心力衰竭患者生存预测与用药经济学研究
Survival Prediction and Pharmacoeconomics in Heart Failure Patients

本文首先基于Cox生存模型和多种机器学习方法来构建心衰患者的生存预测模型，通过正则化方法识别出对心衰疾病具有重要影响的生物标记特征，并对所构建的五种生存预测模型的预测效果进行了比较分析，发现Cox回归模型和随机森林方法对心衰患者的生存预测效果是最为显著的。考虑到合理的用药策略不仅能够改善患者的结局，还能减轻患者的经济负担，因此本文将识别出的与心衰患者生存时间最为相关的五个风险因素进行了特征排名，依据单变量检验法和机器学习算法的排序结果确定出最为关键的两个特征指标，分别是射血分数和血清肌酐，进而通过构建Markov模型来比较和评估调节射血分数的药物的治疗效果和经济效用。总而言之，人工智能与医疗大数据的不断融合与创新为精准预测心衰患者的生存风险和精准识别影响心衰疾病的关键生物指标提供了可能，结合长期用药带来的经济负担，这些识别出的关键生物指标能为用药策略或治疗策略的制定提供有效的指导。
This article initially constructs a survival prediction model for heart failure patients based on the Cox survival model and various machine learning methods. Through regularization methods, it identifies biomarkers that have a significant impact on heart failure disease. A comparative analysis of the predictive effects of the five constructed survival prediction models reveals that the Cox regression model and the random forest method exhibit the most notable predictive effects for heart failure patients’ survival. Considering that a rational medication strategy can not only improve patient outcomes but also alleviate their economic burden, this study ranks the five risk factors most closely associated with the survival time of heart failure patients. Based on the results of univariate tests and the ranking from machine learning algorithms, the two most critical feature indicators are identified as ejection fraction and serum creatinine. Subsequently, a Markov model is constructed to compare and assess the therapeutic efficacy and economic utility of medications that regulate ejection fraction. In summary, the continuous integration and innovation of artificial intelligence with medical big data provide possibilities for the precise prediction of heart failure patients’ survival risks and the accurate identification of key biological indicators affecting heart failure disease. Combining the economic burden of long-term medication, these identified key biological indicators can offer effective guidance for the formulation of medication or treatment strategies.