摘要
目的:构建肺癌根治术后肺部感染的Lasso-logistic回归预测模型,并进行外部验证.方法:将行肺癌根治术治疗的730例肺癌患者,按照7∶3比例随机分为训练组(n=511)、验证组(n=219).统计术后3d内训练组、验证组肺部感染发生情况,并比较发生、未发生肺部感染患者临床资料,在训练组中通过Lasso-logistic回归模型分析肺癌术后肺部感染的影响因素,根据影响因素构建肺部感染Lasso-logistic回归列线图预测模型,在验证组中对Lasso-logistic回归列线图预测模型进行外部验证.结果:本研究730例肺癌患者术后肺部感染发生率为17.81%,训练组、验证组肺部感染发生率分别为17.42%、18.72%,组间比较,差异无统计学意义(P＞0.05);在训练组、验证组中,发生与未发生肺部感染患者的年龄、吸烟史、肿瘤分期、糖尿病、慢性阻塞性肺疾病(COPD)、引流管留置时间、手术时间、机械通气时间、血清C反应蛋白(CRP)、降钙素原(PCT)、淀粉样蛋白A(SAA)、高迁移率族蛋白B1(HMGB1)、可溶性CD14(sCD14)、CD4+/CD8+水平的比较,差异均有统计学意义(P＜0.05);Lasso-logistic回归模型分析显示,以年龄、COPD、机械通气时间、CRP、PCT、SAA、HMGB1、sCD14、CD4+/CD8+作为肺癌根治术后肺部感染的预测因素,可使模型性能优良且影响因素最少;根据Lasso-logistic回归筛选出的9个预测因素构建肺部感染Lasso-lo-gistic回归列线图预测模型,通过外部验证显示该模型预测肺部感染的曲线下面积(AUC)为0.921(95%CI:0.818～0.984),预测敏感度、特异度分别为91.83%、97.26%,具有良好预测效能,且该模型预测肺部感染的校准度良好,与实际观测结果一致性较高.结论:肺癌根治术后肺部感染发生率高,其影响因素较多,根据筛查出的相关预测因素构建肺部感染Lasso-logistic回归列线图预测模型在预测肺部感染发生风险方面具有良好预测价值,有助于指导临床预防术后肺部感染.
Abstract
Objective:To construct a Lasso-logistic regression prediction model for pulmonary infection after radical lung cancer surgery and conduct external validation.Methods:A total of 730 lung cancer patients who un-derwent radical surgery for lung cancer were randomly divided into a training group(n=511)and a validation group(n=219)in a 7:3 ratio.The incidence of pulmonary infection within 3 days after surgery in the training group and the validation group was calculated,and the clinical data of patients with and without pulmonary infec-tion were compared.In the training group,the influencing factors of pulmonary infection after lung cancer sur-gery were analyzed using the Lasso-logistic regression model,and the Lasso-logistic regression nomogram pre-diction model for pulmonary infection was constructed based on the influencing factors and was externally vali-dated in the validation group.Results:The incidence of postoperative pulmonary infection in 730 lung cancer pa-tients in this study was 17.81%,with 17.42%in the training group and 18.72%in the validation group,and there was no statistical significance between groups(P＞0.05).In the training group and the validation group,the age,smoking history,tumor stage,diabetes,chronic obstructive pulmonary disease(COPD),drainage tube retention time,operation time,mechanical ventilation time,serum C-reactive protein(CRP),procalcitonin(PCT),serum amyloid A(SAA),high mobility group protein B1(HMGB1),soluble CD14(sCD14),and CD4+/CD8+levels of patients with and without pulmonary infection were compared,and the differences were statistically significant(P＜0.05).The Lasso-logistic regression model analysis showed that age,COPD,mechanical ventilation time,CRP,PCT,SAA,HMGB1,sCD14,and CD4+/CD8+ were used as predictive factors for lung infection in lung can-cer patients after radical surgery,resulting in excellent model performance with minimal influencing factors;the Lasso-logistic regression nomogram prediction model for lung infection was constructed based on the 9 predic-tive factors screened out by Lasso-logistic regression.External validation showed that the area under the curve(AUC)of the model for predicting pulmonary infection was 0.921(95%CI:0.818-0.984),with predictive sensi-tivity and specificity of 91.83%and 97.26%,respectively.This model had good predictive performance,and its calibration for predicting pulmonary infection was good,which was well consistent with actual observation re-sults.Conclusion:The incidence of pulmonary infection after lung cancer radical surgery is high,and there are many influencing factors.The Lasso-logistic regression nomogram prediction model for pulmonary infection based on the screened relevant predictive factors has good predictive value in predicting the risk of pulmonary in-fection,which is helpful to guide clinical prevention of postoperative pulmonary infection.
基金项目
陕西省医学科学研究课题计划(2020JM4711)
维普
万方数据