Cost-effectiveness Analysis of Breast Cancer Screening Using Mammography in Singapore: A Modeling Study

Breast cancer is the most common cancer and among the leading causes of deaths in women worldwide (1). In Singapore, the age-standardized incidence rate of female breast cancer has risen steeply from 24.6 per 100,000 person-years in 1976 to 1980 to 65.3 in 2011 to 2015 (2). Despite the rising incidence, the mortality rate has remained relatively stable since 1991 to 2015 as reported by the cancer registry annual report (2). Early detection by mammographic screening and advances in treatment are most likely causes of the stable mortality rate (3).

The national breast cancer screening program BreastScreen Singapore (BSS) was launched in 2002, aiming to reduce breast cancer mortality in Singapore. Female Singaporeans and Permanent Residents between 50 and 69 years are invited through personalized letters for biennial screening (4). Women aged 40 to 49 may attend annual screening but are recommended to consult their physician regarding the benefits and harms of screening. Physicians can discuss benefits of screening for eligible women such as better survival when finding an early stage of breast cancer, and harms such as the possibility of false positive test results, overdiagnosis, and anxiety. Eligible women receive subsidized screening mammography at government-funded polyclinics or approved private centers (5).

An essential factor to reduce breast cancer mortality through mammography screening is high attendance rates (6). In addition, attendance at the first screening round is important, because it is a predictor for attending subsequent screening rounds (7). An effective screening program relies largely on the adherence of the target population, which includes two components: the proportion of the target women that undergo screening and the regularity with which they attend the program (8).

Despite its relative affordability with Singaporean women paying only S$50 (45% of total mammogram cost) out-of-pocket, only 11% of the target population aged 50 to 69 attended screening under BSS in 2008 to 2009 (9). In a 2010 national health survey, 40.4% of women aged 50 to 59 and 38.1% of women aged 60 to 69 reported attending mammogram screening in the past two years, which include screening outside the BSS screening program (10). To increase screening uptake, the Singapore Government extended Medisave, a compulsory medical savings scheme, in 2011 to offset the cost of screening mammograms (9).

Currently, the breast cancer screening recommendations are predominantly based on research from Western countries, but these may be less appropriate in Asian settings due to cultural and biological differences. For instance, well-established risk factors for breast cancer such as alcohol, smoking, low intake of fresh fruit and vegetables are in general less prevalent among Asian women than among Western women (11–13). Biological differences between Asian and Western women such as density of the breast could influence the detection of breast cancer when using mammography (14, 15). Furthermore, other potential differences between Western and Asian women could be their perception on breast cancer screening in terms of quality of life as well as the differences in health care context in terms of available monetary resources and human resources (16).

Although many cost-effectiveness analyses have been conducted for breast cancer screening (17), no study has performed an extensive cost-effectiveness analysis of the Singaporean national breast cancer screening program. Therefore, the aim of this study was to assess the cost-effectiveness of the current screening policy and alternative breast cancer screening scenarios in Singapore with different attendance levels.

For this study, we used the MIcrosimulation SCreening ANalysis-Fatal diameter (MISCAN-Fadia) breast model. MISCAN-Fadia model has been used to inform policies and gain insights from clinical trials. For instance, in the United States, a variety of screening strategies have been simulated and evaluated to determine the optimal screening strategy (14). Another example is that MISCAN-Fadia has been used to simulate the UK Breast Screening Frequency trial in order to predict breast cancer mortality using a longer follow-up period, that is, lifetime period (18, 19).

The MISCAN-Fadia model varies the breast tumor growth and fatal diameter among simulated women (20, 21). This unique biological component allows each simulated woman to have a different tumor size at which treatment is no longer effective (i.e., fatal). Furthermore, the model simulates two identical populations: one with and one without screening. By comparing the life histories in the presence of screening with corresponding life histories in the absence of screening, we can estimate harms and benefits of screening.

Women born between 1910 and 1969 were simulated by 5-year birth cohorts (e.g., women born in 1910–1914 cohort were in the year 1971: age 57–61), which takes into account both the national screening program as well as the period prior the screening program. In total, 10 million women were simulated and followed until death. Age-specific all-cause mortalities by 5-year birth cohort were derived by calibrating to observed mortality rates and birth cohort–specific expected life expectancies at age 65, obtained from the Singapore Department of Statistics (Supplementary Table S1).

Incidence data on invasive breast cancers (Fig. 1) and breast cancer deaths from 1971 to 2014 were collected by 5- and 10-year age groups, respectively from the National Registry of Diseases Office (NRDO). Mode of detection (screen- or clinically detected) was included where possible. American Joint Committee on Cancer (AJCC) stage groupings were available from 2003 onwards, while screen detection information was available from 2008 onwards.

In addition, treatment dissemination data were obtained from the NRDO. Percentage of breast cancer cases who received chemotherapy, endocrine therapy, and both treatments combined were retrieved and categorized by year of diagnosis, 5- or 10-year age groups and AJCC staging. Screening attendance data were derived from the NRDO and BSS by 5-year age groups and calendar year. Singaporean data on attendance, specificity (Supplementary Table S1), and treatment were used to model the effect of screening on breast cancer survival. Estimated survival and mortality were compared with observed data.

As local data on phase- and stage-specific health state utilities for breast cancer were unavailable, we adapted values from a study conducted in the Korean general population (22). We assume that the setting of Korea is more similar to Singapore in terms of economic wealth and culture than western countries. Utilities for screening and the diagnostic phase were obtained from de Haes (23). Utility values are shown in Table 1 (see Supplementary Table S1 for calculations).

Singapore's national breast cancer screening program recommends annual screening for women between ages 40 and 49 (upon doctor's advice) and biennial screening for women between ages 50 and 69. In total, 302 scenarios were simulated of which 1 scenario represented Singapore's current screening program at observed attendance levels varying between 3% and 14% across different calendar years and age groups, 1 scenario assumed no screening and 300 scenarios varied by age ranges, screening interval, and attendance levels (Table 2). Starting ages for breast cancer screening were 40, 45, 50, or 55 and stopping ages were 64, 69, 74, or 79. Of 300 scenarios, 240 scenarios included different screening frequencies: annually, biennially, and triennially. The remaining 60 scenarios were hybrid scenarios that allowed women to be screened annually until the switching age, after which they were screened biennially. Switching ages considered in this study were ages 45 and 50. Furthermore, attendance levels of 10%, 30%, 50%, 70%, and 100% were considered for each scenario.

This analysis considered the health care provider perspective. Screening program costs such as screening invitations and mammograms were obtained from BSS. Diagnostic work-up costs (imaging, laboratory investigations, and biopsies) were obtained from the National University Hospital Patient Affordability Simulation System (NUH PASS). Treatment costs were provided by the Ministry of Health and were aggregated by AJCC stages I–IV and phases of care. The phases of care include initial care: the first 12 months after diagnosis of breast cancer, terminal care: the last 6 months prior to death, and continuing care: the period between initial and terminal care (Table 1; ref. 24). Costs were expressed in Singapore dollars using 2010 as a base year, when consumer price index (CPI) = 100 (25). All costs were converted to the year 2002 using the CPI for health care (25), as the breast screening program started in 2002. Both costs and health outcomes were discounted with 3%, in line with recommendations by the Agency for Care Effectiveness Singapore (26). The willingness-to-pay (WTP) threshold was set at one gross domestic product (GDP) per capita of approximately S$70,000/QALY (US$50,000) gained for the purpose of our study. It should be noted that Singapore does not have an explicit WTP threshold.

For each scenario, we calculated the discounted costs and quality-adjusted life years (QALY), average cost-effectiveness ratio (ACER; ratio of differences in costs to differences in QALYs compared with no screening), incremental cost-effectiveness ratio (ICER), and number of mammograms. Furthermore, benefits of screening (e.g., number of breast cancer deaths averted, reduction in breast cancer mortality, and number of life-years gained) were compared against the situation without screening. The harms of screening include number of false positive, overdiagnosis for ductal carcinoma in situ (DCIS), and invasive breast cancers (IBC) combined. The latter was calculated by dividing the number of screen-detected overdiagnosed cases by the total number of screen-detected DCIS and IBC.

We constructed 5 cost-effectiveness planes by attendance level, with efficiency frontiers. The efficiency frontier plot depicts scenarios based on the additional costs and QALYs gained, which can be found on the x-axis and y-axis, respectively. When establishing the efficiency frontier for each attendance level, Singapore's current program at its observed attendance level was excluded. Scenarios on the efficiency frontiers were cost-efficient while those below the efficiency frontier were dominated. ICERs were computed for each scenario on the efficiency frontier as the incremental net costs per incremental QALY gained compared with the previous efficient scenario. Lifetime time horizons for costs and effects of screening were considered for each simulated person. A subanalysis assessed whether the current screening program would be on a frontier across all attendance levels. All outcomes were reported per 100,000 women alive in 2002.

One-way sensitivity analyses were performed to estimate the impact of a lower/higher work-up costs (Table 1) as well as the effect of a continuous care of 5 years on the efficiency frontier.

At observed attendance levels of 3% to 14%, which vary by age groups and calendar year, Singapore's current screening program yielded 1,054 life-years gained and averted 57 breast cancer deaths, while requiring 54,158 mammograms and resulting in 2,101 false positives findings (Table 3). Furthermore, we estimated an increase of 312 QALYs at an additional cost of S$ 3,799,641 (both discounted) compared with no screening. When assuming attendance levels ≥50% for the current screening program, it was near the efficiency frontier (Fig. 2C–E). Increasing the current screening program's attendance level to 50% resulted in an addition of 659,934 mammograms; 10,030 LYG; 479 breast cancer deaths averted and 25,962 false positives when compared with observed attendance level (Table 3). Compared with no screening, the current screening program reduced breast cancer mortality by 3%, 11%, and 19% at attendance levels of 10%, 50%, and 100%, respectively.

Regardless of attendance level (Fig. 2A–E), the majority of screening scenarios on the cost-efficiency frontiers start at ages 40 and 45. At attendance levels <70%, few scenarios on the cost-efficiency frontiers had stopping ages >69.

Few hybrid scenarios were on the cost-efficiency frontiers (Fig. 2A, B, and E). Annual screening scenarios were most common on the cost-efficiency frontier at attendance level ≤70%, while more biennial screening scenarios tend to be on the efficiency frontier when attendance levels increased (Fig. 2D and E). Triennial screening was generally inefficient, with 1 to 2 screening scenarios on each cost-efficiency frontier (Fig. 2A–E).

Scenarios with lowest costs were biennial and triennial screening intervals while scenarios with annual screening interval generally incurred higher costs. More intense screening resulted in higher mortality reductions and more life-years gained. However, it also increased false positives and overdiagnosis rates, which becomes more pronounced at higher attendance rates (more details in Supplementary Table S2).

Assuming a threshold of S$70,000/QALY gained, all scenarios on the frontier were cost-effective for all attendance levels (Table 4). The highest ICER was S$56,306/QALY gained for screening annually from ages 40 to 79, while the lowest ICER was S$10,186/QALY gained for screening triennially from ages 50 to 64, both assuming 10% attendance. The latter is also the least intensive screening scenario with 44,075 mammograms. Compared with no screening, this screening scenario averted 46 breast cancer deaths, reducing breast cancer mortality by 1% and yielded 944 LYG; 1,716 false positives, 39 and 3 overdiagnosis for DCIS and IBC, respectively (Supplementary Table S2).

Varying the work-up costs did not show any major differences in the efficiency frontiers by attendance level (see Supplementary Figs. S1 and S2 and Supplementary Tables S3 and S4). When reducing the continuous care period to 5 years, more scenarios appear near the efficiency frontier compared with the conservative approach in which the continuous care phase includes the period between the initial care and terminal care (Supplementary Fig. S3; Supplementary Table S5).

Our study showed that breast cancer screening in Singapore is cost-efficient when starting screening at ages 40 or 45 for all women. Singapore's current breast cancer screening policy recommends annual screening between ages 40 and 49 and biennial screening between ages 50 and 69 (4). Women age 40 and 49 are recommended to first consult their physician to assess the necessity of screening at these ages. We found that Singapore's current screening recommendation was near the efficiency frontier at attendance levels ≥50%. The current screening policy appears to strike a good balance among cost, benefits (mortality reduction and life-years gained), and harms (false positives and overdiagnosis) given an attendance level ≥50%. Policy makers should therefore focus their efforts on increasing the participation rates of breast cancer screening up to at least 50% (27).

Surprisingly, screening biennially from 50 to 69 was not on the efficiency frontiers. This was striking as this has been recommended by many countries, especially in Europe (28). In Singapore, almost a third of breast cancers were found in women aged 45 to 54 (2). Therefore, starting screening from age 50 may be less effective in Asian countries as a significant number of women with breast cancer are not picked up in a timely manner. This was also observed in our study, as several screening scenarios that involved screening starting age 40 were on the efficiency frontier. However, given the adoption of western life-style patterns in different Asian countries, a shift in incidence could occur over time. This implies that a breast cancer screening program should be periodically reevaluated to optimize the screening ages for future cohorts. Furthermore, some consideration should be made on lowering the screening age (29).

Our study suggests that starting at age 40 will be cost-efficient as several of these scenarios were on the efficiency frontier. However, important to note is that results in this study are conservative estimates as the continuous phase of care includes the entire period between the initial phase of care and terminal phase of care. Our sensitivity analysis results show that when the continuous phase of care is set on 5 years, more scenarios approach the efficiency frontier.

Our analyses are useful to non-Singaporean readers. For example, governments with limited budgets may opt for triennial screening as it has the lowest financial impact. Our study found that triennial screening between ages 50 and 64 has the lowest ICERs varying between S$10,186 and S$10,713. Although ICERs varied little across attendance levels, at 100% attendance, 426 deaths could be averted compared with 46 at a 10% attendance level. Nonetheless, it is important to note that a higher attendance level also increases false positives and overdiagnosis rates. It is therefore essential to find the optimal balance between benefits and harms.

The major strengths of this study were the inclusion of country specific model inputs, such as incidence data, official statistics of the breast cancer screening program, and the extensive range of screening scenarios. This allowed the model to estimate the effects of introducing breast cancer screening in Singapore. However, our research had some limitations. First, a distinguishing characteristic of Singapore is that it has 3 major ethnicities. Yet, we were unable to distinguish effects across different ethnic groups in Singapore as screening rates are very low among Malay women. Future research is required to investigate differences between ethnic groups. Second, registration of data has improved over time, resulting in good quality data after 1995 but less precise for earlier years. Hence, this has created some challenges when calibrating the model. Nonetheless, we were able to achieve a good model fit. Third, generalizations of results in this study to other settings should be done with caution as each country has its own characteristics. Fourth, we have assumed a WTP threshold of S$70,000/QALY gained, but there is no single official WTP threshold for Singapore. Furthermore, this WTP threshold is based on GDP, a demand side approach that has been suggested to be inadequate (30). An alternative approach based on the opportunity cost of health forgone has been proposed. However, such a supply side WTP threshold has not been estimated for Singapore. Hence, the conclusions of our article may change, depending on the actual WTP threshold used by policy makers (31). Fifth, we did not take into account the impact on financial and human resources that are needed to increase the screening attendance among eligible women. Although there are various interventions available to promote screening among women (27). It is important to take context specific factors into account such as cultural belief and attitudes toward screening (32–34). Future studies should focus on the required human resources and financial resources when using context-specific interventions to increase the attendance levels in breast cancer screening. Finally, we did not account for the impact of ageing on quality of life in the QALY estimations. However, for our analyses we have used utilities that were averages across all ages. Accounting for the impact of ageing, would have a stronger effect on the quality of life (QOL) for higher age groups (>70 years) than younger age groups as the QOL would decrease relatively more among older ages than younger ages. In our study, we found that the majority of the scenarios that were cost-effective included scenarios with relatively young stopping ages for screening. We would not expect a change in the ranking of the scenarios on the ICER but a slightly lower ICER. Therefore, adjusting for aging would have a relatively small effect on the results and conclusion. In conclusion, screening all women at age 40 or 45 is cost-efficient in Singapore and a policy change may be considered. At the current screening program attendance of 10%, annual screening will be more likely than biennial or triennial screening to be cost-efficient but with a tradeoff of higher false positives and overdiagnosis. It is thus important to increase the attendance level to gain more benefits in terms of life years and mortality reduction while retaining cost-effectiveness. In the future, focusing on risk stratification and tailoring breast cancer screening policies could enhance the benefits and limit the harms of screening.

S. Chootipongchaivat reports grants from Singapore Ministry of Health, Health Services Research Competitive Research Grant, administered by the National Medical Research Council, and grants from NIH/NCI during the conduct of the study. K. ten Haaf reports grants from Singapore Ministry of Health, Health Services Research Competitive Research Grant, administered by the National Medical Research Council, during the conduct of the study, as well as grants from European Union (Horizon 2020), grants from University of Zurich, Switzerland, nonfinancial support from International Association for the Study of Lung Cancer, grants from Cancer Research UK, nonfinancial support from Russian Society of Clinical Oncology, nonfinancial support and other from BIOMEDICAL RESEARCH IN ENDSTAGE AND OBSTRUCTIVE LUNG DISEASE HANNOVER (BREATH), and grants from NIH/NCI outside the submitted work. N.T. van Ravesteyn reports grants from Singapore Ministry of Health and NIH/NCI during the conduct of the study. No disclosures were reported by the other authors.

S. Chootipongchaivat: Conceptualization, software, formal analysis, validation, visualization, methodology, writing–original draft, project administration, writing–review and editing. X.Y. Wong: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, project administration, writing–review and editing. K. ten Haaf: Conceptualization, formal analysis, methodology, writing–review and editing. M. Hartman: Resources, data curation, funding acquisition, writing–review and editing. K.B. Tan: Resources, data curation, writing–review and editing. N.T. van Ravesteyn: Conceptualization, supervision, funding acquisition, validation, methodology, writing–review and editing. H.-L. Wee: Conceptualization, resources, formal analysis, supervision, funding acquisition, validation, investigation, methodology, project administration, writing–review and editing.

We thank the staff at the National Registry of Disease Office (Singapore) and BreastScreen Singapore for granting access to the datasets, the Breast Cancer Prevention Programme for providing expert opinion and localized data, and the Saw Swee Hock School of Public Health for granting access to the NUH PASS database. Finally, we thank A/P Joanne Ngeow for providing valuable comments on this study. This publication was made possible by the Singapore Ministry of Health, Health Services Research Competitive Research Grant, administered by the National Medical Research Council (grant number HSRG/13MAY006). Support was also provided from the NCI as part of the Cancer Intervention and Surveillance Modeling Network (CISNET), which supported the underlying development of the simulation model utilized (grant number U01 CA199218). The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. M. Hartman was supported by the National Medical Research Council Clinician Scientist Award (Senior Investigator Category, NMRC/CSA-SI/0015/2017), the National University Cancer Institute Singapore Centre Grant Programme (CGAug16M005), the Saw Swee Hock School of Public Health Programme of Research Seed Funding (SSHSPHRes-Prog), and the Asian Breast Cancer Research Fund (N-176-000–023-091).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.