PRIMES: A Biological and Socio-Environmental Approach to Machine Learning for Equitable and Proactive Cancer and Health Screening

该项目是与数学和统计创新研究所(IMSI)合作开展的。该项目首先由国际癌症研究所参与题为《代数统计和我们的变化世界》的2023年秋季IMSI长期计划，然后继续进行一系列三个科学项目以及教育/推广活动，旨在使个人能够有效地评估他们的癌症和健康风险，从而使我们能够在早期阶段主动发现癌症和疾病。利用芝加哥患者与结直肠癌、肺癌和产后健康结果相关的电子医疗记录，以及社会环境信息，将测试一种新颖而公平的机器学习方法，并将其与当前和广泛使用的算法进行比较，以不仅预测癌症和健康结果，而且研究暴露于暴力中对我们健康的影响。对三个研究项目(结直肠癌、肺癌和产后健康结果)的全面数学和统计调查不仅将推动医疗预测建模，还将促进和促进机器学习的使用，以便在医疗保健和其他领域有效、准确和公正地使用预测建模。参与IMSI Long计划将使PI接触到相关领域的最新研究成果和未来的想法，并将提供足够的时间与具有发展新科学合作潜力的研讨会参与者进行讨论，并加强他的学生和合作者的研究。科学和教育活动将改善社会中个人的福祉，减少社会中的不平等现象，减少缺乏服务的社区对健康的不信任，并增加妇女和代表不足的少数群体在STEM中的人数，特别是在数学和统计学方面。正在出现的科学证据表明，社会和社区层面的因素可以引发有毒的、持续的压力反应，从而促进与癌症发展相关的生物学变化。PI将利用芝加哥患者与结直肠癌、肺癌和产后癌症/健康结局相关的电子医疗记录，以及社会环境信息，对分类机器学习方法进行全面调查，即三重判别评分方法，并将其性能与现有和广泛使用的技术，如极端梯度增强和神经网络进行比较。我们使用我们的三重判别评分方法对结直肠腺瘤进行了初步预测建模，该方法依赖于对所有可能的数据变量子集进行的大量模拟和优化测试，显示出对训练数据分布变化的稳健性，而不是极端梯度提升。这三个研究项目，即预测结直肠癌、肺癌和产后癌症/健康结果，将提供三个大型电子病历数据集，我们将在这些数据集上对各种机器学习分类方法进行全面的数学、统计和实证调查，以有效、准确和公正地使用机器学习在医疗保健和其他领域。此外，PI将参加数学和统计创新研究所(IMSI)2023年秋季的长期计划，题为代数统计和我们不断变化的世界，以利用和扩大他的跨学科研究，发展新的研究合作，并加强他的学生和合作者的研究。最后，科学和教育活动将改善社会中个人的福祉，减少社会中的不平等，减少在服务不足的社区中对健康的不信任，并增加STEM中妇女和代表性不足的少数群体的数量，特别是在数学和统计学方面。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

This project is in collaboration with the Institute for Mathematical and Statistical Innovation (IMSI). The project starts with the participation of the PI in the IMSI fall 2023 long program, entitled Algebraic Statistics and Our Changing World, then continues with a series of three scientific projects along with educational/outreach activities aimed to empower individuals to effectively assess their cancer and health risks, and thus enable us to be proactive in detecting cancer and diseases at earlier stages. Using electronic medical records of patients in Chicago, related to colorectal cancer, lung cancer, and postpartum health outcomes, along with socio-environmental information, a novel and equitable machine learning methodology will be tested and compared to current and broadly used algorithms, to not only predict cancer and health outcomes, but to also study the effect of exposure to violence on our health. The full mathematical and statistical investigation in each of the three research projects (colorectal cancer, lung cancer, and postpartum health outcomes) will not only advance healthcare predictive modeling, but also inform and advance the use of machine learning for an effective, accurate, and unbiased use of predictive modeling in healthcare and beyond. The participation in the IMSI long program will expose the PI to the state of the art of research in related fields and ideas for the future, and will provide adequate time for discussion with workshop participants with the potential to develop new scientific collaborations, and enhance the research of his students and collaborators. The scientific and educational activities will improve the well-being of individuals in society, reduce inequities in society, reduce health distrust among underserved communities, and increase the number of women and underrepresented minorities in STEM in general, and in mathematics and statistics in particular. Scientific evidence is emerging suggesting that societal and neighborhood level factors can elicit a toxic and sustained stress response that promotes biological changes associated with the development of cancers. Using electronic medical records of patients in Chicago related to colorectal, lung, and postpartum cancer/health outcomes, along with socio-environmental information, PI will perform a full investigation of a classification machine learning methodology, i.e., the triple discriminant scoring methodology, and compare its performance to existing and broadly used techniques, such as Extreme Gradient Boosting and Neural Networks. Our preliminary predictive modeling of colorectal adenomas using our triple discriminant scoring methodology, which relies on a high number of simulations and optimization tests across all possible subsets of data variables, showed robustness against change in training data distribution, unlike for the Extreme Gradient Boosting. The three research projects, i.e., predicting colorectal, lung and postpartum cancer/health outcomes, will provide three large electronic medical records data sets on which we will conduct a comprehensive mathematical, statistical and empirical investigation of various machine learning classification methods to extract an effective, accurate, and unbiased use of machine learning in healthcare and beyond. In addition, the PI will participate in the Institute for Mathematical and Statistical Innovation (IMSI) fall 2023 long program, entitled Algebraic Statistics and Our Changing World, to harness and expand his interdisciplinary research, develop new research collaborations, and enhance his students and collaborators research. Finally, scientific and educational activities will improve the well-being of individuals in society, reduce inequities in society, reduce health distrust among underserved communities, and increase the number of women and underrepresented minorities in STEM in general, and in mathematics and statistics in particular.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.