冠状动脉粥样硬化性心脏病验前概率机器学习模型

The pretest probability of coronary heart disease determines the clinical pathway of patients with coronary heart disease. The accurate pretest probability model can not only reduce the missed diagnosis of patients with coronary heart disease, but also reduce the unnecessary financial burden and additional risk caused by non-invasive and invasive imaging. The present pretest probability is based on the traditional statistical model, and there is an obvious tendency to overestimate or underestimate the risk of coronary heart disease. With the development of artificial intelligence technology and machine learning algorithm, it is possible to establish a more accurate pretest probability model using multi-variables and large sample data. Samples of this study is 30000 cases, each sample collected more than 40 baseline risk factors for coronary artery disease. We use coronary CT angiography to judge the existence of coronary heart disease or not, using XGBoost algorithm to build the machine learning model for pretest probability of coronary heart disease. Furthermore, the advantages of machine learning model are proved by comparing with existing models. We sought to produce a prediction model for the likelihood of coronary artery disease using machine learning and to support potential and further validation of machine learning derived estimates to guide risk estimates and subsequent management decisions that may positively impact lower risk of downstream unnecessary investigation as well increase diagnostic yield of both non-invasive and invasive testing.

冠心病验前概率决定冠心病患者的临床路径。准确的验前概率模型不仅可以减少冠心病患者的漏诊，而且可以减少不必要的无创及有创影像学带来的经济负担及额外风险。目前存在的验前概率基于传统的统计学模型，存在明显高估或低估冠心病风险的情况。随着人工智能技术及机器学习算法的发展，使利用多变量、大样本数据建立更加准确的验前概率模型成为可能。本课题预期纳入30000例样本冠状动脉CT造影队列，每个样本收集40余项基线危险因素，冠状动脉CT造影判断患者是否存在冠心病，利用XGBoost机器学习模型，建立基于患者基线资料的冠心病验前概率机器学习模型。并进一步和目前已经存在的模型进行对比分析，证明机器学习模型的优势。课题的实施目的就是建立优于目前存在模型的验前概率机器学习模型，更加准确的指导患者的临床诊疗路径，更好的发挥无创及有创影像学在冠心病诊疗路径中的作用。

冠心病验前概率决定冠心病患者的临床路径。准确的验前概率模型不仅可以减少冠心病患者的漏诊，而且可以减少不必要的无创及有创影像学带来的经济负担及额外风险。目前存在的验前概率基于传统的统计学模型，存在明显高估或低估冠心病风险的情况。随着人工智能技术及机器学习算法的发展，使利用多变量、大样本数据建立更加准确的验前概率模型成为可能。本项目基于12321例门诊疑诊冠心病，行冠状动脉CT血管成像的患者，根据患者的CT结果判断患者是否存在冠心病，利用机器学习方法（XGBoost），建立基于患者基线信息的冠心病验前概率模型，通过统计学分析，初步验证模型准确性为81%，且该模型优于传统的验前概率风险评估模型（MDF, CAD consortium score and CONFIRM score），如果利用该模型规范冠状动脉CT血管成像的检查适应证，将会正确改变10.6%患者的临床路径。项目的研究结果有助于优化冠心病的患者的临床路径，使无创影像学及有创影像学的应用更加合理，既可以增加冠心病患者的检出率，又可以减少不必要的影像学检查带来的经济负担及检查风险。本研究作为方法学及模型建立的基础性研究，为未来人工智能技术分流冠心病患者提供模型理论支持。