调强放射治疗肺癌患者放射性肺炎发生的临床预测模型构建

Construction of a clinically predictive model of occurring RP in patients with IMRT

目的:通过分析接受调强放射治疗(IMRT)的肺癌患者发生放射性肺炎(RP)的放射剂量学参数及临床病理特征,建立精确预测RP发生的临床模型。方法:选取医院收治的106例行IMRT的肺癌患者,依照临床表现及影像学检查结果将其分为RP组(44例)和非RP组(62例),分析两组患者与RP相关的剂量学参数和临床病理等特征,并基于全肺总体积的15%、35%、40%(V_(15)、V_(35)、V_(40))、淋巴细胞计数、合并肺部感染、放射治疗剂量、总分期和T分期8个独立预后因素使用机器学习逻辑回归(LR)构建肺癌患者发生RP的临床预测模型(LR预测模型)。采用受试者工作特征(ROC)曲线分析LR预测模型在训练集和验证集中的预测效能。结果:106例患者中发生RP的44例(占41.5%),V_(15)、V_(35)、V_(40)、淋巴细胞计数、合并肺部感染、放射治疗剂量、总分期及T分期为发生RP的独立预后因素。训练集和验证集的ROC曲线下面积(AUC)分别为0.822和0.748,95%CI分别为0.714~0.931和0.488~0.979,模型的校正曲线(Brier score:0.171)和临床决策曲线表明该模型具有较好的预测性和临床有效性。结论:本研究构建的IMRT肺癌患者发生RP的临床LR预测模型,具有较好的预测性能和临床有效性,可为临床预测RP的发生及制定治疗策略提供依据。

Objective:To construct a clinical model that can accurately predict the occurrence of radiation pneumonitis(RP)by analyzing radiation dosimetric parameters and clinically pathological characteristics of patients with lung cancer who occurred RP and received intensity-modulated radiation therapy(IMRT).Methods:A total of 106 patients with lung cancer who underwent IMRT were selected and they were divided into RP group(44 cases)and non-RP group(62 cases)according to clinical manifestations and imaging examination.The characteristic of dosimetric parameters and clinical pathology related to RP of two groups were analyzed.Based on the 8 independently prognostic factors included V_(15),V_(35),V_(40),lymphocyte count,concomitant pulmonary infection,dose of radiation therapy,total staging and T staging,the logistic regression(LR)of machine learning was used to construct the clinically predictive model(LR predictive model)of occurring RP in patients with lung cancer.The receiver operating characteristics(ROC)curve was adopted to analyze the predictive efficiency of LR predictive model in training set and validation set.Results:In the 44 cases(41.5%)who occurred RP of 106 patients,the V_(15),V_(35),V_(40),lymphocyte count,concomitant pulmonary infection,dose of radiation therapy,total staging and T staging were the independently prognostic factors of occurring RP.The area under curves(AUCs)of ROC curve of the training set and the validation set were respectively 0.822(95%CI:0.714-0.931)and 0.748(95%CI:0.488-0.979).The calibration curve(Brier score:0.171)and clinical decision curve of model indicated that this model had favorable predictability and clinical effectiveness.Conclusion:In this study,the constructed clinically LR predictive model for patients with lung cancer who underwent IMRT and occurred RP has favorable predictive predictability and clinical effectiveness,which can provide basis for the clinical prediction of occurring RP and the formulation of treatment strategies.