基于LASSO回归的胶质瘤早期鉴别诊断模型的构建和验证

Construction and validation of early differential diagnosis model for glioma based on LASSO regression

目的利用术前部分血液检查指标、年龄和性别构建预测低级别胶质瘤(LGG)和高级别胶质瘤(HGG)的列线图模型。方法回顾性分析2015年12月至2022年9月在徐州医科大学附属医院接受手术治疗的685例胶质瘤患者的临床数据和检测结果,其中,LGG 260例,HGG 425例。先通过组间比较寻找LGG和HGG组之间的差异因素,然后将P<0.05的16个因素通过LASSO回归算法筛选出系数不为0的因素进行多因素logistic分析建立预测模型构建列线图,并通过受试者工作特征(ROC)曲线、校正曲线和决策曲线分析评估模型。结果通过LASSO回归筛选出系数不为0的5个因素,分别为中性粒细胞、白细胞计数、白蛋白计数、性别和年龄。同时这5个因素构建的列线图结果显示该模型ROC曲线下面积为0.7497(95%CI:0.7112~0.7883),灵敏度为0.7600,特异度为0.6808,同时该列线图模型的一致性和有效性通过Calibration曲线和Decision曲线进行证实。结论该列线图模型可以作为临床鉴别诊断HGG和LGG的量化工具,具有较高的诊断效果。

Objective To construct the nomogram model for predicting low-grade glioma(LGG)and high-grade glioma(HGG)by using some preoperative blood test indicators,age and gender.Methods The clinical data and detection results of 685 patients with glioma receiving surgery in the Affiliated Hospital of Xuzhou Medical University from December 2015 to September 2022 were retrospectively analyzed.Among them,260 cases were LGG and 425 cases were HGG.First,the difference factors between the LGG and HGG groups were found through inter-group comparison,and then the factors whose coefficients were not 0 were secrenned out from the 16 factors with P<0.05 by the LASSO regression algorithm,and the multi-factor Logistic analysis was performed to establish a prediction model to construct a nomogram.The model was evaluated through the receiver operating characteristic(ROC)curve,correction curve and decision curve analysis.Results Through LASSO regression,5 factors with coefficients not 0 were screened out,namely neutrophil ratio,white blood cell count,albumin count,gender and age.At the same time,the results of the nomogram constructed by these 5 factors showed that the area under the ROC curve of the model was 0.7497(95%CI:0.7112-0.7883),the sensitivity was 0.7600,and the specificity was 0.6808.At the same time,the consistency and effectiveness of the nomogram model were confirmed by Calibration curve and Decision curve.Conclusion The nomogram model could be used as a quantitative tool for clinical differential diagnosis of HGG and LGG,and has a high diagnostic effect.