基于机器学习的冠心病影响因素分析

Study of Prediction Model and Influencing Factors for Coronary Heart Disease Based on Machine Learning

摘要: 目的：分析区域性冠心病诱发风险特征因素，探讨冠心病诊断与管理的有效建议。方法：提出基于层次聚类和Fisher评分的双重特征的HC_MFS模型选择方法，采用上海某医院2020年1月至2022年12月的1314例患者数据，以冠心病为例，对其影响因素(C反应蛋白、血小板分布宽度、季节等20个特征)进行分析。结果：HC_MFS方法获得最优性能，最高准确率在随机森林模型中达到83.84%，CRP、LDL、TG、高血压、季节和最低温为重要风险因素，尤其是考虑到样本数据不平衡性并进行处理后，相比其他方法，HC_MFS方法表现更显著，平均准确率提升11.25%，误差最高降低13.16%。结论：崇明区冠心病诱发不仅与CRP、LDL等病理因素强相关，而且还与当地季节气候因素强相关。HC_MFS方法为冠心病分析提供一种新的基于机器学习应用的技术手段，为区域医疗资源建设与健康管理方案制定提供科学决策支持。

Abstract: Objective: To analyze regional risk factors for coronary heart disease and explore effective recommendations for coronary heart disease (CHD) diagnosis and management. Method: The HC_MFS model selection method based on the dual features of hierarchical clustering and Fisher score is proposed, using the data of 1314 patients from a hospital in Shanghai from January 2020 to December 2022, with coronary heart disease as an example, and the influencing factors (C-reactive protein, platelet distribution width, seasons, etc.) were analyzed. Results: The HC_MFS method obtained optimal performance with the highest accuracy of 83.84% in the random forest model, with CRP, LDL, TG, hypertension, season, and minimum temperature as significant risk factors, especially after considering the sample data imbalance and processing it, the HC_MFS method performed more significantly compared to the other methods, with the average accuracy improved by 11.25% and the error reduced by a maximum of 13.16%. Conclusion: Coronary heart disease induced in Chongming District is not only strongly correlated with CRP, LDL and other pathological factors, but also with local seasonal climatic factors. The HC_MFS method provides a new technical means based on machine learning application for the analysis of coronary heart disease, which can provide scientific decision support for the construction of regional medical resources and the development of health management programs.