explainable AI, Data Analytics and Industrial Engineering Methods for Primary Care

用于初级保健的可解释的人工智能、数据分析和工业工程方法

• 批准号: RGPIN-2019-05522
• 负责人: Guergachi, Aziz
• 金额: $ 2.62万
• 依托单位: Ryerson University
• 依托单位国家: 加拿大
• 项目类别: Discovery Grants Program - Individual
• 财政年份: 2022
• 资助国家: 加拿大
• 起止时间: 2022-01-01 至 2023-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=757226
• 关键词: explainable; AI; Data; Analytics; Industrial

To tackle the chronic illnesses challenge, the applicant and his team introduced a new paradigm, referred to herein as Predict + Prevent, or P+P. Two important facts underlie this paradigm --- one is scientific and the other is historic. The scientific fact is that most chronic illnesses are actually preventable, if the appropriate lifestyle interventions are applied. The scientific evidence behind the preventability of chronic illnesses is overwhelming, as was reported during the last two decades in a number of articles in renowned medical journals. As for the historic fact, it is the result of a centuries-old paradigm we inherited from the way medicine (from Latin `medicina', which means `the art of healing') has traditionally been practiced: in general, patients need to wait until they get sick to go see their doctors who will then hopefully heal them. Now, with the availability of large sets of Electronic Medical Records data, and the tremendous advancements in information technology and data analytics, patients don't need to wait until they become sick. It is now possible to predict, at scale, the risk for a person to get ill with a chronic disease long before she contracts this disease. Hence the term "Predict" in the expression Predict + Prevent. The implementation of the P+P paradigm would consist of: (1) identify the high-risk patients, by predicting the likelihood for people to develop chronic diseases using EMR data, and then, (2) transfer lists of high-risk patients to a third-party, such as telehealth providers or human resources companies, who would then reach out to these patients (by e-mail, phone, social media, in-person or any other suitable method) and work with them on improving their lifestyle, to move them away from the onset of chronic diseases. In this proposal, the research will be focused mostly on the prediction side of the P+P paradigm and on limited aspects of prevention. More specifically, the proposed research project will address the following 5 projects listed below, using artificial intelligence methods and data analytics (e.g., decision trees, Markov chains with and without memory, Bayesian nets, mixture models, the EM algorithm, kernels and support vector machines, Monte Carlo simulations, unsupervised learning, and so on) for the first three projects, and industrial engineering methods (decision theory, engineering economics) for the last two projects: 1-Project on Prediction 1 - The single chronic disease cases 2-Project on Prediction 2 - Advanced stages and complications of a chronic disease 3-Project on Prediction 3 - Patients with multiple comorbidities 4-Project on Decision Sciences DS - Assessing the value of biomarkers information collected by doctors and their staff members. 5-Project on Engineering Economics EE - The economics of prevention This research will involve highly qualified personnel (HQP), namely 3 PhD students, 3 Master's students and 1 Lab technician.

为了应对慢性疾病的挑战，申请人和他的团队引入了一个新的模式，在这里称为预测+预防，或P+P。这一范式的基础是两个重要事实——一个是科学的，另一个是历史的。科学事实是，如果采取适当的生活方式干预措施，大多数慢性病实际上是可以预防的。正如过去二十年来著名医学期刊上的一些文章所报道的那样，慢性病可预防的科学证据是压倒性的。至于这一历史事实，这是我们从医学（来自拉丁语“medicina”，意思是“治疗的艺术”）的传统实践方式中继承的一个数百年历史的范式的结果：一般来说，病人需要等到他们生病了才去看医生，然后他们才有希望治愈他们。现在，随着大量电子医疗记录数据的可用性，以及信息技术和数据分析的巨大进步，患者不需要等到他们生病了。现在有可能大规模地预测一个人在感染某种慢性病之前患这种疾病的风险。因此，在“预测+预防”这个表达中有“预测”这个词。P+P范式的实现将包括：(1)通过使用电子病历数据预测人们患慢性病的可能性来识别高风险患者，然后(2)将高风险患者名单转移给第三方，例如远程医疗提供者或人力资源公司，然后这些第三方将通过电子邮件，电话，社交媒体，面对面或任何其他合适的方法与这些患者联系，并与他们一起改善他们的生活方式，使他们远离慢性病的发作。在本提案中，研究将主要集中在P+P范式的预测方面和预防的有限方面。更具体地说，拟议的研究项目将涉及以下5个项目，在前三个项目中使用人工智能方法和数据分析（例如，决策树、带和不带内存的马尔可夫链、贝叶斯网络、混合模型、EM算法、核和支持向量机、蒙特卡罗模拟、无监督学习等），在后两个项目中使用工业工程方法（决策理论、工程经济学）：1-预测项目1-单一慢性疾病病例2-预测项目2-慢性疾病的晚期和并发症3-预测项目3-多重合并症患者4-决策科学项目DS -评估医生及其工作人员收集的生物标志物信息的价值。本研究将涉及高素质人才（HQP），即3名博士生、3名硕士生和1名实验室技术员。