基于多维参数诺模图的局限期小细胞肺癌生存期预测研究

Development of multi-parameter nomogram for predicting survival for limited-stage small cell lung cancer

目的 :基于多维参数诺模图预测局限期小细胞肺癌放疗后的总生存期。方法 :回顾性分析某院放疗科2011—2018年122例接受放疗或放化疗联合治疗的局限期小细胞肺癌(其中112例为小细胞肺癌,10例为混合型小细胞肺癌)患者的病例资料。按照8∶2的比例将患者随机分成训练集和验证集。对于临床参数,先对训练集进行十折交叉验证,然后基于Akaike信息标准,通过完全搜索在所有可能的临床参数组合子集中找到最佳的参数子集,再利用Cox比例风险回归模型构建诺模图预测模型;对于放射组学特征,先进行特征标准化,再利用最小绝对收缩和选择算子(least absolute shrinkage and selection operator,LASSO)-Cox比例风险回归筛选具有预测价值的放射组学特征,最后将放射组学特征和最佳临床参数子集中的临床参数相结合构建多参数诺模图联合预测模型。利用Harrell一致性指数(concordance index,CI)对预测模型进行评估,比较2种预测模型对局限期小细胞肺癌生存期的预测能力。利用Rad_score中位数将患者分为高危组和低危组,同时采用Log-rank检验对高、低危组的Kaplan-Meier(KM)生存曲线进行分析和比较。结果:与仅基于临床参数的诺模图预测模型(CI:0.596;95%CI:0.593~0.599)相比,基于组学特征和临床参数的诺模图联合预测模型(CI:0.641;95%CI:0.610~0.672)的预测能力得到一定程度的提高(P=0.04);高危组和低危组的中位生存期分别为12个月和17个月,2组之间存在显著的统计学差异(P<0.000 1)。结论:基于组学特征和临床参数的联合诺模图能客观地预测局限期小细胞肺癌的总生存期,为患者的个性化治疗奠定前期基础。

Objective To predict the overall survival of limited-stage small cell lung cancer after radiotherapy with multiparameter nomogram.Methods Totally 122 limited-stage small cell lung cancer patients(112 ones of small cell lung cancer and 10 ones of mixed small cell lung cancer)were enrolled in a retrospective study between 2011 to 2018,who were divided into training and validation cohort according to the ratio of 8∶2 randomly.For clinical factors,10-fold cross-validation was performed on the training cohort firstly,and the best subset of parameters was found in all possible subsets of clinical parameters by complete search based on the Akaike information criterion,then a nomogram predictive model was constructed using the Cox proportional risk regression model;for radiomic features,the features were first standardized,and then the radiomic features were screened for predictive value using the least absolute shrinkage and selection operator(LASSO)-Cox proportional risk regression,and finally the radiomic features were combined with the clinical parameters in the best clinical parameter subset to develop a joint multi-parameter nomogram predictive model.The predictive models were evaluated using the Harrell concordance index(CI)to compare the predictive ability of the 2 prediction models for survival in limited-stage small cell lung cancer.The patients were divided into high-risk and low-risk groups using the Rad_score median,and the Kaplan-Meier(KM)survival curves of the high-and low-risk groups were also analyzed and compared using the Log-rank test.Results The combined nomogram predictive model based on radiomic features and clinical parameters(CI:0.641;95%CI:0.610-0.672)gained advantages over the nomogram predictive model based on clinical parameters only(CI:0.596;95%CI:0.593-0.599)in predictive ability(P=0.04);the median survival was 12 and 17 months in the high-risk and low-risk groups, respectively, with statistically significant differences between the 2 groups (P<0.0001). Conclusion Multi-parameter nomogram based on radiomics and clinical parameters can objectively predict the overall survival of limited-stage small cell lung cancer to lay a foundation for personalized treatment of patients.