Patients and treatment options
Breast cancer patients were registered in the Breast Cancer Information Management System of West China Hospital, Sichuan University (Sichuan, China) since 1989. Their medical history, pathological diagnosis, and treatment information were prospectively collected by oncologists. Each patient was followed by outpatient visit or telephone at 3 to 4-month intervals within 2 years after diagnosis, 6-month intervals within 3 ~ 5 years, and then annually. Written informed consent was provided by all the patients. Ethical permission was granted by the Ethics Committee, West China School of Medicine/West China Hospital, Sichuan University(approval number 2017 − 255).
From 2011 to 2017, a total of 5,070 early breast cancer patients were diagnosed in the West China Hospital of Sichuan University, of which 4,408 were treated with three main types of surgery. There were 206 cases of MAST + RECON, 425 cases of BCT and 3777 cases of MAST.
Because the baselines of the three groups were not consistent, we used R software to match the propensity scores. Based on the MAST + RECON group, the nearest-neighbor method was used for 1:1 matching. The rest of the statistics were performed using SPSS 25.0 software. The measurement data were analyzed by analysis of variance, the unordered counting data were tested by row × list chi-square tests, and the ordered counting data were tested by rank sum. All the tests were two-sided, and p < 0.05 indicated statistical significance. There was no statistically significant difference in breast lesion location before propensity score matching. The variables of age, medical insurance type, histology type, TNM stage, BMI, HR, patient source, whether to use neoadjuvant chemotherapy, and whether to use targeted therapy were statistically significant (p < 0.05). After matching, there were 206 cases in each of the three groups. There was no significant differences in the general information orclinical characteristics, indicating that the data of the three groups were balanced after matching (Table 1). We conducted survival follow-up for the 3 groups of patients, and the deadline was April 2019.
Table 1
Comparison of general information after matching of propensity scores of early breast cancer patients with different surgical treatment approaches
| | MAST + RECON | BCT | MAST | Statistics | p value |
Age,year | | 38.63 ± 6.94 | 38.60 ± 7.58 | 39.03 ± 7.21 | 0.222 | 0.801 |
Health-care insurance | Provincial medical insurance | 13 | 12 | 11 | 4.367 | 0.359 |
City medical insurance | 102 | 100 | 119 | | |
Other | 90 | 93 | 75 | | |
Lesion location | Left breast | 110 | 112 | 109 | 2.268 | 0.687 |
Right breast | 95 | 93 | 95 | | |
Double breast | 0 | 0 | 1 | | |
Histology type | Ductal carcinoma in situ | 10 | 5 | 6 | 3.888 | 0.692 |
Invasive ductal carcinoma | 158 | 168 | 164 | | |
Invasive lobular carcinoma | 5 | 3 | 2 | | |
Other | 32 | 29 | 33 | | |
TNM | 0 | 4 | 3 | 7 | 1.336 | 0.513 |
Ⅰ | 6 | 5 | 6 | | |
Ⅱ | 195 | 197 | 192 | | |
BMI,Kg/m2 | ༜18.5 | 21 | 15 | 15 | 2.687 | 0.261 |
18.5 ~ 23.9 | 144 | 164 | 148 | | |
༞23.9 | 40 | 26 | 42 | | |
HR | Positive | 105 | 106 | 195 | 0.749 | 0.993 |
Negative | 19 | 17 | 21 | | |
Mixed | 73 | 75 | 73 | | |
Unkonwn / missing | 8 | 7 | 6 | | |
Patient source | In this city | 126 | 129 | 131 | 1.072 | 0.899 |
In this province | 71 | 65 | 66 | | |
Other | 8 | 11 | 8 | | |
Neoadjuvant chemotherapy | No | 165 | 169 | 169 | 0.349 | 0.840 |
Yes | 40 | 36 | 36 | | |
Targeted therapy | No | 167 | 172 | 11 | 0.482 | 0.786 |
Yes | 38 | 33 | 34 | | |
Model Structure
The Markov model of early breast cancer identified in this study has four states: disease-free survival, local recurrence, distant metastasis, and death. Patients with disease-free survival can develop local recurrence and distant metastases, patients with local recurrence can develop distant metastases, and only patients with distant metastases may have breast cancer-related deaths. All of the states can result in death from other causes. Once a patient has died, they cannot transition to other states, so death is also an absorbed state (Fig. 1).
There are three alternative surgical options for confirmed early breast cancer patients: MAST + RECON, BCT, and MAST. The initial age of the cohort after the propensity score in this study was 39 years. Therefore, the Markov model simulates the 60-year outcome of patients after receiving the three surgical routes. The status of all the patients entering the model was the disease-free survival status.
This study analyzed the costs and effectiveness of the three surgical treatment paths from the perspective of society as a whole. The utility analysis used quality-adjusted life years (QALYs) and then weighed the advantages and disadvantages of the three surgical treatment paths. The evaluation index of this method was the ICER, which was the incremental cost-effectiveness ratio, that is, the ratio of the difference between the relative costs and effects of the intervention plan and those of the control plan. In this study, it was necessary to calculate the cost of each additional quality-adjusted life year compared to different treatment strategies. When comparing the ICER with the threshold, if the ICER is less than the threshold, it means that the solution is cost-effective; if the ICER is greater than the threshold, then the solution does not have cost-utility. The threshold for this study, which was WTP, used 3 times China's GDP per capita in 2018[19, 20], which was $27,931.04.
TreeAge Pro 2011 (TreeAge Software, Inc., Williamstown, MA, USA) was used to build and analyze the Markov model. This software is professional software for decision trees and Markov models. This study used a 3% discount rate to discount costs and utility values and applied a half-cycle correction.
Transition Probability
In this study, the transition probability was determined by survival analysis to obtain the time to transition from one state to another state, and then the transition probability was calculated by the formula. The survival analysis curve for each state transition in the study cohort is shown in Fig. 2. If a patient had both recurrence and metastasis, it was counted as metastasis. According to the calculation formula of transition probability, i.e.,r=-[ln(1-P1)]/t1 and P = 1-e-rt2, the transition probability was calculated (Table 2). For example, the follow-up time from modified disease-free survival to local recurrence in this study was 94 months, with a cumulative recurrence-free probability P1 of 0.994. A Markov cycle was 12 months in a year, and the unit of follow-up time was converted from month to year to obtain the parameter t1.
Table 2
One-way sensitivity analysis parameter change range and probability sensitivity analysis probability distribution setting
| | Range | Distribution |
Parameters | Baseline | Upper boundary | Lower boundary | |
Cost | | | | |
Cost of local recurrence hospitalization(first year) | 10147.18 | 2971.96 | 17322.39 | Lognormal |
Cost of local recurrence outpatient(first year) | 3728.43 | 2305.50 | 5151.35 | Lognormal |
Cost of distant metastasis hospitalization per year | 10652.42 | 9233.85 | 12070.99 | Lognormal |
Cost of distant metastasis outpatient per year | 2984.55 | 2649.14 | 3319.96 | Lognormal |
Cost of hospitalization for 3 months before death | 3585.82 | 2533.61 | 4638.03 | Lognormal |
Cost of outpatient for 3 months before death | 874.56 | 713.96 | 1035.15 | Lognormal |
Cost of MAST + RECON hospitalization(first year) | 10221.30 | 9192.74 | 11249.86 | Lognormal |
Cost of MAST + RECON outpatient (first year) | 6070.98 | 5233.55 | 6908.42 | Lognormal |
Cost of BCT hospitalization(first year) | 7475.54 | 4659.50 | 10291.58 | Lognormal |
Cost of BCT outpatient (first year) | 6905.61 | 6051.87 | 7759.35 | Lognormal |
Cost of MAST hospitalization(first year) | 6630.24 | 6114.63 | 7145.85 | Lognormal |
Cost of MAST outpatient (first year) | 4210.88 | 3608.80 | 4812.97 | Lognormal |
Annual cost of follow-up for MAST + RECON | 1448.12 | 1166.11 | 1730.13 | Lognormal |
Annual cost of follow-up for BCT | 1198.82 | 961.66 | 1435.99 | Lognormal |
Annual cost of follow-up for MAST | 770.22 | 770.22 | 1090.77 | Lognormal |
Transportation cost of MAST + RECON | 253.97 | 203.17 | 304.76 | Lognormal |
Cost of losing work of MAST + RECON | 1202.54 | 962.03 | 1443.04 | Lognormal |
Transportation cost of BCT | 219.34 | 175.47 | 263.20 | Lognormal |
Cost of losing work of BCT | 997.85 | 798.28 | 1197.42 | Lognormal |
Transportation cost of MAST | 230.88 | 184.70 | 277.06 | Lognormal |
Cost of losing work of MAST | 1125.78 | 900.62 | 1350.93 | Lognormal |
Utilities | | | | |
Local recurrence (first year) | 0.779 | 0.641 | 0.917 | Beta |
Distant metastasis | 0.737 | 0.657 | 0.817 | Beta |
Disease-free for MSAT + RECON(first year) | 0.868 | 0.694 | 1 | Beta |
Disease-free for MSAT + RECON(subsequent year) | 0.933 | 0.746 | 1 | Beta |
Disease-free for BCT(first year) | 0.872 | 0.823 | 0.921 | Beta |
Disease-free for BCT(subsequent year) | 0.923 | 0.903 | 0.943 | Beta |
Disease-free for MSAT (first year) | 0.785 | 0.729 | 0.842 | Beta |
Disease-free for MSAT (subsequent year) | 0.900 | 0.883 | 0.918 | Beta |
Transition probability | | | | |
Local recurrence of MAST + RECON | 0.002299 | 0.001839 | 0.002759 | Beta |
Distant metastasis of MAST + RECON | 0.016980 | 0.013584 | 0.020376 | Beta |
Distant metastasis after local recurrence of MAST + RECON | 0.000000 | 0.000000 | 0.100000 | Beta |
Death after distant metastasis of MAST + RECON | 0.112083 | 0.089666 | 0.134500 | Beta |
Local recurrence of BCT | 0.002666 | 0.002133 | 0.003199 | Beta |
Distant metastasis of BCT | 0.006892 | 0.005514 | 0.008270 | Beta |
Distant metastasis after local recurrence of BCT | 0.201328 | 0.161062 | 0.241594 | Beta |
Death after distant metastasis of BCT | 0.016473 | 0.013178 | 0.019768 | Beta |
Local recurrence of MAST | 0.000768 | 0.000614 | 0.000922 | Beta |
Distant metastasis of MAST | 0.010451 | 0.008361 | 0.012541 | Beta |
Distant metastasis after local recurrence of MAST | 1.000000 | 0.800000 | 1.000000 | Beta |
Death after distant metastasis of MAST | 0.334939 | 0.267951 | 0.401927 | Beta |
Discount rate | 3% | 0 | 5% | Constant |
Because the annual local recurrence probability was calculated, t2 = 1 was taken. The local recurrence probability of MAST was calculated by the formula as 0.000768. Table 2 summarizes the transition probability parameters used for model input.
Cost
This research considered the direct and indirect costs from the perspective of the whole society. All costs were expressed in US dollars ($), and the exchange rate was US $1 = 6.93 yuan (January 13, 2020). Table 3 summarizes the cost data used for model inputs. The direct cost was calculatedas the direct medical costs and the patient's transportation expenses, and the indirect cost included the patient's lost time. Direct medical costs were derived from all inpatient and outpatient records of patients in the electronic medical record system and were collected according to the state Markov model.
This study also considered the first year of transportation costs for patients in different surgical treatment groups. The calculation of transportation costs was considered as the sum of the number of inpatient and outpatient visits × the average transportation cost per visit. The average transportation cost of each visit referred to the related literature published by Chengdu, China, on health economics evaluation[21]. Based on taxi fares, the transportation cost was set at 80 yuan/time.
The calculation of the cost of lost work in this study was based on the sum of the average number of days of hospitalization and the number of outpatient visits in the first year of treatment for patients in different surgical treatment groups × average daily lost time. By calculation, the loss time in the MAST + RECON group was 47 days, the loss time in the BCT group was 39 days, and the loss time in the MAST group was 44 days. According to the announcement issued by the Statistics Bureau of Sichuan Province of China, the average daily wage of employees in all units of Sichuan Province in 2018 was $9338.67/year, calculated as $25.59/day. Therefore, the lost labor cost of the MAST + RECON group was calculated to be $1202.54, the lost labor cost of the BCT group was $940.71, and the lost labor cost of the MAST group was $1125.78.
Health Utility
It was necessary to determine the health utility value of the patients of the three surgical treatment plans within one year of treatment, after the second year or more, the cases of relapsed breast cancer within one year (state R) and those of metastatic cancer (state M). The EQ-5D-5L scale was used to investigate the health utility value of 446 Chinese breast cancer patients. The health utility value of recurrent breast cancer within one year (state R) was 0.779, and the health utility value of metastatic cancer (state M) was 0.737. The health utility values of patients undergoing BCT and MAST were also obtained from the survey. Since only 3 of the 446 patients surveyed underwent MAST + RECON, the health utilities of this surgical treatment group could not be calculated. Therefore, we used the health utility mapping model established earlier in this research group to map the value of FACT-B to EQ-5D-5L to obtain the health utility of this type of patient[22]. The value of FACT-B in breast cancer patients undergoing breast reconstruction surgery was taken from the literature[23, 24], and we calculated the average value of FACT-B reported in these studies. The final health utility data used for model input are shown in Table 2.
Sensitivity Analysis
A one-way sensitivity analysis was performed to test the robustness of the economic model and the impact of the key input parameters on the results. The upper and lower 95% CI limits were used as the upper and lower limits of the parameter change, and the remaining parameters adopted ± 20% of the baseline value as the upper and lower limits of the parameter change. The discount rate used 0% and 5% as the upper and lower limits, respectively. The parameter settings are listed in Table 2.To evaluate the results of the basic cost utility, a Monte Carlo simulation was performed on the Markov model, and 1,000 samples were simulated to resolve the uncertainty strategy.