We aimed to estimate costs for breast cancer treatment due to suboptimal breastfeeding from a healthcare provider or a government perspective in Hong Kong. We only focused on direct medical cost to the government so as to inform investment of healthcare resources. As such our evaluation did not consider individual costs and societal costs, such as loss of productivity due to absence from work.

An individual-based Monte Carlo method (Fig. 1) adopted from previous studies [6, 10] was constructed to simulate the development of breast cancer over a woman’s lifetime in a hypothetical birth cohort aged 20 years in 2018 (n = 33500), with which we predicted the cases and deaths of breast cancer in the base case (with actual breastfeeding rate of 26% in 2018) and two hypothetical optimal scenarios (scenario 1: 90% parous women exclusively/partially breastfeed for at least 12 months or scenario 2: 90% parous women exclusively breastfeed for six months). Monte Carlo methods are a computational technique that uses repeated random sampling of input parameters to predict probability of outcomes. In this present economic evaluation, the outcome was incidence of breast cancer and the input parameters for simulation were best available data mainly from the government statistics. (Table 1) The fertility rates of Hong Kong women and mortality rates were obtained from the Hong Kong Census and Statistics Department. Rates of exclusive and partial breastfeeding were obtained from biennial breastfeeding surveys in Hong Kong conducted by the Department of Health. Incidence rates of different breast cancer stages and surviving rates of each stage of breast cancer were obtained from Hong Kong Cancer Registry operated by the Hospital Authority. Medical costs for treating breast cancer in Hong Kong were obtained from a local cost study [11]. Disability weights were obtained from Global Burden of Disease Study 2017 [12]. We adopted the relative risks of breast cancer by duration of any (exclusive or partial) breastfeeding and exclusive breastfeeding summarised in a recent meta-analysis on 65 studies [13] as local studies on the associations between breastfeeding and breast cancer are lacking.

Adapted from: Bartick et al., Cost analysis of maternal disease associated with suboptimal breastfeeding. Obstet Gynecol. 2013;122(1):111–119. https://doi.org/10.1097/AOG.0b013e318297a047 Bartick et al., Suboptimal breastfeeding in the United States: Maternal and pediatric health outcomes and costs Matern Child Nutr. 2017;13(1):e12366. https://doi.org/10.1111/mcn.12366

Diagram of simulation model.

In base case and each scenario, we simulated a cohort of Hong Kong women aged 20 years old in 2018. The proportion of parous women in each year was simulated according to 20 to 49 year-old age-specific fertility rates in Hong Kong in 2018 in which total number of births was 53,700. The maximum parity was set at two because average fertility rate was as low as 1.1–1.3 during 2010–2019. For each simulated woman, the risk of developing breast cancer in each year from age 20 to 80 years was simulated based on her age, cumulative lifetime breastfeeding history and age-specific incidence rates of breast cancer in 2018. The case-fatality was simulated based on the age, stage at diagnosis and ten-year survival rates extrapolated from the stage-specific relative 1-to-5-year survival rates from the Hong Kong Cancer Registry, but not their breastfeeding history. The mortality of women without breast cancer was simulated using the age-specific death rates among females in Hong Kong. We assumed steady-state rates of disease incidence, disease survival, fertility, and the cost of treatment. We did not consider transitions between stages as early diagnosed breast cancers are mainly treated and we set out to provide a conservative estimate. Deaths for women surviving beyond 10 years from diagnosis were considered unrelated to breast cancer.

We calculated the treatment cost and DALYs associated with breast cancer in each scenario of breastfeeding rate using stage-specific aggregated one-time treatment costs reported in a local study [11] and DALYs information from the Global Burden of Disease Study [12], at an annual discount rate of 3%. Costs were converted to US dollars based on the exchange rate in 2018 (1 USD = 7.8 HKD). The terminal care cost would be applied when the death due to breast cancer occurred in the simulation. We assumed all the diagnosed breast cancer cases would receive treatment. We also assumed treatment cost in public sector, which is heavily subsidised by the government is the same as private hospitals, where more expensive treatment options are maybe available.

For DALYs calculation, the number of years lived with disability (YLD) and the number of years of life lost (YLL) were deduced with the formula in the DALY calculator for R. The information to calculate YLL and YLD, including age of onset and duration of disease, were derived from the simulation of the disease outcome for each woman. Survivors beyond ten years were considered cured [12]. Two sequelae (diagnosis and primary therapy, and controlled phase) were assumed for cases that were cured, and four sequelae (diagnosis and primary therapy; controlled phase; metastatic phase; terminal phase) were assumed for those that did not survive beyond ten years. Age-weighting was not applied.

The simulated outcomes were based on 500 iterations each with parameters including relative risks, age-specific incidence of breast cancer, aggregated treatment costs, and disability weights randomly generated from specified distributions. (Table 1) The results from probabilistic sensitivity analysis were validated by comparing deduced incidence rates from the model for the base case with the actual rates. We also carried out deterministic sensitivity analysis to assess the main cost drivers (by changing one parameter at a time) and the most/least cost-saving simulations (by changing parameters at the same time to achieve most/least cost-saving scenario).

Programming was performed using Python and R statistical software version 4.1.0 (Vienna, Austria; R Core Team, 2021).