决策树联合Logistic回归模型在结肠腺瘤性息肉危险因素分析中的应用

The application of decision tree model and logistic regression model in analyzing the risk factors for colonic adenomatous polyps

摘要:

目的 通过构建腺瘤性息肉影响因素的多因素一般Logistic回归模型和决策树模型，研究影响结肠腺瘤性息肉发生的危险因素。方法 回顾性分析2019年6月—2024年5月皖南医学院附属铜陵市人民医院533例结肠息肉手术患者的临床资料，依据病理结果分为腺瘤性息肉组（393例）和非腺瘤性息肉组（140例）。收集患者临床资料，包括肠镜检查结果、息肉病理结果、年龄、性别、吸烟史、饮酒史、高血压病史、糖尿病史、甘油三酯、总胆固醇、低密度脂蛋白、尿酸、胆红素、丙氨酸氨基转移酶（ALT）、门冬氨酸氨基转移酶（AST）、碱性磷酸酶（ALP）及γ-谷氨酰转肽酶（GGT）水平等。采用多因素一般Logistic回归模型分析影响结肠腺瘤性息肉发生的危险因素，采用卡方自动交互检测算法构建腺瘤性息肉影响因素的决策树模型，并绘制两种模型的受试者工作特征（ROC）曲线。结果 腺瘤性息肉组年龄≥60岁占比、吸烟史率、饮酒史率、高甘油三酯血症率、高低密度脂蛋白血症率和高尿酸血症率均高于非腺瘤性息肉组（P <0.05）。多因素一般Logistic回归分析结果表明，年龄≥60岁［O^R=1.825（95% CI：1.146，2.907）］、高甘油三酯血症［O^R=1.506（95% CI：1.002，2.264）］和高尿酸血症［O^R=2.023（95% CI：1.193，3.431）］均为结肠腺瘤性息肉发生的危险因素（P <0.05）。决策树模型结果显示，年龄≥60岁、高甘油三酯血症、高尿酸血症是影响结肠腺瘤性息肉发生的危险因素，其中高尿酸血症是最重要的影响因素。ROC曲线结果显示，多因素一般Logistic回归模型预测结肠腺瘤性息肉的曲线下面积（AUC）为0.656，敏感性为58.0%，特异性为67.1%；决策树模型预测结肠腺瘤性息肉的AUC为0.644，敏感性为60.3%，特异性为63.6%，预测效果相近。结论 年龄≥60岁、高甘油三酯血症、高尿酸血症是结肠腺瘤性息肉发生的危险因素，构建的多因素一般Logistic回归模型和决策树模型对结肠腺瘤性息肉的预测具有一定的临床价值。

Abstract:

Objective To investigate risk factors for the development of colorectal adenomatous polyps by constructing a multivariable logistic regression model and a decision tree model for associated factors.Methods A retrospective analysis was conducted on 533 patients who underwent colonic polypectomy at Tongling People's Hospital between June 2019 and May 2024. Patients were categorized into adenomatous polyp (n = 393) and non-adenomatous polyp groups (n = 140) based on pathological findings. Clinical data were collected, including colonoscopy findings, polyp pathology results, age, sex, history of smoking, alcohol consumption, hypertension, and diabetes mellitus, and levels of triglycerides, total cholesterol, low-density lipoprotein, uric acid, bilirubin, alanine aminotransferase (ALT), aspartate aminotransferase (AST), alkaline phosphatase (ALP), and gamma-glutamyl transpeptidase (GGT). Multivariable logistic regression analysis was employed to identify risk factors influencing the occurrence of colonic adenomatous polyps. The Chi-square automatic interaction detection (CHAID) algorithm was used to construct a decision tree model on risk factors for adenomatous polyps. Receiver operating characteristic (ROC) curves were plotted for both models, with the area under the curve (AUC) calculated.Results The adenomatous polyp group exhibited significantly higher proportions of individuals aged ≥ 60 years, those with history of smoking and alcohol consumption, and those with hypertriglyceridemia, hyperlipidemia, and hyperuricemia compared to the non-adenomatous polyp group (P < 0.05). Multivariable logistic regression analysis revealed that age ≥ 60 years [O^R = 1.825 (95% CI: 1.146, 2.907) ], hypertriglyceridemia [O^R = 1.506 (95% CI: 1.002, 2.264) ], and hyperuricemia [O^R = 2.023 (95% CI: 1.193, 3.431) ] were identified as risk factors for the development of colonic adenomatous polyps (P < 0.05). The decision tree model constructed using the CHAID algorithm indicated that age ≥ 60 years, hypertriglyceridaemia, and hyperuricemia were risk factors for the occurrence of colonic adenomatous polyps, with hyperuricemia being the most significant influencing factor. The results of the ROC curve analysis indicated that the multivariable logistic regression model predicted colonic adenomatous polyps with an AUC of 0.656, a sensitivity of 58.0%, and a specificity of 67.1%, and that the decision tree model predicted colonic adenomatous polyps with an AUC of 0.644, a sensitivity of 60.3%, and a specificity of 63.6%, indicating comparable predictive performance between the two models.Conclusion Age ≥ 60 years, hypertriglyceridemia, and hyperuricemia constitute risk factors for the development of colonic adenomatous polyps. The constructed multivariable logistic regression model and decision tree model demonstrate clinical utility in predicting colonic adenomatous polyps.