Comparison of Breast Cancer Risk Predictive Models and Screening Strategies for Chinese Women

Abstract

Background:

Previous studies have shown that organized mammographic screening implementation in China may not be cost-effective. Our aim was to develop a valid predictive mathematical model for selecting high-risk groups eligible for mammography examinations (MAMs) and cost-effective strategies for breast cancer screening among Chinese women.

Methods:

Between 2009 and 2012, 13,355 eligible women aged 30–65 years were enrolled from the community in Chengdu City. All subjects were administered a valid questionnaire and given MAMs. Using biopsies and 1-year follow up, we compared the accuracy indexes of three predictive models (back-propagation artificial neural network [BP-ANN], logistic regression [LR], and Gail) and four serial screening strategies (BP-ANN→MAM, LR→MAM, Gail→MAM, and MAM alone). We also evaluated the benefits of the four strategies by comparing their incidence-adjusted positive predictive value (PPV). All analyses were conducted with three age-based subgroups: 30–39, 40–49, and 50–65.

Results:

The BP-ANN₁, in conjunction with additional continuous risk factor variables, was the best predictive model, with the highest sensitivity (SEN, 76.99%) and specificity (SPE, 54.20%). The BP-ANN₁→MAM strategy was best for the 40–49 age group, with the highest adjusted PPV (9.80%) and reasonable SEN (81.82%).

Conclusion:

We found that the BP-ANN model performed the best and was the most accurate for predicting high risk for breast cancer among Chinese women, and the BP-ANN→MAM screening strategy was most effective among the 40–49 age group. However, mammography alone may be a sufficient screening strategy for women aged 50–65.