胃神经内分泌肿瘤患者一年内发生进展预测风险模型的建立及其预测价值

Establishment of risk model for predicting the progression within 1 year of patients with gastric neuroendocrine neoplasms and its value of prediction

目的建立胃神经内分泌肿瘤（gNEN）1年内肿瘤发生进展的预测风险模型并探讨其预测价值。方法回顾性收集1999年1月至2015年1月间南方医科大学附属南方医院（63例）和中山大学附属第一医院（64例）收治的共计127例gNEN患者的临床资料，剔出25例失访病例，共102例纳入分析。根据患者确诊gNEN后1年内出现肿瘤增大、数目增加、原发灶切除后复发以及肿瘤转移等定义为肿瘤进展。采用Logistic回归分析判断gNEN患者在1年内发生进展的影响因素，求得回归方程，得到gNEN患者1年内肿瘤发生进展的概率并进行预测分级：Ⅰ级进展概率〈25．0％，Ⅱ级为25．0％-50．0％，Ⅲ级为50．0％～75．0％，Ⅳ级≥75．0％。Spearman相关分析探讨预测分级与gNEN患者1年内是否发生疾病进展的相关性。绘制不同预测方法对肿瘤1年内发生进展的ROC曲线，并比较曲线下面积（AUC）、灵敏度及特异度。结果102例gNEN患者确诊后1年内出现肿瘤进展者56例（进展组），无进展46例（无进展组）。多因素Logistic回归分析结果显示，肿瘤大小（OR=1．048，95％CI：1．014-1．083，P=0．005）、Ki-67指数（OR=2．045，95％CI：1．261～3．316，P=0．004）以及是否手术切除原发病灶（OR=0．074．95％CI：0．011～0．497，P=0．070）是影响本组gNEN患者1年内有无进展的独立危险因素；得到回归方程：P（Y）=1／[1＋e-（-0．934＋0．047a＋0．715b-2．597c）]（a为肿瘤大小，b为Ki-67指数，c为是否手术切除原发灶）。按照回归方程进行预测分级：Ⅰ级28例（29．2％），Ⅱ级9例（9．4％），Ⅲ级24例（25．0％），Ⅳ级35例（36．5％）；Ⅰ、Ⅱ、Ⅲ、Ⅳ级患者1年内发生肿瘤进展的概率分别为10．7％（3／28）、5／9、58．3％（14／24）和91．4％（32／35），差异有统计学意义（χ2=41．236，P：0．000）；预测分级与gNEN患者1年内是否发生肿瘤进展呈正相关（r=0．644，P=0．000）。ROC曲线显示，预测分级预测gNEN患者1年内肿瘤进展的AUC为0．857，灵敏度为85．2％，特异度为69．0％。采用DeLong法比较预测分级与Ki-67分级和TNM分期的AUC，结果显示，预测分级对1年内肿瘤进展的预测价值与TNM分期差异无统计学意义（P=0．303）；显著优于Ki-67分级预测方式（P=0．006）。结论预测风险分级模型可较为准确地预测gNEN患者1年内肿瘤发生进展的风险，分级越高，表示患者1年内肿瘤发生进展的可能性越大，复诊频率应越高。

Objective To establish the risk model for predicting the progression within 1 year of patients with gastric neuroendocrine neoplasms（gNEN） and to evaluate its value of prediction. Methods Clinical data of 127 gENE patients with histologically comfirmed sporadic gNEN from January 1999 to February 2015 in Nanfang Hospital of Southern Medical University （n = 63） and The First Affiliated Hospital of Sun Yat-sen University （n = 64） were collected retrospectively. Twenty-five patients without follow-up were excluded, so a total of 102 cases were enrolled in the analysis. Tumor size enlargement, lesion number increase, recurrence after resection of primary tumor and emergence of tumor metastasis were defined as tumor progression. Patients were divided into progression group （above definitions occurred within 1 year, n = 56） and non-progression group （above definitions did not occur within 1 year, n = 46）. Logistic regression analysis was used to identify the influencing factors of progression within 1 year and the regression equation was acquired, then the probability of progression within 1 year of gNEN patients was obtained to predict the grading： grade Ⅰ ： the probability of tumor progression within 1 year was 〈 25.0%; grade Ⅱ： this probability was from ≥25.0% to 〈 50.0%; grade m ： this probability was from ≥ 50.0% to 〈75.0%; grade Ⅳ：this probability was ≥ 75.0%. Spearman correlation analysis was used to study the correlation between predictive grading and the occurrence of disease progression in patients with gNEN NET within 1 years. The ROC curve of different prediction methods was drawn, then the area under the curve （AUC）, sensitivity and specificity were calculated and compared. Results Multivariate regression analysis showed that tumor size （OR = 1.048, 95%CI： 1.014-1.083,P=0.005）, Ki-67 index （OR =2.045, 95%CI： 1.261-3.316, P = 0.004）, and surgical resection of the primary lesion （OR = 0.074, 95%CI：0,011-0.497, P = 0.070） were independent influencing factors of the progression of gNET within 1 year. The regression equation was as below： P （Y）=1/[1＋e-（-0.934＋0.047a＋0.715b-2.597c）]. （a： tumor size, b： Ki-67 grading, c： undergoing the surgical resection of the primary lesion）. Prediction grading was based on regression equation： 28 cases （29.2%） belonged to grade Ⅰ , 9 cases （9.4%） to grade Ⅱ, 24 cases （25.0%） to grade Ⅲ, and 35 cases （36.5%） to grade Ⅳ. The probability of progression within 1 year of patients in grade Ⅰ , Ⅱ,Ⅲand Ⅳwas 10.7% （3/28）,5/9, 58.3%（14/24）, and 91.4%（32/35） respectively, with significant difference （χ2 = 41.236, P = 0.000）. Prediction grading was positively correlated with the occurrence of tumor progression in gNEN patients within 1 year （r = 0.644, P = 0.000）. The AUC of prediction grading to predict the progression of gNEN within 1 year was 0.857, while the sensitivity was 85.2% and the specificity was 69.0%. DeLong method was used to compare the AUC values of prediction grading, Ki-67 grading and TNM staging. Comparison result revealed that the predictive value of prediction grading was not significantly different with that of TNM staging （P= 0.303）, but was better than that of Ki-67 grading （P = 0.006）. Conclusions Our grading standard can objectively and accurately reflect the probability of progression within one year in gNEN patients. The later the grading, the higher the probability of progression within 1 year is for gNEN patients.