胸腔镜下非小细胞肺癌肺叶切除术中转开胸的预测模型构建与验证

Establishment and Validation of the Prediction Model for Conversion from Thoracoscopic Lobectomy to Thoracotomy for Non-Small Cell Lung Cancer

目的:建立胸腔镜下肺叶切除术的非小细胞肺癌(NSCLC)患者中转开胸的风险预测模型,并验证模型的预测效果。方法:回顾性分析2017年1月至2022年1月我院收治的463例胸腔镜下肺叶切除的NSCLC患者为研究对象,按照7∶3比例随机分为建模队列(n=324)与验证队列(n=139),采用自制病例资料调查表收集患者临床资料,将建模队列按照是否发生中转开胸分为中转开胸组与胸腔镜组,采用二分类Logistic回归确定中转开胸的危险因素,并基于危险因素建立列线图预测模型,采用受试者工作特征曲线评估模型区分度,Hosmer-Lemeshow拟合优度检验、校准曲线评估模型的一致性,绘制决策曲线分析(decision curve analysis, DCA)评估模型的临床应用价值,并将验证队列数据代入进行外部验证。结果:单因素分析结果显示,开胸组与胸腔镜组在年龄、肺结核病史、淋巴结增大情况、胸膜是否粘连、肿瘤位置方面差异有统计学意义(χ^(2)=6.604,P=0.010;χ^(2)=12.543,P <0.001;χ^(2)=8.501,P=0.014;χ^(2)=5.652,P=0.017;χ^(2)=10.462,P=0.001);多因素分析结果显示,有肺结核病史、肺门淋巴结增大、胸膜粘连、肿瘤位于肺上叶是中转开胸的独立危险因素(P=0.003、<0.001、0.002、0.023、<0.001);基于多因素分析结果建立的列线图预测模型,ROC曲线下面积为0.838(95%CI:0.761~0.915),最大约登指数所对应的临界值0.23,即预测概率> 23%(对应总分> 130分)的患者为中转开胸的高危人群;Bootstrap法抽样1 000次,校准曲线图预测的中转开胸风险与实际发生风险高度一致,Hosmer-Lemeshow拟合优度检验χ^(2)=3.447,P=0.841;验证队列的校准曲线图显示预测的中转开胸风险与实际发生风险一致性较好,ROC曲线下面积为0.800(95%CI:0.709~0.890)。DCA显示预测模型的阈概率为0.1~0.95,模型表现为正的净收益。结论:通过术前病史及术前CT检查指标确立的NSCLC患者胸腔镜下肺叶切除术中转开胸的风险预测模型具有良好的预测效能与临床应用价值,列线图的可视化展示形式有助于医护人员快捷、方便地筛选出中转开胸的高危人群,为临床决策提供支持。

Objective : To establish and validate the model predicting the risks in conversion from thoracoscopic lobectomy to thoracotomy for non-small cell lung cancer(NSCLC) patients. Methods : 463 NSCLC patients who underwent thoracoscopic lobectomy in our hospital from January 2017 to January 2022 were selected as subjects for retrospective analysis, then, assigned to the modeling queue(n = 324) and the validation queue(n = 139) with a ratio of 7 : 3. The patients’ clinical data were collected by the self-made case questionnaire in this study. The modeling queue was, further, divided into two groups(the thoracotomy group and the thoracoscopy group) based on whether thoracoscopic lobectomy was converted to thoracotomy intraoperatively. Binary Logistic regression was utilized to determine the risk factors for conversion to thoracotomy, and a nomogram prediction model was established based on these risk factors. The receiver operating characteristic curve was used to evaluate the model discrimination. The Hosmer Lemeshow test and calibration curves were used to evaluate the consistency of the model. The decision curve analysis(DCA) was used to evaluate the clinical application value of the model, and the validation queue data was into the model for external validation. Results : Univariate analysis showed that there were statistically significant differences between the thoracotomy group and the thoracoscopic group in terms of age, tuberculosis history, lymph node enlargement, pleural adhesion and tumor location (χ^(2) = 6.604, P = 0.010;χ^(2) = 12.543, P < 0.001;χ^(2) = 8.501, P = 0.014;χ^(2) = 5.652, P = 0.017;χ^(2) = 10.462, P = 0.001). Multivariate analysis showed that the history of tuberculosis, hilar lymph node enlargement, pleural adhesion and tumor located in the upper lobe of the lung were independent risk factors for conversion to thoracotomy(P = 0.003, P < 0.001, P = 0.002, P = 0.023). The nomogram prediction model based on the results of multivariate analysis showed that the area under the curve was 0.838(95% CI: 0.761-0.915). The critical value corresponding to the maximum Yodon index was 0.23, indicating that more than 23% patients(corresponding to the total score of over 130 points) were predicted to be the high-risk group for conversion to thoracotomy. The Bootstrap method was then used to sample 1,000 times. We found that the predicted risk of conversion to thoracotomy evaluated by the calibration curve was highly consistent with the actual risk(Hosmer Lemeshow test: χ^(2) = 3.447, P = 0.841). The calibration curve of the validation queue also showed that the predicted risk of conversion to thoracotomy was consistent with the actual risk, and the area under the curve was 0.800(95% CI: 0.709-0.890). The DCA results showed that the threshold probability of the prediction model was 0.1-0.95, and the model shows positive net proceeds. Conclusion: The risk prediction model for conversion from thoracoscopic lobectomy to thoracotomy in NSCLC patients, which was established with preoperative medical history and CT indicators, has good efficiency and clinical application value. In addition, the visualization of nomogram is helpful for medical staff to quickly and conveniently screen out high-risk groups of conversion to thoracotomy, so as to formulate effective clinical strategies.