基于Logistic回归分析的胃癌周围神经侵犯独立危险因素分析以及诊断预测模型建立

Analysis of Independent Risk Factors and Establishment of Diagnostic Prediction Model for Perineural Invasion in Gastric Cancer Based on Logistic Regression Analysis

摘要: 目的：探讨胃癌周围神经侵犯的独立危险因素，同时建立诊断预测模型。方法：收集自2017年01月至2020年12月在延安大学附属医院行胃癌根治术的543例胃癌患者临床病理资料，根据术后病检结果将胃癌患者分为周围神经侵犯阴性组(275例)和周围神经侵犯阳性组(268例)两组。对纳入本研究的临床病理指标进行单因素Logistic回归分析，得到具有统计学意义的指标。将这些指标进行多因素Logistic回归分析，继而得出胃癌周围神经侵犯的独立危险因素，此外应用Hosmer-Lemeshow拟合优度检验拟合效果。建立胃癌周围神经侵犯的诊断预测模型，同时应用ROC曲线和单因素Logistic回归分析验证该模型的诊断价值。结果：单因素Logistic回归分析结果表明体重、BMI、淋巴细胞、单核细胞、红细胞、血红蛋白、血小板、NLR、PLR、LMR、白蛋白、CEA、CA19-9、CA72-4、肿瘤最大直径、肿瘤分化程度、肿瘤组织类型、肿瘤浸润深度、淋巴结转移以及癌结节是具有统计学意义的指标。多因素Logistic回归分析结果提示肿瘤最大直径(P = 0.010, OR: 1.226, 95% CI: 1.049~1.433)、肿瘤分化程度(P < 0.001, OR: 3.000, 95% CI: 1.660~5.423)、肿瘤浸润深度(P < 0.001, OR: 6.323, 95% CI: 2.766~14.453)、淋巴结转移(P < 0.001, OR: 42.884, 95% CI: 11.470~160.338)以及癌结节(P < 0.001, OR: 17.454, 95% CI: 6.759~45.071)是胃癌周围神经侵犯的独立危险因素。此外应用Hosmer-Lemeshow拟合优度检验得知该五项临床病理指标拟合效果较好(χ2 = 7.581, P = 0.475)。根据独立危险因素可以建立胃癌周围神经侵犯的诊断预测模型为：P = ex/(1 + ex)，x = −4.118 + 0.204 × 肿瘤最大直径 + 1.099 × 肿瘤分化程度 + 1.844 × 肿瘤浸润深度 + 3.759 × 淋巴结转移 + 2.860 × 癌结节。ROC曲线的曲线下面积为0.944，渐近95% CI为0.926~0.961。预测模型的诊断临界值为0.5122，该临界值对应的敏感度为0.836，特异性为0.935。此外该诊断模型的预测正确总体百分比87.1%以及单因素Logistic回归分析结果均提示该模型预测准确率较高。结论：肿瘤最大直径、肿瘤分化程度、肿瘤浸润深度、淋巴结转移以及癌结节是胃癌周围神经侵犯的独立危险因素。本研究建立的胃癌周围神经侵犯诊断预测模型其敏感度和特异性均较高，且预测准确率较高，提示该预测模型的诊断价值良好，具有较好的实践应用价值。

Abstract: Objective: To explore the independent risk factors of perineural invasion in gastric cancer, and to establish the diagnostic prediction model. Methods: The clinicopathological data of 543 patients with gastric cancer who underwent radical gastrectomy in the Yan’an University Affiliated Hospital from January 2017 to December 2020 were collected. According to the postoperative pathological examination results, the gastric cancer patients were divided into two groups: the perineural invasion negative group (275 cases) and the perineural invasion positive group (268 cases). Univariate Logistic regression analysis was performed on the clinicopathological indicators included in this study, and statistically significant indicators were obtained. Multivariate Logistic regression analysis was performed on these indicators to determine the independent risk factors for perineural invasion in gastric cancer. In addition, Hosmer-Lemeshow goodness of fit was used to test the fitting effect. The diagnostic prediction model for perineural invasion in gastric cancer was established, and the diagnostic value of the model was verified by ROC curve and univariate Logistic regression analysis. Results: Univariate Logistic regression analysis showed that body weight, BMI, lymphocytes, monocytes, erythrocytes, hemoglobin, platelet, NLR, PLR, LMR, albumin, CEA, CA19-9, CA72-4, maximum tumor diameter, degree of tumor differentiation, tumor tissue type, depth of tumor invasion, lymph node metastasis and tumor deposit were statistically significant indicators. Multivariate Logistic regression analysis showed that maximum tumor diameter (P = 0.010, OR: 1.226, 95% CI: 1.049~1.433), degree of tumor differentiation (P < 0.001, OR: 3.000, 95% CI: 1.660~5.423), depth of tumor invasion (P < 0.001, OR: 6.323, 95% CI: 2.766~14.453), lymph node metastasis (P < 0.001, OR: 42.884, 95% CI: 11.470~160.338) and tumor deposit (P < 0.001, OR: 17.454, 95% CI: 6.759~45.071) were independent risk factors for perineural invasion in gastric cancer. Hosmer-Lemeshow goodness of fit test showed that the five clinicopathological indicators had good fitting effect (χ2 = 7.581, P = 0.475). According to the independent risk factors, the diagnostic prediction model for perineural invasion in gastric cancer could be established as follows: P = ex/(1 + ex), x = −4.118 + 0.204 × maximum tumor diameter + 1.099 × degree of tumor differentiation + 1.844 × depth of tumor invasion + 3.759 × lymph node metastasis + 2.860 × tumor deposit. The area under the ROC curve was 0.944, and the asymptotic 95% CI was 0.926~0.961. The diagnostic cut-off value of the predictive model was 0.5122, which corresponded to a sensitivity of 0.836 and a specificity of 0.935. In addition, the prediction accuracy of this diagnostic model was 87.1% and the results of univariate Logistic regression analysis indicated that the prediction accuracy of this diagnostic model was high. Conclusion: Maximum tumor diameter, degree of tumor differentiation, depth of tumor invasion, lymph node metastasis and tumor deposit are independent risk factors for perineural invasion in gastric cancer. The diagnostic prediction model for perineural invasion in gastric cancer established in this study has high sensitivity and specificity, and high prediction accuracy, suggesting that the prediction model has good diagnostic value and good practical application value.