胰腺神经内分泌肿瘤预后模型的建立与验证

目的 探讨影响胰腺神经内分泌肿瘤预后的危险因素，构建胰腺神经内分泌肿瘤预后预测模型。方法 选取SEER数据库中2004年1月至2015年12月经病理明确诊断为胰腺神经内分泌肿瘤的患者3606例，按照3∶1分割为训练集（n=2704）和验证集（n=902），在训练集中通过Cox比例风险模型筛选影响胰腺神经内分泌肿瘤预后的危险因素，进一步构建其预后模型并绘制列线图。分别在训练集和验证集中对模型的预测效能进行内部及外部验证。结果 单因素Cox回归分析显示，性别、年龄、婚姻状态、肿瘤部位、分化程度、TNM分期、美国癌症联合委员会（American Joint Committee on Cancer，AJCC）分期、手术均是影响胰腺神经内分泌肿瘤预后的危险因素（P<0.05）；多因素Cox回归分析显示，年龄、性别、婚姻状态、分化程度、TNM分期、手术均是影响胰腺神经内分泌肿瘤预后的危险因素（P<0.05）。最终将年龄、性别、分化程度、肿瘤部位、TNM分期、手术、婚姻状态等变量纳入预测模型并绘制列线图。在训练集和验证集中，预测模型的C指数分别为0.8579和0.8572。训练集和验证集中3年、5年生存率的校准曲线显示，预测生存率与实际生存率存在较好的一致性。结论 构建的胰腺神经内分泌肿瘤预测模型具有良好的预测价值。

Prognostic prediction model construction and validation of pancreatic neuroendocrine tumor

Objective To investigate the risk factors affecting the prognosis of pancreatic neuroendocrine tumors, and to construct a prognosis prediction model of pancreatic neuroendocrine tumors. Methods A total of 3606 patients diagnosed with pancreatic neuroendocrine tumor from January 2004 to December 2015 were selected from the SEER database. They were divided into training set (n=2704) and validation set (n=902) according to the ratio of 3:1. In training set, the Cox proportional hazards model was used to screen the effects of pancreatic neuroendocrine tumors prognostic risk factors, further construct its prognostic model and draw nomogram. The predictive performance of the model was validated internally and externally on the training set and validation set, respectively. Results Univariate Cox regression analysis showed that gender, age, marital status, tumor location, degree of differentiation, TNM stage, American Joint Committee on Cancer (AJCC) stage, and surgery or not were risk factors for the prognosis of pancreatic neuroendocrine tumors (P<0.05). Multivariate Cox regression analysis showed that age, gender, marital status, degree of differentiation, TNM stage and surgery were the risk factors affecting the prognosis of pancreatic neuroendocrine tumors (P<0.05). Finally, variables such as age, gender, degree of differentiation, tumor location, TNM stage, surgery, and marital status were incorporated into the prediction model and nomograms were drawn to predict 3-year and 5-year survival rates. In training set and validation set, the C-index of the prediction model was 0.8579 and 0.8572, respectively. The calibration curves of the 3-year and 5-year survival rates in the training set and the verification set showed that the predicted survival rate was in good agreement with the actual survival rate. Conclusion The constructed pancreatic neuroendocrine tumor prediction model has good predictive value.