D2^(+)根治性胃腺癌切除术后组织Rictor蛋白阳性患者无进展生存预测列线图模型的构建及预测效果分析

Construction and predictive effect analysis of a nomogram model for predicting the progression⁃free survival of patients with positive Rictor protein in tissues of D2^(+)radical gastric adenocarcinoma resection

目的构建预测D2^(+)根治性胃腺癌切除术后病变组织检测Rictor蛋白阳性患者无进展生存(PFS)的列线图模型,并分析其预测价值。方法回顾性收集2005年5月至2020年12月于山西省肿瘤医院接受D2^(+)根治性切除术的1366例胃腺癌患者组织样本,使用免疫组织化学SP法检测Rictor蛋白表达。Rictor阳性676例,按7∶3比例进行简单随机分组,其中训练队列496例,验证队列180例。采用多因素Cox比例风险模型分析Rictor蛋白表达情况及其他临床病理因素与Rictor阳性患者PFS的关系,确定PFS独立影响因素。根据PFS独立影响因素构建预测胃腺癌患者3年和5年PFS率的列线图。进行10000次重复采样的内部交叉验证,检验预测模型的准确度。通过校准曲线、时间受试者工作特征(ROC)曲线下面积(AUC)和决策曲线分析(DCA)进一步评估模型的内部和外部预测能力。应用列线图模型对训练队列496例PFS进行评分,采用X‑tile软件获得评分的最佳临界值,根据最佳临界值将总体队列、训练队列和验证队列病例分为低危组(≤最佳临界值)和高危组(>最佳临界值),采用Kaplan‑Meier法分析各队列低危组和高危组PFS差异。结果多因素Cox回归分析显示,性别、年龄、pT分期、阳性淋巴结数、神经侵犯、肿瘤长径、网膜侵犯、Clavien‑Dindo术后并发症分级、CGA的表达为Rictor阳性胃腺癌训练队列PFS的独立影响因素。依据上述指标构建预测Rictor阳性胃腺癌患者3年和5年PFS率的列线图。内部验证和外部验证的校准曲线显示列线图预测与实际PFS吻合较好。时间ROC曲线显示,内部验证和外部验证模型预测3年PFS率的AUC分别为0.834(95%CI 0.746~0.823)、0.799(95%CI 0.699~0.868),预测5年PFS率的AUC分别为0.817(95%CI 0.718~0.821)、0.795(95%CI 0.675~0.895)。模型整体预测的C指数为0.795(95%CI 0.764~0.825),优于美国癌症联合委员会(AJCC)第8版TNM分期的0.693(95%CI 0.662~0.723);外部验证DCA示,模型预测的C指数为0.769(95%CI 0.718~0.821),优于AJCC第8版TNM分期的0.715(95%CI 0.669~0.760)。X‑tile软件分析示训练队列模型PFS评分的最佳临界值为265.08,所有队列、训练队列和验证队列低危组分别有457、337、120例,高危组分别有219、159、60例。Kaplan‑Meier生存分析显示,所有队列、训练队列和验证队列低危组中位PFS时间均未达到,高危组中位PFS时间分别为24、24、28个月,各队列低危组和高危组间PFS差异均有统计学意义(均P<0.001)。结论首次成功构建了针对Rictor阳性表达的胃腺癌患者PFS预测的列线图模型,该模型能较好区分PFS风险低危和高危患者。对于Rictor表达阳性的高危胃腺癌患者,除了控制肿瘤的转移和术后并发症,还应该注意针对Rictor表达阳性的靶向治疗。

Objective To construct a nomogram model for predicting the progression‑free survival(PFS)of patients with positive Rictor protein tested in lesion tissues of D2^(+)radical gastric adenocarcinoma resection and to analyze its predictive value.Methods The tissue samples of 1366 gastric adenocarcinoma patients who underwent radical resection in Shanxi Province Cancer Hospital from May 2005 to December 2020 were retrospectively collected,and the Rictor protein expression was detected by using the immunohistochemical SP method.The 676 Rictor‑positive cases were grouped in a 7∶3 ratio by simple randomization,including 496 cases in the training cohort and 180 cases in the validation cohort.The correlation of Rictor protein expression and other clinicopathological factors with PFS of Rictor‑positive patients was analyzed by using a multifactorial Cox proportional risk model to determine the independent influencing factors of PFS.The nomogram for predicting the 3‑year and 5‑year PFS rates of patients with gastric adenocarcinoma was constructed based on the independent influencing factors of PFS.The constructed nomogram was bootstrapped with 1000 resamplings for internal validation to test the accuracy of the prediction model.The internal and external predictive efficacy of the model was further assessed by calibration curves,area under the curve(AUC)of time‑dependent receiver operating characteristic(ROC)and decision curve analysis(DCA).The nomogram model was applied to score the PFS of 496 cases in the training cohort,and the X‑tile software was used to obtain the best cut‑off value for the score.The overall cohort,training cohort and validation cohort cases were divided into low‑risk group(≤best cut‑off value)and high‑risk group(>best cut‑off value)according to the best cut‑off value,and the Kaplan‑Meier method was used to analyze the difference in PFS between the low‑risk group and high‑risk group.Results Multifactorial Cox regression analysis showed that gender,age,pT stage,number of positive lymph nodes,neural invasion,tumor longest diameter,omental invasion,Clavien‑Dindo classification of postoperative complications,and CGA expression were independent influencing factors for PFS of the training cohort with Rictor‑positive gastric adenocarcinoma.The nomogram for predicting the 3‑year and 5‑year PFS rates of patients with Rictor‑positive gastric adenocarcinoma was constructed based on the above indicators.The calibration curve for internal validation and external validation showed good agreement between the prediction of nomogram and actual PFS.The time‑ROC curve showed that the AUC of the internally validated and externally validated models for predicting the 3‑year PFS rate was 0.834(95%CI 0.746-0.823)and 0.799(95%CI 0.699-0.868),and the AUC for predicting the 5‑year PFS rate was 0.817(95%CI 0.718-0.821)and 0.795(95%CI 0.675-0.895).The C index of the model for overall prediction was 0.795(95%CI 0.764-0.825),which was better than the 8th edition of American Joint Committee on Cancer(AJCC)TNM staging[0.693(95%CI 0.662-0.723)];the external validation DCA showed that the C index of the model for prediction was 0.769(95%CI 0.718-0.821).The X‑tile software analysis showed that the best cut‑off value for the PFS score of the training cohort model was 265.08,with 457,337 and 120 cases in the low‑risk group and 219,159 and 60 cases in the high‑risk group for the overall cohort,training cohort and validation cohort,respectively.Kaplan‑Meier survival analysis showed that the median PFS time was not reached in the low‑risk group for the overall cohort,training cohort and validation cohort,and the median PFS time was 24,24 and 28 months in the high‑risk group,and there were statistical differences in PFS between the low‑risk and high‑risk groups for each cohort(all P<0.001).Conclusions For the first time,a nomogram model for PFS prediction in gastric adenocarcinoma patients with Rictor‑positive expression is successfully constructed,which could better distinguish between patients with low‑risk and high‑risk of PFS.For high‑risk patients with Rictor‑positive gastric adenocarcinoma,in addition to controlling tumor metastasis and postoperative complications,attention should be paid to the targeted therapy for positive expression of Rictor.