Abstract

Abstract

Female breast cancer has become a prominent public health issue across the world. Its epidemiological features, including incidence and mortality, present with geographic variations. A comprehensive understanding of the local epidemic could provide essential references for formulating region-specific prevention and control strategies for breast cancer. However, there was little research towards the epidemiological characteristics of female breast cancer in Macau. In face with the clipping burden of female breast cancer, lots of countries or regions have carried out breast cancer screening programs for decades. From the perspective of cost-effectiveness, the risk-stratified screening program has been the hotspot in the academia and policy research. However, there is still no any female breast cancer screening program in Macau, which is of great urgency to initiate as soon as possible. Notably, the risk prediction model, providing the ground for classifying participants into various risk groups, is the core item for the successful implementation of the risk-stratified screening program. Therefore, this study aimed to investigate the incidence and mortality characteristics to learn the burden of female breast cancer in Macau, including their status quo, temporal trends and projections. As regards to breast cancer screening, this study aimed to analyzed the distribution features of mammographic results of local females and then developed a risk prediction tool for identifying high-risk population. Multiple approaches have been adopted in our analysis. Firstly, the annual incidence and mortality data of breast cancer from 2003 to 2012 was retrieved from Macao Health Bureau. Descriptive, Join-point regression, and autoregressive integrated moving average (ARIMA) were applied to explore the status quo, temporal trends, and projections of breast cancer in Macau. Secondly, the mammography records at the individual level were collected from Macau Health Bureau from August 2018 to July 2023. Descriptive and chi-square tests were used to present and compare the differences among different mammography results according to the Breast Imaging Reporting and Data System (BI-RADS). Thirdly, in order to develop a risk prediction model, the original dataset was randomly separated into training set (60%) and testing set (39%). Synthetic minority over-sampling technique (SMOTE) and random undersampling (RUS) were applied to deal with the unbalanced issue in the training set. Logistic regression, decision tree and random forest was used to constructed the risk prediction model on the 2 sets of training data. The discrimination and several evaluation metrics of these models on the same testing set was calculated, compared and selected for identifying high-risk population of breast cancer in Macau. In 2021, breast cancer was the most common and the third fatal cancer type among females in Macau. The crude incidence presented with increasing tendency from 2003 to 2021 (average annual percentage change (AAPC): 3.1%), especially since the year 2016 (annual percentage change (APC): 11.01%). Besides, age groups of 50- 69 and >70 demonstrated growing trends in the incidence rate with the AAPC values of 2.4% and 2.6%, respectively. The crude mortality rate remained a relatively stable increasing tendency (AAPC=1.05%). The crude incidence and mortality rates were estimated to keep rising in the following three years. Overall, 12,091 participants with available mammography records were included in our analysis. The proportions of BI-RADS 1-6 were 52.41%, 37.76%, 8.97%, 0.03%, 0.79% and 0.04%, respectively. The distribution of mammographic results was significantly different by age group, age at menarche, age of first live birth, breast implant, breast biopsy, nipple discharge and breast lumps. For the construction of the risk prediction model, BI-RADS 1 and 2 were classified as normal mammography, while others as abnormal mammography. Age, age at menarche, age of first live birth, breast biopsy, and breast lump were identified as significant predictors. For the testing set, logistic regression together with SMOTE had the best discrimination with an AUC value of 0.7270 (95% CI: 0.6982-0.7559), as well as a relatively higher value of balanced accuracy of 0.6964. This model was selected as the final risk prediction model, and the threshold was 0.342 for the high-risk on the basis of the Youden index. A nomogram was designed to visualized the coefficients of this model. To conclude, female breast cancer was a serious health issue in Macau, whose burden would possibly become even greater in the following years because of the lasting growth in the incidence and mortality rates. The mammography results significantly differed by age, reproductive and clinical factors. Supported by current data, age, age of menarche, age of first live birth, breast biopsy and breast lump could be used to evaluate the risk of breast cancer. Individuals assessed as high-risk might be suggested for breast cancer screening.

Keywords: female breast cancer, incidence, mortality, risk prediction, BI-RADS, Macau