Risk-based Breast Cancer Screening and Surveillance in Community Practice - Admin Supplement for P3

项目摘要 本申请是为了响应被标识为NOT-CA的特别利益通知（NOSI）而提交的- 20-038.该补充的目标是推进实施基于风险的成像的进展 在临床实践中监测乳腺癌。我们建议改进方法， 制定乳腺癌影像监测结果的风险模型，包括监测发现的 第二次乳腺癌（受益）和间隔浸润性乳腺癌（失败），并告知一个 为患有原发性乳腺癌的个体女性提供最佳的基于风险的成像监测策略。这 该提案基于乳腺癌监测联盟（BCSC）的资源， 60，000名有乳腺癌个人史的妇女和超过330，000例监测乳房X光检查 考试研究人员将利用现代数据自适应建模方法，特别是 正则化回归模型和机器学习方法，可以潜在地增强预测 准确性，开发监测结果的风险模型（目标1）。调查人员提出了一个 通过多种指标进行全面的内部验证，以评估通过替代方法开发的风险模型 充分了解其效用的方法和模型之间的权衡，以改善乳腺癌 生存率，同时保持临床可用性和可解释性（目标2.1）。具体而言，调查人员将 评价受试者工作特征曲线下面积（AUC）和每种风险的校准 目标1中开发的模型，并使用净重新分类改进和 可变重要性度量此外，建议使用R Markdown创建在线教程， 加速采用最佳实践，用于其他癌症的现代风险模型开发和验证（Aim 2.2）。本补编中的替代方法学建模方法的评价和传播 将直接为乳腺癌和其他癌症类型的风险分层监测算法的开发提供信息。

PROJECT SUMMARY This application is being submitted in response to the Notice of Special Interest (NOSI) identified as NOT-CA- 20-038. The goals of this supplement are to advance progress toward implementing risk-based imaging surveillance for breast cancer in clinical practice. We propose to improve methodological approaches for developing risk models for breast cancer imaging surveillance outcomes, including surveillance detected second breast cancer (benefit) and interval invasive breast cancer (failure), and inform the development of an optimal risk-based imaging surveillance strategy for individual women with primary breast cancer. This proposal builds on the resources of the Breast Cancer Surveillance Consortium (BCSC) from more than 60,000 women with a personal history of breast cancer and more than 330,000 surveillance mammography examinations. The investigators will leverage modern data-adaptive modeling approaches, specifically regularized regression models and machine learning methods which can potentially enhance prediction accuracy, to develop risk models of surveillance outcomes (Aim 1). The investigators propose a comprehensive internal validation with multiple metrics to evaluate the risk models developed via alternative methods for a full understanding of their utilities and trade-offs between models in improving breast cancer survivorship while maintaining clinical usability and interpretability (Aim 2.1). Specifically, the investigators will evaluate the area under the receiver operating characteristic curve (AUC) and the calibration of each risk model developed in Aim 1, and conduct comparison across models using net reclassification improvement and variable importance measures. Additionally, an online tutorial created using R Markdown is proposed to accelerate uptake of best practices for modern risk model development and validation in other cancers (Aim 2.2). The evaluation and dissemination of alternative methodological modeling approaches in this supplement will directly inform development of risk-stratified surveillance algorithms in breast and other cancer types.