The Mathematics of Breast Cancer Overtreatment: Improving Treatment Choice through Effective Communication of Personalized Cancer Risk

项目总结/摘要 这个职业发展应用程序提供有针对性的课程和指导研究，使亲， 在数学建模和人的高度交叉学科领域的独立研究的回归， 乳腺癌的治疗每年，美国有近6万名妇女在确诊后接受根治性手术， 筛查发现乳腺原位癌（BCIS），但其中多达45,000人接受了乳腺癌治疗， 在其一生中不会进展为浸润性乳腺癌的微小病变。由此产生的过度治疗 非进展性BCIS病变可造成实质性伤害，并显著降低患者的生活质量 而不会降低乳腺癌死亡率。尽管对患有BCIS的女性的普遍过度治疗是好的， 在人口水平上记录，其在患者水平上的预防受到当前治疗标准的阻碍， adigm规定，几乎所有患者都要立即接受治疗。这反过来又使缺乏 评估除立即治疗以外的管理策略所需的数据，如积极的 监视为了解决这个难题，已经启动了主动监测的随机对照试验， 但只是最近才在欧洲出现预计这些试验即使成功，也不会产生临床效果。 至少10年的可操作数据然而，与此同时，有大量现有的临床和生物- BCIS上分散在大量数据和知识源中的逻辑数据。在没有 量化模型，使这些分散的来源的整合，现有的大部分数据重新- 患者无法使用的电源。因此，为了使BCIS患者能够做出明智的决策， 迫切需要（i）开发预测模型，整合可用的患者和肿瘤特异性数据， 针对不同管理策略的个性化风险和不确定性预测，以及（ii）有效地将 将这些个性化的预测传达给患者。在缺乏量化和交流工具的情况下， 尽管个性化风险预测的应用，患者和医生仍然很难权衡权衡 与不同的管理策略相关联，并做出明智的，基于证据的决定， 减少了对BCIS过度治疗的潜在危害。长期目标是发展个性化的... 最大化知情决策和最小化BCIS患者过度治疗的决策辅助工具。的 这项建议的总体目标包括实现这一目标的前三个步骤：㈠发展个性化的 针对BCIS不同管理策略的风险预测模型，（ii）使用这些预测来制定一个 个性化的决策辅助，以及（iii）评估其在无乳腺癌病史的女性测试队列中的影响。 癌我们的中心假设是，基于模型的个性化风险预测的沟通会导致 更好地理解与BCIS不同管理策略相关的权衡。那个... 拟议研究的主要目的是，通过个性化的结果估计，患者可以获得 形成所需的证据为基础的决定，是符合他们的个人风险承受能力。具体 本K99/R 00的指导（K）和独立（R）研究阶段的目标如下。 目标K1：在BCIS页面中发现与个性化风险预测相关的数据和知识源- 数据库，并将它们分别整理成一个统一的数据存储和知识库。 目标K2：开发数学模型，使用数据存储和知识库来计算个性化 针对不同BCIS管理策略的风险预测，包括主动监测。 目的K3：设计一项两阶段研究，以开发、完善和评估基于模型的个性化决策辅助工具， BCIS患者通过认知访谈（第1阶段）和RCT（第2阶段）。 目标R1：执行模型验证和不确定性量化，以最大化模型置信度。 目标R2：第1阶段：进行认知访谈，以开发和完善交互式决策辅助工具， 在BCIS患者中进行个性化风险预测的有效沟通。 目标R3：第2阶段：实施随机对照试验，以检验使用个性化决策辅助工具的主要假设 导致（一）认为积极监测是可行的妇女比例增加 BCIS的管理策略;及（ii）增加对相关风险权衡的了解。 交付成果将包括一个数据驱动的数学建模框架，预计将产生最佳的位置， 针对不同BCIS管理策略的可行患者特异性风险预测。交互式决策辅助 预计将提供与不同的风险和不确定性相关的直观理解， BCIS管理策略。此外，这种方法将在其他屏幕上有广泛的应用- 检测到未知进展风险的病变，例如越来越多地诊断为前列腺、甲状腺 和肺申请人已完成物理学（理学硕士）和数学（博士）的研究生学习，并具有 与在早期乳腺癌方面具有丰富经验的主要导师一起进行的项目，包括 BCIS。基于他与临床医生成功合作研究的历史，申请人是独一无二的 这是一个弥合数学建模和个性化癌症护理之间鸿沟的平台。

Project Summary/Abstract This Career Development Application provides targeted coursework and mentored research to enable pro- gression to independent research in the highly cross-disciplinary areas of mathematical modeling and person- alized breast cancer care. Every year, close to 60,000 women in the US undergo radical surgery after diagno- sis with screen-detected breast carcinoma in situ (BCIS), yet as many as 45,000 of these are treated for be- nign lesions that would not progress to invasive breast cancer in their lifetime. The resulting overtreatment of non-progressive BCIS lesions can cause substantial harms and significantly reduce the patient's quality of life without reducing breast cancer mortality. Although the widespread overtreatment of women with BCIS is well documented at the population level, its prevention at the patient level is hindered by the current treatment par- adigm, which dictates that virtually all patients undergo immediate treatment. This in turn perpetuates the lack of data needed for the evaluation of management strategies other than immediate treatment, such as active surveillance. To resolve this conundrum, randomized controlled trials on active surveillance have been initiated, but only recently and only in Europe. It is anticipated that these trials, even if successful, will not yield clinically actionable data for at least 10 years. At the same time, however, there is a wealth of existing clinical and bio- logical data on BCIS that is dispersed across a large number of data and knowledge sources. In the absence of quantitative models that enable the integration of these dispersed sources, the bulk of the existing data re- mains inaccessible to patients. Thus, to enable informed decision making among patients with BCIS, there is a critical need (i) to develop predictive models that integrate available patient- and tumor-specific data to make personalized risk and uncertainty projections for different management strategies, and (ii) to effectively com- municate these personalized projections to patients. In the absence of tools for the quantification and commu- nication of personalized risk projections, it remains difficult for patients and physicians to weigh the trade-offs associated with different management strategies and to make an informed, evidence-based decision that re- duces the risk of potentially harmful overtreatment of BCIS. The long-term goal is to develop personalized de- cision aids that maximize informed decision-making and minimize overtreatment in patients with BCIS. The overall objective of this proposal comprises the first three steps towards this goal: (i) to develop personalized risk projection models for different management strategies of BCIS, (ii) to use these projections to develop a personalized decision aid, and (iii) to evaluate its impact in in a test cohort of women without a history of breast cancer. Our central hypothesis is that communication of model-based personalized risk projections leads to an improved understanding of the trade-offs associated with different management strategies for BCIS. The ra- tionale for the proposed research is that with personalized outcome estimates, patients gain access to the in- formation needed for an evidence-based decision that is aligned with their personal risk tolerance. The specific aims for the mentored (K) and independent (R) research phases of this K99/R00 are as follows. Aim K1: Discover data and knowledge sources that are relevant for personalized risk projections in BCIS pa- tients, and curate them into a harmonized data store and knowledge base, respectively. Aim K2: Develop mathematical models that use the data store and knowledge base to compute personalized risk projections for different BCIS management strategies, including active surveillance. Aim K3: Design a two-stage study to develop, refine and evaluate a model-based personalized decision aid for BCIS patients through cognitive interviews (Stage 1) and a RCT (Stage 2). Aim R1: Perform model validation and uncertainty quantification to maximize model confidence. Aim R2: Stage 1: Conduct cognitive interviews to develop and refine an interactive decision aid for the effec- tive communication of personalized risk projections in BCIS patients. Aim R3: Stage 2: Implement a RCT to test the main hypothesis that the use of personalized decision aids leads to (i) an increase in the proportion of women who would consider active surveillance as a viable management strategy for BCIS, and (ii) an increase in knowledge of the associated risk trade-offs. The deliverables will include a data-driven mathematical modeling framework, expected to yield the best pos- sible patient-specific risk projections for different management strategies of BCIS. The interactive decision aid is expected to provide an intuitive understanding of the risks and uncertainties that are associated with different BCIS management strategies. Moreover, this approach will have widespread application in other screen- detected lesions of unknown progression risk, such as those increasingly diagnosed in the prostate, thyroid and lung. The applicant has completed graduate studies in physics (MSc) and mathematics (PhD) and has ini- tiated projects with the primary mentor who has extensive experience in early stage breast cancer, including BCIS. Based on his history of successful collaborative research with clinicians, the applicant is in the unique position to bridge the divide between mathematical modeling and personalized cancer care.