基于TARGET数据库的预测神经母细胞瘤患儿预后的列线图构建

Establishment and validation of a survival nomogram for neuroblastoma children based upon TARGET database

目的构建并验证用于评估神经母细胞瘤患儿无事件生存的临床预测模型。方法收集1986年至2012年来源于TARGET数据库诊断为神经母细胞瘤的817例患儿临床资料,将患儿以2∶1的近似比按随机数字表法分成训练组(544例)和验证组(273例)。LASSO回归分析确定预后相关的影响因素,选取重要的影响因素绘制列线图。绘制Kaplan-Meier生存曲线评价列线图对患儿的危险分层能力,利用一致性指数(C指数)、受试者操作特征(receiver operator characteristic,ROC)曲线及其曲线下面积(area under the curve,AUC)和校准曲线评估列线图的辨别和校准能力,决策曲线评价列线图的临床实用性。结果LASSO回归分析表明年龄、性别、MYCN基因状态、组织学分类、DNA倍体情况和病理类型为神经母细胞瘤患儿无事件生存的独立影响因素。将上述变量纳入预测模型并绘制预测患儿3年、5年、10年的无事件生存率的列线图。预测模型在训练组和验证组中的C指数分别为0.612和0.655,训练组3年、5年、10年ROC曲线的AUC值分别为0.642、0.691、0.715,验证组分别为0.701、0.705、0.776,表明模型具有较好的辨别能力。校准曲线显示,列线图预测的患儿3年、5年、10年无事件生存率与实际结果具有良好的一致性。决策曲线分析表明列线图的预测概率阈值在0.2~0.6时,使用列线图预测神经母细胞瘤患儿的预后更有益。结论基于TARGET数据库中神经母细胞瘤患儿资料构建的列线图预测模型可以对患儿预后风险进行分层,同时计算出患儿的3年、5年和10年无事件生存率,其能辅助临床决策,建议临床推广。

Objective To create and validate a clinical predictive model for event-free survival(EFS)in children of neuroblastoma(NB).Methods From 1986 to 2012,the relevant clinical data for 817 NB children were collected from the database of TARGET.They were randomized by a ratio of approximately 2∶1 into two groups of training(n=544)and validation(n=273).LASSO regression analysis was performed for determining the prognostic factors and important factors were selected for creating a nomogram.Kaplan-Meier survival curve was plotted for evaluating the nomogram's risk stratification capability.Receiver operating characteristic(ROC)curve,area under the curve(AUC)and calibration curve were employed for evaluating the nomogram's discrimination and calibration capability.And decision curve was utilized for evaluating the nomogram's clinical utility.Results LASSO regression analysis revealed that age,gender,MYCN status,histological classification,DNA ploidy and pathological type were independent prognostic factors for EFS.These variables were included into the prediction model and a nomogram was created for predicting the EFS at Year 3/5/10.The nomogram had a C-index of 0.612 in training group and 0.655 in validation group.AUC values were 0.642,0.691 and 0.715 at Year 3/5/10 for training group.The validation group had AUC values of 0.701,0.705 and 0.776 at Year 3/5/10,indicating an excellent discriminative capability of the model.The calibration curve demonstrated an excellent agreement between nomogram-predicted 3/5/10-year EFS and actual results.The decision curve analysis indicated that using nomogram for predicting the prognosis of NB children was more beneficial when the prediction probability threshold was(0.2~0.6).Conclusions Constructed with information from NB children in the TARGET database,the nomogram may stratify the prognostic risks of NB children.It can also calculate the 3/5/10-year EFS rates.Its capability of assisting in clinical decision-making is recommended for wider clinical popularization.