A Markov model consisting of three health states (i.e., progression-free state, progressive state, and death) was constructed in Microsoft Excel (Table 1 for model parameters). Transition probabilities between states were calculated using survival data from the ASPECCT trial, a head-to-head, non-inferiority phase 3 randomized trial that compared panitumumab and cetuximab in mCRC patients with wild-type KRAS who had failed (i.e. disease progression) or were intolerant to irinotecan-based and oxaliplatin-based therapy. The model assumed a 2-year time horizon, given the limited life expectancy of the target population. Patients were assumed to take either panitumumab or cetuximab until disease progression. Upon disease progression, all patients were assumed to receive either regorafenib or TAS-102 with equal probability, before receiving best supportive care.
The cycle unit of the model was 1 month, with a half-month correction. The model considered a 61-year-old male as the base-case patient, to match the baseline patient characteristics of the ASPECCT trial. All dosage and other relevant parameters were based upon this base-case. The primary endpoint of the model is quality-adjusted life-years (QALYs). Other outputs included discounted and undiscounted life-months and costs. We calculated both incremental cost-effectiveness ratios (ICER) and incremental net monetary benefits (INMB).
Transition probabilities
Median overall survival (OS) and median progression-free survival (PFS) of patients taking panitumumab and cetuximab were obtained directly from the ASPECCT trial. The decreasing exponential approximation of life expectancy (DEALE) method was used to calculate the monthly probabilities based on these medians. The DEALE method assumes that patients have a constant hazard of death throughout the time period being modeled [8, 9]. This assumption was appropriate in our case, given the short life expectancy of the patient population. In addition, the transition probabilities were adjusted for background all-cause mortality which can be obtained from the Actuarial Life Table of Social Security website, which in this case reflected the mortality rate of a 61-year-old US male.
Direct costs
Direct costs considered by the model included the cost of drugs, administration, and management of adverse events (AEs) associated with treatment that was incurred during the progression-free state, and drug costs and other medical costs during the progressive state. All costs were converted into 2017 US dollars using the medical services component of the consumer price index.
Progression-free state
The prices of the two EGFR antibodies were obtained from the 2017 ASP Drug Pricing Files from Centers for Medicare and Medicaid Services, and we used ASP + 6% in calculating the drug acquisition cost [10]. Required doses of the two agents were calculated according to the prescribing information. Panitumumab is given at the dose of 6 mg/kg every 2 weeks, and cetuximab is given at 400 mg/m2 at the initial dosing and followed by 250 mg/m2 for every week. An alternative schedule (500 mg/m2 every 2 weeks) of cetuximab, which is more commonly used in clinical practice, was tested in the sensitivity analysis. The height and weight of the base-case needed to calculate the dose were obtained from CDC National Health Statistics Reports [11]. The cost of administration and other infusions was obtained from the Medicare Coding and Payment for Drug Administration Services under the Physician Fee Schedule [12]. Premedication drug costs and any related administration costs were also captured in the model.
The model considered costs for managing AEs with grade 3 or higher severity, except for infusion reactions where all grades were included. Since the reported incidence rate for grade 3 or higher AEs in the ASPECCT trial was very similar between the two treatment groups, the costs would be fully offset when calculating the incremental costs. Hence, the model only included AEs that differed between the two groups by over 1%, i.e. skin rash, hypomagnesaemia, hypokalemia and infusion reaction. Assumptions for managing AEs were based on recent published guidelines and literature [13,14,15].
Progressive state
The model considered the drug cost of regorafenib or TAS-102, and the cost of best supportive care after the disease progressed. Again, the drug costs were obtained from the Medicare ASP and fee schedule. As for the cost of best supportive care (BSC), Lang et al. estimated a lifetime, as well as phase-specific costs for CRC patients with different stages in the US [16]. For patients with metastatic CRC and those with an expected survival of less than 13-months, the annual BSC cost estimate was $31,980 after adjusting for inflation. This was applied in our model as total direct cost of the best supportive care in progressive state.
Indirect costs
Caregiver’s time and lost alternative compensation due to caring for the mCRC patients were included in the model as indirect costs. Van Houtven et al. reported the estimated economic burden of informal caregivers of CRC patients during different phases [17]. Their study reported the cost of caring for patients during the initial phase (first year after diagnosis, not within 6 months of death), continuing phase (after 1 year, not within 6 months of death), and terminal phase (within 6 months of death) of disease. In our model, we assumed the caregiver costs in the progression-free state to be the cost of caring for patients in the continuing phase, which was estimated to be $24,423 over the accumulated 16.7 months after adjusting for inflation. This is justifiable because the patients in our model were receiving third- or further lines of chemotherapy, and should no longer be considered in the initial disease phase. The caregiver costs in the progressive disease state was assumed to be the cost of caring for terminal phase patients, which was reported as $17,645 over a 7-month period.
Health state utilities
The utilities of the progression-free state for both panitumumab and cetuximab used in the model were reported from clinical trials. Bennett et al. analyzed patient report outcomes from a phase 3 clinical trial (NCT00339183) evaluating panitumumab plus FOLFIRI compared to FOLFIRI alone as second line therapy for mCRC [18]. The patients were all wild-type KRAS carriers and the HRQOL was measured by the EQ-5D Health State Index. The baseline EQ-5D was 0.769 in the panitumumab + FOLFIRI group, and according to the mixed model, the change from baseline score until disease progression was − 0.024. For cetuximab, we used utilities reported by patients using Health Utilities Index 3 (HUI-3) from a phase 3 trial evaluating cetuximab versus best supportive care among patients with chemo-refractory mCRC [19]. The mean of utilities assessed at different weeks until progression from the trial was used as the model input for cetuximab.
The utility of progressive state in the model was set to be same between the two arms, considering no more treatment toxicities after stopping EGFR inhibitor and both groups should develop similar disease patterns. The utility value was referenced to Ramsey et al. which has reported a HUI-3 utility of CRC survivors with less than 1-year expected survival [20].
Sensitivity analysis
Both univariate sensitivity analysis and probabilistic sensitivity analysis were performed to evaluate the robustness of the model and address uncertainty in the estimation of model parameters. Survival data were varied over their 95% confidence interval reported from the trials; costs and utilities were varied within ± 20% of their baseline values. The alternative dose schedule of cetuximab was also tested.