高龄胃肠癌患者术后并发症的 危险因素：多因素分析与 机器学习相结合的多中心 嵌套病例对照研究

Risk Factors for Postoperative Complications in Elderly Patients with Gastric and Colorectal Cancer: A Multicenter Nested Case-Control Study Combining Multivariate Ananlysis and Machine Learning

摘要: 目的：本研究通过多因素分析和机器学习确定老年胃癌和结直肠癌术后并发症的危险因素。方法：对2213例老年患者(371例有并发症，1842例无并发症)的临床资料进行回顾性分析。单变量分析比较基线特征。Boruta算法确定了潜在的核心特征，然后将其包括在多变量Logistic回归模型中。SHAP (SHapley加性解释)分析量化变量重要性，并按肿瘤类型进行分层分析。结果：单因素分析发现并发症与肿瘤类型、性别、合并症、新辅助化疗、ASA分级、术中出血量和手术类型等变量显著相关(P < 0.05)。Boruta和SHAP的分析强调新辅助化疗、合并症和术中失血是核心危险因素。多因素回归分析显示：新辅助化疗(OR = 1.71)、合并症(OR = 2.34)、术中出血(OR = 1.002)、ASA分级2级(OR = 1.49)是独立关联因素。失血量为78毫升的患者风险增加(OR = 1.43)。亚组分析显示，合并症(OR = 4.35)、新辅助化疗(OR = 1.68)、术中出血(OR = 1.002)是胃癌的主要危险因素，ASA分级接近2级。新辅助化疗(OR = 1.60)、合并症(OR = 1.51)、术中出血(OR = 1.001)、ASA分级2级(OR = 1.44)也是结直肠癌的独立危险因素。结论：新辅助化疗、合并症和术中出血是老年胃肠道肿瘤患者并发症的核心、可改变的危险因素，其影响因肿瘤类型而异。建议个体化治疗，以降低风险，改善预后。

Abstract: Objective: This study aimed to identify risk factors for postoperative complications in elderly patients with gastric and colorectal cancer through multivariate analysis and machine learning. Methods: A retrospective analysis was conducted on clinical data from 2213 elderly patients (371 with complications and 1842 without complications). Baseline characteristics were compared using univariate analysis. The Boruta algorithm identified potential core features, which were subsequently included in a multivariate Logistic regression model. SHAP (SHapley Additive exPlanations) analysis quantified variable importance, with stratified analysis performed by tumor type. Results: Univariate analysis revealed that complications were significantly associated with tumor type, sex, comorbidity status, neoadjuvant chemotherapy, ASA classification, intraoperative blood loss, and surgery type (P < 0.05). Boruta and SHAP analyses emphasized neoadjuvant chemotherapy, comorbidity status, and intraoperative blood loss as core risk factors. Multivariate regression analysis showed that neoadjuvant chemotherapy (OR = 1.71), comorbidity status (OR = 2.34), intraoperative blood loss (OR = 1.002), and ASA grade II (OR = 1.49) were independently associated factors. Patients with blood loss of 78 mL showed increased risk (OR = 1.43). Subgroup analysis demonstrated that comorbidity status (OR = 4.35), neoadjuvant chemotherapy (OR = 1.68), and intraoperative blood loss (OR = 1.002) were major risk factors for gastric cancer, with ASA grade approaching II. Neoadjuvant chemotherapy (OR = 1.60), comorbidity status (OR = 1.51), intraoperative blood loss (OR = 1.001), and ASA grade II (OR = 1.44) were also independent risk factors for colorectal cancer. Conclusion: Neoadjuvant chemotherapy, comorbidity status, and intraoperative blood loss are core, modifiable risk factors for complications in elderly patients with gastrointestinal tumors, with their impact varying by tumor type. Individualized treatment is recommended to reduce risk and improve prognosis.