Project 3

项目摘要--项目3 幸存者的第二次乳腺癌的发病率是没有这种病史的妇女的第一次乳腺癌发病率的7倍；这些癌症与发病率和乳腺癌特有的死亡率的增加有关。建议幸存者进行年度监测乳房X光检查，以便及早发现第二次乳腺癌，以降低发病率和死亡率。然而，监视乳房X光检查并不完美。幸存者的监测失败率(乳房X光检查阴性后12个月内确诊的癌症)比没有乳腺癌病史的女性高4倍。项目3‘S的总体目标是通过公平预测高危女性监测失败(即第二期乳腺癌)，通过人工智能(AI)改进癌症检测，以及将健康的社会决定因素作为监测失败的多层次驱动因素和未来干预的目标，来减少乳腺癌幸存者的监测失败。在广大幸存者群体中拥有准确的监测失败风险预测工具，将支持定向监测成像的未来。了解不同种族的预测绩效是否存在差异，对于验证准确的风险预测模型至关重要。外部验证也是确保模型通用性的关键。此外，未来基于风险的监测策略对转诊医生和妇女的可接受性是重要的，需要考虑，但尚未进行评估。在目前的P01中，我们开发了第一个可以应用于个体幸存者的监测失败的风险模型。在下一个资金周期，我们的目标1将通过评估种族群体的外部概括性和可接受性来评估绩效公平，从而解决采用BCSC 5年监测失败风险预测模型的可行性。我们的目标2将评估人工智能算法是否在与BCSC放射科医生相同的固定特异度下提高敏感度。商业上可用的用于乳房X光检查解释的人工智能算法已经被实施。然而，这些算法是在筛查人群中开发的，尚未在幸存者中得到验证。最后，我们的目标3将确定导致监测失败的选定的多层次健康社会决定因素的相对贡献。多项研究表明，包括收入、教育、住房和种族在内的健康的社会决定因素与较高的二次乳腺癌发病率和死亡率有关。以前没有研究评估过可能受到干预的多层次社会健康决定因素与存活者的间歇性第二乳腺癌之间的关系。了解多水平因素对监测失败的相对贡献可用于指导未来有针对性的干预措施，以减少监测失败。总而言之，该项目将产生证据，通过多管齐下的努力，在风险分层和成像解释方面减少监测失误，并全面审查作为监测失误的多层面驱动因素和未来干预目标的健康的关键社会决定因素，从而改善监测。

PROJECT SUMMARY – Project 3 The incidence of second breast cancer in survivors is seven times higher than first breast cancer incidence in women without this history; these cancers are associated with increased morbidity and breast-cancer specific mortality. Annual surveillance mammography is recommended for survivors for early detection of a second breast cancer to reduce morbidity and mortality. However, surveillance mammography is imperfect. Survivors have 4x higher surveillance failure rates (cancers diagnosed within 12 months of a negative mammogram) than women without a history of breast cancer. Project 3’s overarching goal is to reduce surveillance failures in breast cancer survivors through equitably predicting women at high risk of a surveillance failure (i.e., interval 2nd breast cancer), improving cancer detection through artificial intelligence (AI), and examining social determinants of health as multilevel drivers of surveillance failures and targets for future interventions. Having an accurate surveillance failure risk prediction tool across broad populations of survivors would support a future of targeted surveillance imaging. Understanding whether predictive performance differs by race is critical to validating an accurate risk prediction model. External validation is also key to ensuring the generalizability of the model. Additionally, acceptability of future risk-based surveillance strategies for referring physicians and women is important to consider and has not yet been evaluated. In the current P01, we developed the first risk model of surveillance failures that can be applied to individual survivors. In the next funding cycle, our Aim 1 will address feasibility of adopting the BCSC 5-year surveillance failure risk prediction model by assessing performance equity by race group external generalizability, and acceptability. Our Aim 2 will evaluate whether AI algorithms improve sensitivity at a fixed specificity equivalent to that of BCSC radiologists. Commercially available AI algorithms for mammography interpretation are already being implemented. However, these algorithms were developed in screening populations and have not yet been validated in survivors. Lastly, our Aim 3 will identify the relative contribution of selected multilevel social determinants of health that contribute to surveillance failures. Multiple studies have shown social determinants of health including income, education, housing, and race are associated with higher rates of second breast cancers and mortality. No prior study has evaluated how multilevel social determinants of health that may be amenable to intervention are associated with interval second breast cancers in survivors. Understanding the relative contribution of multilevel factors to surveillance failures can be used to guide future targeted interventions to decrease surveillance failures. Together, this Project will generate evidence to improve surveillance through a multipronged effort to reduce surveillance failures at the points of risk stratification and imaging interpretation, and an overall examination of key social determinants of health as multilevel drivers of surveillance failure and targets for future interventions.