A prediction model to simultaneously estimate personal risk of breast cancer and death from other causes in women aged 55 and older

乳腺癌的发病率随着年龄的增长而增加。虽然乳房X光检查可以减少乳腺癌 女性死亡率40-74岁，其疗效在女性中为75岁未知，平均为10.7岁 在1000名妇女中有1人避免因接受筛查而死于乳腺癌的前几年。这也有风险。 进行筛查，包括过度诊断(检测到非致命性肿瘤)。因此，指南建议 临床医生在决定进行筛查时会考虑老年女性的乳腺癌风险和预期寿命。然而， 临床医生发现很难评估老年妇女的乳腺癌风险和健康如何相互作用来确定 筛查的潜在好处是否大于风险。很可能是因为现有的模型不是 专为老年女性开发，没有一种模型可以同时预测乳腺癌和非乳腺癌 癌症(卑诗省)的死亡有助于为这些决定提供信息。为了改善老年女性的乳腺癌预测，我们 之前开发了一个新的模型来预测绝经后女性5年内患乳腺癌的风险&GT；55 使用竞争风险回归(CRR)和护士健康研究(NHS)的数据。然后我们 检查了我们的模型在妇女健康倡议(WHI)参与者中的表现。我们的模型考虑到 年龄、家族史、健康行为、生殖因素和健康状况在评估乳腺癌风险中的作用。我们 发现我们的模型准确地预测了55-74岁女性的乳腺癌，但低估了 妇女&GT；在WHI有75人。我们模型的辨别能力类似于乳腺癌风险评估工具 (BCRAT，初级保健中使用的最常见的风险模型)，但我们的模型准确地进行了风险分层 比BCRAT年长的女性更多。在实施我们的模型之前，我们的目标是将其扩展到预测10年的风险 使用CRR和NHS的非乳腺癌死亡数据，因为考虑到年龄较大 女性的10年预期寿命与考虑她们患乳腺癌的风险一样重要，在制定适当的 乳房X光检查的决定。我们还打算通过使用Black来提高模型的泛化能力 女性健康研究数据和CRR确定乳腺癌特定种族风险因素风险比 (BC)和非BC死亡在我们的模型中使用，并将我们的模型校正到基于人群的乳腺癌 和非卑诗省的死亡率。我们将在两个不同的独立模型中检查我们最终模型的性能 队列(WHI和多种族队列)，并将比较其在预测乳腺癌方面的表现 BCRAT和Tyrer-Cuzick(TC)型号，因为TC也越来越多地被推荐用于一般 人口。在敏感性分析中，我们将检查将乳房密度和基因组数据添加到我们的 模特。为了使我们的模型易于访问，我们将把它添加到我们广泛使用的ePrognosis网站，并 确保我们的模型的网页是用户友好的，我们将测试其与最终用户的可接受性。减少过度- 在老年妇女中筛查乳腺癌是NCI的优先事项，我们预计我们的模型的使用将有所帮助 优化老年妇女对乳房X光检查的使用，从而改善她们的护理和生活质量。

The incidence of breast cancer increases with age. While mammography screening reduces breast cancer mortality in women 40-74 years its efficacy in women >75 years in not known and on average it takes 10.7 years before 1 in 1,000 women avoids breast cancer death as a result of being screened. There are also risks to being screened, which include overdiagnosis (detection of non-lethal tumors). Thus, guidelines recommend that clinicians consider older women's breast cancer risk and life expectancy when deciding on screening. Yet, clinicians find it difficult to assess how an older woman's breast cancer risk and health interact to determine whether the potential benefits of screening outweigh the risks. Likely because existing models were not developed for use with older women and no model simultaneously predicts breast cancer and non-breast cancer (BC) death to help inform these decisions. To improve breast cancer prediction in older women, we previously developed a novel model to predict 5-year breast cancer risk among postmenopausal women >55 years using competing risk regression (CRR) and data from the Nurses' Health Study (NHS). We then examined our model's performance among Women's Health Initiative (WHI) participants. Our model considers age, family history, health behaviors, reproductive factors, and health in assessing breast cancer risk. We found that our model accurately predicted breast cancer in women 55-74 but underpredicted breast cancer in women >75 in WHI. Our model's discrimination was similar to that of the Breast Cancer Risk Assessment Tool (BCRAT, the most common risk model used in primary care) in WHI but our model accurately risk-stratified more older women than BCRAT. Before implementing our model, we aim to extend it to predict 10-year risk of death from causes other than breast cancer using CRR and NHS data because consideration of older women's 10-year life expectancy is as important as considering their breast cancer risk in making appropriate mammography screening decisions. We also aim to improve our model's generalizability by using Black Women's Health Study data and CRR to determine race-specific risk factor hazard ratios for breast cancer (BC) and non-BC death to use in our model and by calibrating our model to population based breast cancer and non-BC death incidence rates. We will examine our final model's performance in two diverse independent cohorts (WHI and the Multiethnic cohort) and will compare its performance in predicting breast cancer to that of the BCRAT and the Tyrer-Cuzick (TC) models since TC is also increasingly recommended for use in a general population. In sensitivity analyses, we will examine the effect of adding breast density and genomic data to our model. To make our model easily accessible we will add it to our widely used ePrognosis website and to ensure our model's webpage is user-friendly we will test its acceptability with end users. Reducing over- screening for breast cancer in older women is an NCI priority and we anticipate that use of our model will help optimize older women's use of mammography and as a result will improve their care and quality of life.