胰头癌和远端胆管癌术前列线图诊断模型的构建与验证

Construction and validation of the preoperative nomogram diagnosis model for pancreatic head cancer and distal cholangiocarcinoma

目的探究胰头癌和远端胆管癌术前检验指标的差异性, 构建并验证列线图诊断模型, 为临床决策提供参考。方法回顾性分析2015年1月1日至2019年12月31日解放军总医院第一医学中心连续收治的接受开腹根治性胰十二指肠切除术的患者243例, 其中包括177例男性和66例女性, 年龄(58.9±8.9)岁。根据术后病理结果, 将243例患者分为胰头癌组113例和远端胆管癌组130例。收集两组患者的临床资料, 采用最小绝对值收敛和选择算子回归方法筛选出胰头癌的最佳预测指标, 运用rms程序包构建列线图模型, 以k折交叉进行内部验证。结果筛选出7个胰头癌与远端胆管癌鉴别诊断的指标:年龄、体重下降速率、主胰管直径>3 mm、中性粒细胞/淋巴细胞比值、直接胆红素/总胆红素比值、癌胚抗原、肿瘤糖类抗原125。基于上述指标构建列线图预测模型, k折交叉验证的C指数为0.868, 表明模型有较好的区分度;校准曲线显示, 预测概率与理想概率之间具有良好的一致性(P=0.728)。结论本研究建立的列线图预测模型有助于术前对影像学难以区分的胰头癌和远端胆管癌进行鉴别诊断。

Objective To study the differences in preoperative examination indexes between pancreatic head cancer and distal bile duct cancer,and to establish a preliminary prediction model to provide reference for clinical decision-making.Methods Retrospective analysis was conducted on 243 consecutive patients who underwent open radical pancreaticoduodenectomy from January 1,2015 to December 31,2019 at the Chinese PLA General Hospital.There were 177 males and 66 females,aged(58.9±8.9)years old.Based on postoperative pathological results,these 243 patients were divided into the pancreatic head cancer group(n=113)and the distal bile duct cancer group(n=130).The clinical data of the two groups were collected.Minimum absolute value convergence and selection operator regression were used to screen the best predictors of pancreatic head cancer.The rms package was used to construct the nomogram model,and k-fold cross was used for internal validation.Results Seven best predictive indexes are selected:age,rate of weight loss,main pancreatic duct diameter,neutrophil/lymphocyte,DTR(DBil/TBil),carcinoembryonic antigen and CA125.Based on these indexes,a nomogram prediction model was constructed with the C-index of 0.868 after k-fold cross-validation,indicating that discrimination of the model to be acceptable.Validation using calibration curve exhibited good concordance between the predicted probability with the ideal probability(P=0.728).Conclusion The nomogram prediction model established in this study effectively predicted patients with pancreatic head cancer,and helped preoperatively to differentiate pancreatic head cancer from distal cholangiocarcinoma.