Global, Regional, and National Burden of Breast Cancer and Its Attributable Risk Factors Among Women, 1990–2017

Introduction: Breast cancer is the most common cancer in women worldwide. However, no comprehensive study has been conducted to compare the incidence, mortality, and disability-adjusted life years (DALYs) for female breast cancer among different countries. The current study examined the level and trends of incidence, death, and DALYs for breast cancer and its attributable risk factors among women in 195 countries from 1990 to 2017 by age, socio-demographic index (SDI; a composite of socio-demographic factors), and healthcare access and quality (HAQ; an indicator of health system performance) index. Methods: Vital registration, verbal autopsy, and cancer registries were used across the globe to generate estimates. Incidence, mortality, and DALYs were estimated. All estimates are presented as counts and age-standardised rates per 100,000 person-years. Results: Between 1990 and 2017 the global incidence of breast cancer increased signicantly by 17.1% (95% uncertainty interval [UI]: 7.1–23.6; with 1.9 million incidences in 2017 [95% UI 1.9–2.0]; age-standardised rate of 45.9 [95% UI: 44.2–47.4]). However, over this same period the age-standardised death rate signicantly decreased by 10.6% (95% UI: -19.5 to -4.4), with 600.7 thousand deaths in 2017 (95% UI: 578.7–630.0; age-standardised death rate of 14.1 95% UI: 13.6–14.8). Global DALYs also signicantly decreased between 1990 and 2017 by 9.3% (95% UI: -19.9 to -1.6) with 17.4 million DALYs reported in 2017 (95% UI: 16.6–18.4; age-standardised rate of 414.7; 95% UI: 395.5– 437.6). Lebanon [138.3 (95% UI: 106.5–170.7)], the Netherlands [109.8 (95% UI: 97.4–122.7)], and the UK [102.6 (95% UI: 99.6–105.8)] had the three highest age-standardised


Mortality estimation
As there is less cancer mortality than incidence data available globally, mortality-to-incidence ratios (MIR) were used to transform incidence data to mortality estimates. These ratios were modeled by linear-step mixed effect models in locations where both incidence and death data existed for the same year and were adjusted for age, sex, and HAQ. Spatiotemporal Gaussian process regression (ST-GPR) was used to smooth estimates across space and time [15]. Mortality estimates were calculated by multiplication of 5- year age group, sex, and location-speci c MIRs and incidence estimates. These mortality estimates as well as direct mortality data from VR and VA sources were included in the Cause of Death Ensemble Model (CODEm) [15]. This model uses various individual models to provide a single model with the best t using all available data and covariates. Covariates used in CODEm are listed in Appendix Table 2.
Incidence, prevalence, and disability estimation

Risk factors
Risk factors that had evidence of causation with breast cancer were selected in the present study [16].
The percentage of breast cancer DALYs attributable to alcohol consumption, high fasting plasma glucose, high body mass index, secondhand smoke, smoking, and low physical activity are reported.
De nitions for these risk factors and their relative risk for breast cancer can be found elsewhere [16].

Global level
Breast cancer among women accounted for 1.9 million (95% UI:  Table 3). The number of incident cases increased from 870,183 in 1990 to 1,937,574 in 2017 and east Asia, western Europe, and high-income North America were the regions with the highest number of incident cases between 1990 and 2017 ( Fig. 2A, Table 1 Table 3).

National level
The age-standardised incidence rate of breast cancer in 2017 ranged from 16

Burden of breast cancer by SDI and HAQ
The current study identi ed a non-linear association between age-standardised DALY rate of breast cancer for GBD regions and SDI between 1990 and 2017. The global burden of breast cancer was higher than expected in the initial years of measurement but decreased to lower than expected levels after 2003.
High-income North America, western Europe, Australasia, the Caribbean, southeast Asia, Oceania, and western sub-Saharan Africa had a higher than expected disease burden based on their SDI level throughout the estimated times series. Whereas central Europe, eastern Europe and central Asia, southern Latin America, Tropical Latin America, and central Asia had lower than expected levels in recent years (Fig. 5A). The association between age-standardised DALY rate and SDI in 2017 was also examined for 195 countries and non-linear associations were shown between two variables. There were countries with various level of SDI that had higher than expected levels of breast cancer burden, such as the Bahamas, Pakistan, Tonga, Nigeria, Lebanon, and Fiji. In contrast, some countries such as Mongolia, Guatemala, China, Maldives, Kuwait, and South Korea had much lower than expected breast cancer burden (Fig. 5B).
Interestingly, non-linear and negative associations were found between age-standardised DALY rate and HAQ for 195 countries. There were many countries with higher than expected disease burden based on their HAQ, such as the Bahamas, Pakistan, Tonga, Nigeria, Lebanon, and Fiji, reporting the highest difference with their expected levels. In contrast, some countries such as Mongolia, Guatemala, China, Maldives, Kuwait, and South Korea had a much lower than expected level of breast cancer burden (Appendix Fig. 2).

Risk factors
Although the percentage of attributable breast cancer DALYs for risk factors were different in the GBD regions, alcohol consumption (9.2% [95% UI: 7.7-10.7]), high fasting plasma glucose (6.1% [95% UI: 1.1-13.6]), and high body mass index (4.5% [95% UI: 1.4-8.5]) were the top three contributors to percent of breast cancer DALYs globally (Fig. 6). Alcohol consumption had the greatest attributable burden in most of the GBD regions but high body mass index and high fasting plasma glucose were found to have higher attributable burden than alcohol consumption in Southeast Asia, East Asia and Oceania (Fig. 6) Fig. 3).

Discussion
The current study reports the incidence, deaths,  [18]. Although the age-standardised rates reported in GLOBOCAN and GBD could not be compared due to differences in the standard population, the ranking of countries could be compared.
Belgium, Luxembourg, and the Netherlands had the highest age-standardised incidence rates of breast cancer in GLOBOCAN 2018; while, in 2017, Lebanon, the Netherlands, and the UK were found to have the highest age-standardised incidence rate in our study. Age-standardised death rates were highest in Fiji, Barbados, and Somalia in GLOBOCAN 2018, while our 2017 data found Pakistan, Tonga, and the Bahamas to have the highest age-standardised death rates [18]. These differences may be attributed to different data sources and estimating methodology. However, other studies have also reported differences in breast cancer incidence and death rates for selected countries over time, across different iterations of the GLOBOCAN project, which are not comparable with our ndings as their estimates are up to 2012 [4, 7, 9-12, 19, 20]. An important point is that both GLOBOCAN and GBD studies indicate that incidence rates of breast cancer were highest in developed countries, while death rates were highest in some developing countries and some of these differences between the developed and developing countries could be explained by different prevalence of aetiological factors (i.e. alcohol consumption and high body-mass index), as well as the variations in cancer prevention, screening practice, advances in diagnosis, treatment, and oncology care, and management of health resources [12].
The trend of age-standardised incidence rate is also reported in some studies across the selected countries. A study reported increasing trends of age-standardised incidence rate for breast cancer in Australia, Ireland, Japan, the Netherlands, England, Sweden, and the USA during 1980-2009 [7]. While our study found that all of the aforementioned countries had increasing trends in age-standardised incidence rate during 1990-2017, except the USA (-13% [95% UI: -17 to -9]) and the UK (-5% [95% UI: -8 to -1]), which showed signi cant decreasing trends in the measurement period. Another study [11] examined the trend in age-standardised incidence and death rate for some selected countries with various time intervals, mainly during 1980-2010. They reported that age-standardised incidence rate increased for the USA, Denmark, Italy, Australia, Russia, Costa Rica, and China. Again, our study identi ed somewhat con icting results nding that age-standardised incidence rates increased for all of the aforementioned countries, with the exception of the USA (-13% [95% UI: -17 to -9]) between 1990 and 2017. They also reported that age-standardised death rates are decreasing for Denmark, Italy, the USA, and Australia. In contrast, Japan, Russia, the Philippines, and China experienced an increasing trend in their study period (mainly 1980-2010) [11]. The decline in breast cancer mortality among women in developed countries may be explained by a number of reasons, most notably the introduction of a screening program (mammography) and improvement in therapy [21]. Besides a considerable decrease in the mortality of breast cancer in the world during the period 1990-2017, global incidence increased over the period of study. The signi cant differences in breast cancer burden observed between countries could be explained by different applied methodology and the varying data quality worldwide. Also, the considerable variations in breast cancer burden can be attributed to differences in prevalence of risk factors, differences in cancer prevention and treatment, and management of health resources. Therefore, it is imperative that policy makers consult multiple data sources and consider results that provide the most representative and comprehensive information available for their country. Global-or regional-level estimates of breast cancer burden may be misleading for national policy, as there are some countries that follow completely different patterns compared to their regional estimates. When making public health policy decisions, individual countries should consider trends in breast cancer burden as well as their expected levels of burden as part of informed judgments. Modi able risk factors such as alcohol use [22], high fasting plasma glucose [23], high body mass index [24], smoking [25], second hand smoke exposure [26], and low physical activity [27], are all reported to be associated with breast cancer burden. Breast cancer incidence and mortality increase with age and have a distinctive age-speci c curve, with the age group 50-54 dividing the trends of mortality into premenopausal and postmenopausal periods. According to our results, this observation could be related to, besides the hormonal milieu, exposure to all of the described risk factors which generally increased after the age of 50. Therefore, country-level health professionals could be promoting community awareness regarding prevention programs and associated risk factors through using appropriate health education plans to decrease the breast cancer burden as much as possible. Data from this study can be used to determine trends over time, benchmarking with other similar countries and regions. In addition, this evidence can be used by policy makers to determine factors contributing to country-speci c differences within regions and subsequently inform a matrix of policy options that will best meet the economic, public health, political, and societal needs of the country.
Associations between the socioeconomic development level of each country and breast cancer incidence/death rate has been studied previously, in which a positive association between breast cancer incidence and a country's development level in 2012 was reported [28]. These results should be interpreted with caution as income alone was used as the determining factor, which on its own is an inadequate indicator of a country's level of development. In our study we used SDI to examine socioeconomic development, which more comprehensively re ects socioeconomic development because it is a composite indicator of education, income, and fertility rate. In addition to SDI, we also report breast cancer burden by HAQ, which shows the burden of breast cancer may be decreasing with successive increments of HAQ. This suggests that publicly funded interventions such as breast cancer screening and appropriate treatment services can decrease breast cancer burden. The previous studies indicate that all of the observed reductions could not be attributable to screening alone [29][30][31] and it need to be considered along with education, treatment and care. For example, a study in Norway found that screening itself accounted for only about a fourth of the total reduction [32]. However, some studies reported larger contribution of screening measures, up to 60%, in decreasing the breast cancer mortality rates [33,34]. The national-level analysis found that although SDI of countries may not be associated with breast cancer age-standardised DALY rate, it is negatively associated with HAQ. Therefore, this approach allowed us to compare the observed breast cancer burden with corresponding expected levels based on SDI (in regional-and country-level) and HAQ (in country-level) for the rst time across the globe.
The present study is the most comprehensive and up-to-date study to examine level and trends of incidence, mortality, and DALYs associated with breast cancer during the last three decades globally. As with all research, this study had a number of limitations, including the fact that some variations in incidence and mortality may be due to detection biases, such as changes in screening protocols. GBD accounts for ascertainment bias by adjusting single cause estimates to the all-cause mortality envelope.
There was data sparsity in some countries such as Iraq and Afghanistan and their estimations were subsequently conducted based on predictive covariates and neighboring locations and should be interpreted with caution. Moreover, HAQ was not available for 2017 and as such we had to examine the association between HAQ and corresponding age-standardised DALY rate for 2016. Finally, there are several other and stronger risk factors for breast cancer, e.g. age at menarche, age at rst pregnancy, breast feeding history, use of contraceptives and low number of pregnancies that were not considered in the paper because their associations with breast cancer risk are not fully established.

Conclusion
Remarkable inter-country variation was observed in the burden of breast cancer. Globally, the agestandardised rates of death and DALYs due to breast cancer is decreasing, while the corresponding agestandardised incidence rate is increasing. Despite clear global patterns, national-level data does not always t the same regional or global trends, particularly among countries such as Mauritius, the Philippines, and the Dominican Republic that showed remarkable increasing trends in both agestandardised incidence and death rates. Prevention measures should be planned based on national-level estimates in every country and strengthened through early detection and treatment and increasing population awareness regarding reducing exposure to modi able risk factors, especially in countries with high level or increasing trends in disease burden. A broad population-wide approach to breast cancer indicates a contribution of exposures to selected risk factors, while leaving space for future examinations of the contribution of genetic and some other lifestyle factors, as well as the diagnosis and treatment patterns to the differences on regional and country level.       Supplementary Files