Decision model
A Markov model (Figure 1) was built using an in-house code to simulate the clinical history of one hypothetical cohort of women who received PMRT with a prescribed dose of 50 Gy in 25 fractions. The cohort consisted a population of 55-year-old postmenopausal women with Stage III breast cancer (node positive with tumor diameter smaller than 2cm) after mastectomy, and the planning target volume (PTV) for these patients included the left chest wall, left supraclavicular and axillary area, and internal mammary chain area. Markov simulation allowed these patients to transition between different health states, including no evidence of disease (NED), distant metastasis, local recurrence, late radiogenic side effects, and death, in a fixed increment of time (one year). The primary endpoints of this study included quality-adjusted life years (QALYs) from a payer perspective over a 15-year horizon. Treatment strategies associated with lower costs and higher QALY were considered dominant. Incremental cost-effectiveness ratios (ICERs), which was defined as the incremental cost divided by the incremental QALY gained, were calculated in scenarios where there was no dominant strategy. We will determine whether a PMRT modality is cost-effective by comparing ICER with common willingness-to-pay (WTP) thresholds of $50,000/ QALY and $100,000/QALY. [20, 21]
Model data input
The transition probabilities for SOC PMRT were taken from literature. [12, 18, 22-27] Table 1 shows transition probabilities for 55-year-old PMRT patients receiving SOC PMRT: each baseline value of transition probability was summed over years been studied and the value divided by the number of years was used in the Markov model. Table 1 also shows the utilities values for each health state in the Markov model.
For advanced PMRT techniques, the transition probabilities were largely lacking in the literature. In this study, probabilities of tumor coverage including local recurrence and metastasis after advanced PMRT were assumed to be the same as those after SOC PMRT, while probabilities of radiogenic side effects after advanced PMRT were calculated based on normal tissue complication probability (NTCP), [28] lifetime attributable risk (LAR) of second cancers [29, 30] and risk of coronary events (RCE) [12, 31] models for a 55-year-old cohort as we previously reported [5, 8] (Table 2). We assumed that all lung and cardiac events start from year 11 after radiotherapy, [12, 32] while contralateral breast events start from year 6 after radiotherapy. [26]
The annual mortality rates due to breast cancer were derived from Early Breast Cancer Trialists' Collaborative Group (EBCTCG) [1]. The normal death rates were based on the United States life tables. [30] The death from the radiogenic side effects was mainly caused by cardiac toxicity and second cancers. [25]
Table 3 shows the costs for PMRT patients using different technologies from payer perspective and these costs were based on local Medicare charges. Costs of treating late effects were also included in Table 3. All costs and utilities were discounted at 3% per year as recommended by US Panel on Cost-Effectiveness in Health and Medicine. [33]
Model calibration & validation
CancerMath is the latest web-based breast cancer prognostic tool that can predict mortality rate in each year for the first 15 years after the current SOC treatment, and the external validity of our model was assessed by comparing 15-year overall survival and breast cancer mortality of patients who received SOC PMRT with the predicted results from CancerMath.
Sensitivity analyses
We performed a series of one-way sensitivity analyses to determine the variability in the ICER as a function of the probabilities, utilities, and treatment costs of contralateral breast cancer, lung cancer and heart toxicities for seven advanced PMRT techniques versus SOC PMRT.
We also performed probability sensitivity analyses (PSA). The probabilities, utilities and costs were varied simultaneously across their distributions using a second-order Monte Carlo simulation. Transition probabilities and utilities were modeled using a beta-distribution and cost was modeled using a gamma distribution as recommended in the literature.[34] The cost-effectiveness acceptability was calculated based on the result of 100,000 simulations for each PMRT technique at different WTP thresholds.