肺结节中西医结合恶性风险预测模型的建立与验证
Construction and Verification of An Integrated Traditional Chinese and Western Medicine Model for Predicting Malignant Risk of Pulmonary Nodules
杨倩 1肖晶旻 1陈远彬 1吴蕾1
作者信息
- 1. 中医证候全国重点实验室,广州中医药大学第二临床医学院,广州 510120
- 折叠
摘要
目的:基于临床数据探索肺结节恶性风险的危险因素,构建肺结节中西医结合恶性风险预测模型并使用列线图将预测结果可视化.方法:基于横断面调查研究设计,纳入2023年4月至2024年1月于广东省中医院呼吸与心胸外科住院的肺结节患者,按7∶3随机将数据集分成训练集和验证集.在训练集中,通过单因素Logistic回归分析及最小绝对值压缩选样模型(LASSO)回归分析选择预测因素,经Logistic回归方法建模,对西医及中西医结合预测模型的区分度、校准度及临床决策曲线等性能进行比较,筛选出最优模型,并使用列线图对模型进行可视化呈现.结果:本研究共纳入366例患者,划分为训练集(258例)及验证集(108例),可纳入年龄、嗜食肥甘厚腻、环境或职业暴露史、气虚质、阳虚质、结节密度及结节直径等7个预测因素.采用Logistic回归构建模型,仅以年龄、环境或职业暴露史、结节密度及结节直径为预测因素建立的西医模型定义为model1,而在以上基础上增加嗜食肥甘厚腻、气虚质和阳虚质为预测因素建立的中西医结合模型定义为model2,model2在训练集及验证集中具有更好的预测效能,其训练集准确率为0.740,精确率为0.825,召回率为0.829,F1分数为0.827,曲线下面积(AUC)为0.865[95％置信区间(95％CI)0.815～0.915],布里尔评分为0.122;验证集准确率为0.731,精确率为0.776,召回率为0.831,F1分数为0.803,AUC为0.852(95％CI 0.776～0.927),布里尔评分为0.149,校准曲线及决策曲线分析显示该模型在预测中具有较好的一致性与临床效用性.构建方程为肺结节恶性概率p=eY/(1+eY),其中Y=-4.187+0.022×(年龄)+1.079×(环境或职业暴露史)+0.718×(嗜食肥甘厚腻)+1.403×(气虚质)+0.732×(阳虚质)+0.349x(纯磨玻璃密度结节)+1.304×(部分实性密度结节)+0.148×(结节直径),并采用直观的列线图进行展示.结论:该研究开发了一个肺结节恶性风险的综合评估模型,结合了中医和西医的因素,将中医元素嵌入预测模型可以多维度评估结节恶变的相关危险因素,具有良好的诊断效能,为早期识别肺癌提供一定的参考价值.
Abstract
Objective:This study explored the risk factors for malignant risks of pulmonary nodules based on clinical data,constructed an integrated traditional Chinese and Western medicine model for predicting malignant risks of pulmonary nodules,and visualized the prediction results by using a nomogram.Methods:Based on a cross-sectional survey study design,patients with pulmonary nodules who were hospitalized in the Department of Respiratory and Cardiothoracic Surgery of Guangdong Provincial Hospital of Traditional Chinese Medicine from April 2023 to January 2024 were included.The dataset was randomly divided into a training set and a validation set according to7∶3.In the training set,predictive factors were selected through univariate Logistic regression analysis and Least Absolute Shrinkage and Selection Operator(LASSO)regression analysis,and Logistic regression models were built.The discriminative ability,calibration,and clinical decision-making curves of the Western medicine model and the integrated traditional Chinese and Western medicine prediction model were compared to select the optimal model,which was then visualized in a nomogram.Results:This study included a total of 366 patients,and they were divided into a training set(258 cases)and a validation set(108 cases).Seven predictive factors were considered including age,preference for fatty and greasy foods,history of environmental or occupational exposure,Qi deficiency,Yang deficiency,nodule density,and nodule diameter.A Logistic regression model was constructed.A Western medicine model,defined as model 1,was created using only age,history of environmental or occupational exposure,nodule density,and nodule diameter as predictive factors.In addition,an integrated traditional Chinese and Western medicine model,defined as model2,was created by adding preference for fatty and greasy foods,Qi deficiency,and Yang deficiency as predictive factors.Model2 demonstrated better predictive performance in both the training and validation sets.Its accuracy in training set was 0.740,with precision of 0.825,recall of 0.829,F1 score of 0.827,the area under the curve(AUC)of 0.865(95％confidence interval(CI):0.815-0.915),and a Brier score of 0.122.The accuracy in validation set was 0.731,with precision of 0.776,recall of 0.831,F1 score of 0.803,AUC of 0.852(95％CI:0.776-0.927),and a Brier score of 0.149.The calibration curve and decision-making curve analysis showed that this model exhibited good consistency and clinical utility in prediction.The equation for the malignant probability of pulmonary nodules was defined as p=eY/(1+eY),where Y=-4.187+0.022×(age)+1.079×(history of environmental or occupational exposure)+0.718×(preference for fatty and greasy foods)+1.403×(Qi deficiency)+0.732×(Yang deficiency)+0.349×(pure ground glass density nodules)+1.304×(partially solid density nodules)+0.148×(nodule diameter).A nomogram was used for intuitive representation.Conclusion:This study develops a comprehensive assessment model for the malignant risk of pulmonary nodules,which integrates factors from traditional Chinese medicine and Western medicine.Embedding traditional Chinese medicine elements in the prediction model allows for a multidimensional assessment of relevant risk factors for nodule malignancy.This demonstrates good diagnostic performance and provides a reference for early identification of lung cancer.
关键词
肺结节/中西医结合/预测模型/危险因素/列线图Key words
pulmonary nodule/integrated traditional Chinese and Western medicine/prediction model/risk factor/nomogram引用本文复制引用
出版年
2025
国家科技期刊平台
NSTL
万方数据