Validating Breast Cancer Risk Prediction Models in the Korean Cancer Prevention Study-II Biobank

Abstract

Background: Risk prediction models may be useful for precision breast-cancer screening. We aimed to evaluate the performance of breast cancer risk models developed in Europeanancestry studies in a Korean population. Methods:We compared discrimination and calibration of three multivariable risk models in a cohort of 77,457 women from the Korean Cancer Prevention Study (KCPS)-II. The first incorporated US breast-cancer incidence and mortality rates, US risk-factor distributions, and relative risk (RR) estimates from European-ancestry studies. The second recalibrated the first by using Korean incidence and mortality rates and Korean risk factor distributions, while retaining the European-ancestry RR estimates. Finally, we derived a Korea-specific model incorporating the RR estimates from KCPS. Results:The US European-ancestry breast cancer risk model was well calibrated among Korean women <50 years (Expected/Observed=1.124 (0.989, 1.278)) but markedly overestimated the risk for those ≥50 years (E/O=2.472 (2.005, 3.049)). Recalibrating absolute risk estimates using Korean breast cancer rates and risk distributions markedly improved the calibration in women ≥50 (E/O=1.018 (0.825, 1.255)). The model incorporating Korean-based RRs had similar but not clearly improved performance relative to the recalibrated model. Conclusions:The poor performance of the US European-ancestry breast cancer risk model among older Korean women highlights the importance of tailoring absolute risk models to specific populations. Recalibrating the model using Korean incidence and mortality rates and risk factor distributions greatly improved performance. Impact:The data will provide valuable information to plan and evaluate actions against breast cancer focused on primary prevention and early detection in Korean women.

Polygenic scoring accuracy varies across the genetic ancestry continuum

“…However, the variation in PGS performance across different genetic ancestries and/or socio-demographic features (for example, sex, age and social determinants of health) 2 poses a critical equity barrier that has prevented widespread adoption of PGSs. Similar portability issues have also been reported for non-genetic clinical models [16][17][18] . The interpretation and application of PGSs are further complicated by the conflation of genetic ancestries with social constructs such as nationality, race and/or ethnicity.…”

“…Finally, we highlight that, just like PGS, the traditional clinical risk assessment may suffer from limited portability across diverse populations 18 . For examples, the pooled cohort equation overestimates atherosclerotic cardiovascular disease risk among non-European populations 16 ; and a traditional clinical breast cancer risk model developed in the European population in the USA overestimated the breast cancer risk among older Korean women 17 . Here we focus on genetic prediction potability owing to the wide interest and attention from both the research community and society.…”

Machine Learning Algorithms to Predict Recurrence within 10 Years after Breast Cancer Surgery: A Prospective Cohort Study

“…Although many forecasting models for predicting outcomes after breast cancer surgery have been proposed in recent years, models for predicting recurrence within 10 years after breast cancer surgery have had major shortcomings: (1) recently proposed forecasting models have lower prediction accuracy compared to conventional models [ 6 , 7 ], (2) proposed forecasting models require use of health insurance claims data, which may be unavailable for real-time use in clinical settings [ 8 , 9 ], and (3) predictions of postoperative recurrence after breast surgery do not consider demographic characteristics, clinical characteristics, quality of care and preoperative health-related quality of life [ 10 , 11 ]. Successful applications of statistical data mining and machine learning methods have been demonstrated in the medical field [ 7 , 8 , 9 , 10 , 11 ]. Clinical and genetic information can be used to improve precision in estimating prognosis and to obtain a comprehensive overview of a disease.…”

Polygenic Risk Scores for Prediction of Breast Cancer in Korean women

“…We previously used Individualized Coherent Absolute Risk Estimation (iCARE) [17] to validate three risk prediction models (the U.S.-based European-ancestry model, a recalibrated model, and a fully Korean-based model) based on classical breast cancer risk factors in a Korean population [3]. Here we evaluate the predictive capacity of four PRS developed using Asian or European training samples; two PRS were restricted to genome-wide significant SNPs (GRS-11 ASN and GRS-136 EUR ) and two included sub-genome-wide significant SNPs (PRS-42 ASN PRS-209 EUR ).…”

“…While the incidence of breast cancer in Asian women is currently lower than that in Western countries, the age distribution of breast cancer incidence in Asian women is markedly different from that in the Western countries, with a peak at 45–49 years in the Asian countries vs. 60–70 years in the Western countries 1,2 . We previously found this age difference led to overestimation of risk in Korean women when conventional breast cancer risk models developed in European-ancestry populations were used 3 . This underscores the need to validate Western-derived risk prediction models in Asian women and adapt them to improve their predictive ability.…”