基于机器学习构建化学疗法期间结直肠癌肝郁脾虚证诊断模型OA
Development of diagnostic model for liver depression and spleen deficiency syndrome in colorectal cancer patients during chemotherapy based on machine learning
目的 基于标准化中医证候要素(症状、体征),通过 LASSO-logistic 回归算法构建化学疗法(简称"化疗")期间结直肠癌患者肝郁脾虚证的量化诊断模型.方法 收集400例化疗期结直肠癌患者的相关数据,判定中医证型,通过LASSO-logistic回归分析构建肝郁脾虚证诊断模型并进行评估.结果 ①400例患者中,肝郁脾虚证占比相对更高(27.75%),其后依次为脾气亏虚证(19.75%)、脾肾阳虚证(14.25%)、湿热蕴结证(11.25%).②LASSO回归分析筛选出12个非零回归系数变量,进一步的单因素logistic回归分析显示其中11个变量是结局的影响因素;将上述11个变量纳入多因素logistic回归分析,最终得到6个变量因素(膝软、胁肋胀痛、情绪不佳、嗳气吞酸、脉数、脉弦),用于构建诊断模型,生成列线图.③各项性能评估(受试者操作特征曲线、校准曲线和决策曲线)结果显示,肝郁脾虚证预测模型在不同数据集上都具有一定的准确性和可靠性.结论 研究构建的肝郁脾虚证诊断模型通过数据驱动实现了中医证候的"经验-数据"双轨诊断模式,不仅为化疗期间结直肠癌的中医精准诊疗提供了工具,更示范了机器学习在中医药现代化中的方法论价值.未来可通过多模态数据融合和外部验证,推动其向临床实践转化.
Objective To develop a quantitative diagnostic model based on standardized traditional Chinese medicine(TCM)syndrome elements(symptoms and physical signs)for liver depression and spleen deficiency(LDSD)syndrome in colorectal cancer patients undergoing chemotherapy using the LASSO-logistic regression algorithm.Methods Relevant clinical data of 400 colorectal cancer patients undergoing chemotherapy were collected,and their TCM syndromes were identified.A diagnostic model for LDSD syndrome was established via LASSO-logistic regression analysis and subsequently evaluated.Results ①Among the 400 patients,the proportion of LDSD syndrome is relatively higher(27.75%),followed by spleen qi deficiency syndrome(19.75%),spleen-kidney yang deficiency syndrome(14.25%),and damp-heat accumulation syndrome(11.25%).② LASSO regression analysis identified 12 variables with non-zero regression coefficients,and further univariate logistic regression analysis indicated that 11 of these variables were influencing factors for the outcome.These 11 variables were included in a multivariate logistic regression analysis,and 6 variables(knee weakness,hypochondriac distension and pain,poor emotional status,belching with acid regurgitation,rapid pulse,and wiry pulse)were finally identified and used to develop the diagnostic model,with a nomogram generated accordingly.③The results of performance evaluations(receiver operating characteristic curve,calibration curve and decision curve analysis)showed that the predictive model for LDSD syndrome had certain accuracy and reliability across different datasets.Conclusions The diagnostic model for LDSD syndrome developed in this study enables a data-driven"experience-data"dual-track diagnostic mode for TCM syndrome differentiation.It not only provides a practical tool for the precise TCM diagnosis and treatment of colorectal cancer during chemotherapy,but also demonstrates the methodological value of machine learning in the modernization of traditional Chinese medicine.In the future,multi-modal data integration and external validation can be adopted to promote its translation into clinical practice.
王紫薇;王朝伟;孙云川;何新颖;毕凌;王炎
上海中医药大学附属曙光医院肿瘤科(上海 201203)上海中医药大学附属曙光医院肿瘤科(上海 201203)河北省沧州中西医结合医院肿瘤科(河北 沧州 061013)河北省沧州中西医结合医院肿瘤科(河北 沧州 061013)上海中医药大学附属曙光医院肿瘤科(上海 201203)上海中医药大学附属曙光医院肿瘤科(上海 201203)
结直肠癌中医证候机器学习人工智能诊断模型肝郁脾虚证
colorectal cancertraditional Chinese medicine syndromemachine learningartificial intelligencediagnostic modelliver depression and spleen deficiency syndrome
《上海中医药杂志》 2026 (6)
1-7,7
国家自然科学基金项目(82474230)
评论