联合应用NLR、V20和Dmean预测肺癌放射性肺炎发生——一项外部验证研究

Combined use of NLR, V20 and Dmean to predict radiation-induced lung injury in lung cancer patients:an external validation study

目的用已建立的中性粒细胞、淋巴细胞绝对值比值(NLR)联合双肺V20及双肺Dmean预测肺癌放疗≥3级放射性肺炎模型,对另外一批在我院接受放疗的肺癌患者进行外部验证。方法选择参与建立模型的166例肺癌患者作为内部验证集,选择2016-2018年间在我院接受调强放疗的85例肺癌患者作为外部验证集。比较2个验证集患者≥3级放射性肺炎的发生率;对NLR、V20、Dmean行Logistic多因素预测分析;利用ROC曲线分析联合应用NLR、V20和Dmean预测放射性肺炎的区分度;利用Hosmer-Lemeshow检验来评价预测模型的校准度。结果内部验证集和外部验证集患者≥3级放射性肺炎发生率分别为23.8%和22.9%。多因素预测分析结果显示内部验证集中NLR、V20、Dmean均有统计学意义(P=0.032、0.006、0.005);外部验证集中仅V20具有统计学意义(P=0.038)。模型预测放射性肺炎区分度和校准度在内部验证集和外部验证集均相近(P>0.05)。内部验证集NLR、V20、Dmean及三者联合预测的曲线下面积值分别为0.611、0.646、0.682和0.775,外部验证集相应值分别为0.544、0.702、0.658和0.754。内部验证集校准度P=2.797、0.834。外部验证集的P=2.452、0.653。结论联合应用NLR、V20和Dmean的预测模型可以较好地预测肺癌放疗患者≥3级放射性肺炎的发生,且在外部数据集中得到了验证。

Objective To externally validate the accuracy of combined use of neutrophil-lymphocyte ratio (NLR),V20,and Dmean in predicting the incidence of grade Ⅲ or higher radiation-induced lung injury (RILI) in lung cancer patients.Methods A total of 166 lung cancer patients,who participated in the model establishment were selected into the internal validation group,and 85 lung cancer patients who received intensity-modulated radiotherapy in our department between June 2016 and June 2018 were assigned into the external validation group.The incidence rate of grade 3 or higher RILI was statistically compared between the internal and external validation groups.Multivariate logistic analysis was performed for NLR,V20 and Dmean.The discrimination degree of the predictive model was evaluated by using ROC curve in combination with NLR,V20 and Dmean.The calibration degree of the predictive model was assessed by Hosmer-Lemeshow test.Results The incidence rate of grade 3 or higher RILI in the internal and external validation groups was 23.8% and 22.9%.Multivariate logistic analysis demonstrated that NLR,V20 and Dmean significantly differed in the internal validation group (P=0.032,0.006 and 0.005).However,only V20 significantly differed in the external validation group (P=0.038).The discrimination and calibration degree of RILI was almost consistent between the internal and external validation groups (both P>0.05).The area under the curve (AUC) predicted by NLR,V20,Dmean and the combination of three indexes were 0.611,0.646,0.682 and 0.775 in the internal validation group,and 0.544,0.702,0.658 and 0.754 in the external validation group,respectively.The calibration degree in the internal validation group was P=2.797 and 0.834,P=2.452 and 0.653 in the external validation group.Conclusion Combined application of NLR,V20 and Dmean can accurately predict the incidence of grade Ⅲ or higher RILI in lung can cancer patients,which has been validated by external dataset.