Article type
Abstract
"Background: The burden of postmenopausal breast cancer is rising worldwide. Previous works found that existing risk prediction models can estimate women’s breast cancer risk. However, these models used the additional inputs derived from costly or invasive procedures and may have limited applicability.
Objectives: To find out the life-behavior related risk factors that affect the incidence of postmenopausal breast cancer, and build a postmenopausal breast cancer risk prediction model, identify high-risk individuals, and help the early screening of postmenopausal breast cancer.
Methods: Data for this study were derived from the UK Biobank (UKB), a large prospective cohort of approximately 500,000 UK residents recruited from 21 assessment centres between 2006 and 2010. With reference to postmenopausal breast cancer screening guidelines and high-quality risk factors research literature, the identification criteria of high-risk groups in this study were determined. The risk factors of postmenopausal breast cancer were screened by univariate analysis and full subset regression, and the optimal fitting logistics model was finally selected according to Mallows' Cp values. The sample was randomly divided into training set and test set according to the ratio of 7:3. In the test set, we measured the predictive performance and compared with models with polygenic risk score (PRS), including area under the receiver-operating-characteristics curve (AUC) and Calibration curves for each model.
Results: The study included 231,739 people, and a total of 4318 postmenopausal breast cancers were diagnosed during the follow-up time. Seven independent variables were selected to establish the logistic model: age, family history of breast cancer, physical activity, body mass index, hormone replacement therapy history, alcohol consumption, and mammogram history (P<0.05). The AUC was 0.659 (95 % CI= 0.647- 0.672), and the calibration curve shows that the model has a good effect. After adding the PRS factor, the AUC was 0.669 (95 % CI= 0.645- 0.674).
Conclusion: In this study, we established a large population postmenopausal breast risk prediction model, which showed good performance in discriminating ability, and provided a tool for developing standardized screening strategies for postmenopausal breast cancer. More population-based prospective follow-up studies can be used for the validation to improve the current model.
"
Objectives: To find out the life-behavior related risk factors that affect the incidence of postmenopausal breast cancer, and build a postmenopausal breast cancer risk prediction model, identify high-risk individuals, and help the early screening of postmenopausal breast cancer.
Methods: Data for this study were derived from the UK Biobank (UKB), a large prospective cohort of approximately 500,000 UK residents recruited from 21 assessment centres between 2006 and 2010. With reference to postmenopausal breast cancer screening guidelines and high-quality risk factors research literature, the identification criteria of high-risk groups in this study were determined. The risk factors of postmenopausal breast cancer were screened by univariate analysis and full subset regression, and the optimal fitting logistics model was finally selected according to Mallows' Cp values. The sample was randomly divided into training set and test set according to the ratio of 7:3. In the test set, we measured the predictive performance and compared with models with polygenic risk score (PRS), including area under the receiver-operating-characteristics curve (AUC) and Calibration curves for each model.
Results: The study included 231,739 people, and a total of 4318 postmenopausal breast cancers were diagnosed during the follow-up time. Seven independent variables were selected to establish the logistic model: age, family history of breast cancer, physical activity, body mass index, hormone replacement therapy history, alcohol consumption, and mammogram history (P<0.05). The AUC was 0.659 (95 % CI= 0.647- 0.672), and the calibration curve shows that the model has a good effect. After adding the PRS factor, the AUC was 0.669 (95 % CI= 0.645- 0.674).
Conclusion: In this study, we established a large population postmenopausal breast risk prediction model, which showed good performance in discriminating ability, and provided a tool for developing standardized screening strategies for postmenopausal breast cancer. More population-based prospective follow-up studies can be used for the validation to improve the current model.
"