intensity and treatment duration for each treatment option were obtained from trials in previously treated ALK+ population.21,23 Body surface area (BSA) used to estimate required chemotherapy doses was based on clinical practice experience by the investigators.24
Cost of AEs
Grade 3/4 AEs were included if they were reported in ≥ 5% of patients for at least one treatment in the model. The rates of AEs were based on rates reported in the ASCEND-8 trial for ceritinib,7 and the PROFILE 1014 trial for crizotinib.16 Patients were assumed to incur a one-time cost for the management of AEs. AE management procedures were based on the clinical opinion of the investigators. The costs of AEs were obtained from the list of charges of the Hong Kong Hospital Authority.25
Disease management cost by health state
Medical costs incurred during stable and progressed disease states were determined by monthly frequencies based on expert opinion and unit costs from the Hong Kong Hospital Authority databases.25 Resource use during stable disease included visits to cancer nurses, pharmacists and outpatient visits. Tests and procedures performed included complete blood counts, computed tomography scans, magnetic resonance imaging, X-rays, and serum chemistry testing on a regular basis. In the progressed disease state, resource use for cancer nurse visits, pharmacist visits, outpatient visits, palliative medications, and home oxygen were included as well as all other tests and procedures from the pre-progression state. With respect to terminal care costs, all patients who transitioned to death were assumed to incur a one-time terminal care cost of $24,089.26 This figure was based on a study of cost effectiveness of biennial mammography in women HK, that estimated terminal care cost during the 6 months before death.26
Utility inputs
In the base case, utilities associated with stable disease for ceritinib were based on EQ-5D data from the ceritinib arm in the ASCEND-4 trial6 and the utilities for crizotinib were obtained from an analysis of EQ-5D data from PROFILE 1014 reported by Felip et al.27 For utilities in the progressed disease state, EQ-5D scores were not collected systematically after treatment discontinuation in ASCEND-4 or PROFILE 1014 trials. Given this lack of information, for all treatment arms, the base-case utility value for progressive disease (0.64) was estimated based on the utility study by Chouaid et al.28 To avoid double-counting the impact of AEs on quality of life, AE disutilities were not applied.
Model outputs
Total costs and effectiveness outcomes were estimated for each arm during the 20-year model time horizon. Effectiveness outcomes included total life years (LYs), and quality-adjusted life years (QALYs). The incremental cost-effectiveness ratio (ICER) of ceritinib compared to crizotinib was evaluated using incremental cost per QALY gained and incremental cost/LY gained.
Sensitivity analyses
A deterministic sensitivity analysis (DSA) was performed to test the robustness of the model. The model uncertainty was assessed by varying one model input or assumption at a time. These included drug cost unit cost, efficacy for both arm, relative dose intensity, resource use, utility, discount rate, and time horizon. The detailed list of DSA inputs and their variation are illustrated in Figure 3.