基于SEER数据库的原发胰腺淋巴瘤预后预测模型的构建

Prognostic prediction model construction of primary pancreatic lymphoma based on SEER database

目的:分析影响原发胰腺淋巴瘤(PPL)预后的因素,构建PPL预后预测模型。方法:收集SEER数据库中1975至2016年的PPL病例,最终纳入830例,以7∶3的比例分为训练集(n=582)和验证集(n=248)。在训练集中,通过Cox比例风险回归模型筛选影响PPL预后的因素,在回归分析结果基础上构建预后预测模型并绘制列线图。分别在训练集和验证集中对模型的预测效能进行内部和外部验证。结果:Cox回归分析结果显示高龄(≥60岁)、未婚、Ann Arbor分期(Ⅳ期)、未手术、未化疗的患者预后差,HR(95%CI)分别为2.823(2.147~3.713)、1.556(1.255~1.929)、1.581(1.206~2.071)、1.531(1.043~2.247)、2.207(1.505~3.237)。将上述变量纳入预测模型并绘制列线图,进行3 a、5 a生存率预测;在训练集和验证集中C指数(95%CI)分别为0.701(0.685~0.715)和0.643(0.592~0.701),训练集中ROC曲线的AUC(95%CI)分别为0.743(0.682~0.804)、0.752(0.674~0.830),验证集中分别为0.727(0.685~0.769)、0.733(0.690~0.776)。预测3 a、5 a生存率的校正曲线显示预测的生存率与实际生存率有着较好的一致性。结论:构建的PPL预后预测模型具有良好的预测价值。

Aim:To analyze the factors affecting the prognosis of primary pancreatic lymphoma(PPL)and to construct a prognostic prediction model.Methods:A total of 830 cases of PPL from 1975 to 2016 in SEER database were collected,and were allocated into training cohort(n=582)and validation cohort(n=248)at a ratio of 7∶3.Factors affecting the survival of PPL patients were screened by Cox proportional risk regression model.A prognostic prediction model of PPL was constructed based on the factors and plotted in a nomogram.Internal and external validation of the model was performed in the training cohort and the validation cohort,respectively.Results:Cox regression analysis showed that the patients with advanced age(≥60 years),unmarried,Ann Arbor stage(stageⅣ),not undergoing surgery,and not receiving chemotherapy had poor prognosis with HR(95%CI)of 2.823(2.147-3.713),1.556(1.255-1.929),1.581(1.206-2.071),1.531(1.043-2.247),2.207(1.505-3.237),respectively.The prediction model and its nomogram were used to predict 3-year and 5-year survival rate,C-index(95%CI)was 0.701(0.685-0.715)and 0.643(0.592-0.701)in the training and validation cohorts,respectively,AUC(95%CI)of ROC curve was 0.743(0.682-0.804)and 0.752(0.674-0.830)in the training cohort,and 0.727(0.685-0.769)and 0.733(0.690-0.776)in the validation cohort,respectively.The 3-and 5-year survival prediction calibration curves showed good agreement between the predicted survival rates and the actual survival rates.Conclusion:The constructed prognostic prediction model has good predictive value for PPL.