目的

探讨胆囊肿瘤性息肉发生相关因素，并基于随机森林算法构建胆囊肿瘤性息肉预测模型。

方法

收集2015 年1 月至2023 年8 月在11 家医疗中心行胆囊切除术的745 例胆囊息肉患者临床病理资料。患者均签署知情同意书，符合医学伦理学规定。其中男286 例，女459 例；年龄18～80 岁，中位年龄46 岁。胆囊息肉长径为10～15 mm，中位直径11 mm。胆囊肿瘤性息肉发生相关因素的单因素分析采用χ² 或Mann-Whitney U 检验。根据患者入院时间不同分为训练集（588 例）和测试集（157 例），训练集用于随机森林预测模型的构建，测试集用于预测模型验证。采用ROC 曲线下面积（AUC）及混淆矩阵评估模型的预测能力。

结果

本研究中非肿瘤性息肉者占87.2%（650/745），其中胆固醇息肉518 例，炎性息肉55 例，腺瘤样增生47 例；肿瘤性息肉占12.8%（95/745），其中胆囊腺瘤83 例，T1 期胆囊癌12 例。单因素分析显示，息肉数量、息肉长径、息肉短径、基底情况、息肉部位、回声强度与胆囊肿瘤性息肉发生有关（χ²=20.675，Z=-4.694，Z=-2.595，χ²=6.692，Z=3.935，Z=-2.690；P＜0.05）。基于胆囊肿瘤性息肉发生的危险因素及重要度排序结果构建随机森林预测模型，模型训练集和测试集AUC 分别为0.79、0.69，敏感度分别为0.74、0.63，特异度分别为0.75、0.68。基于胆囊肿瘤性息肉的随机森林预测模型混淆矩阵分析，模型训练集和测试集准确率分别为75%、68%。

结论

胆囊肿瘤性息肉发生与息肉个数、息肉大小、息肉基底情况、息肉部位、回声强度等具有明显相关性，基于随机森林算法构建的预测模型有助于胆囊肿瘤性息肉的识别，为胆囊息肉患者的外科诊疗及随访策略提供决策支持。

Objective

To explore the risk factors of neoplastic gallbladder polyps, and construct a prediction model for neoplastic gallbladder polyps based on random forest algorithm.

Methods

Clinicopathological data of 745 patients with gallbladder polyps who underwent cholecystectomy in 11 medical centers from January 2015 to August 2023 were collected.The informed consents of all patients were obtained and the local ethical committee approval was received.Among them, 286 patients were male and 459 female, aged from 18 to 80 years, with a median age of 46 years.The maximum diameter of gallbladder polyps was ranged from 10 to 15 mm, and the median diameter was 11 mm.Univariate analysis of the risk factors of neoplastic gallbladder polyps was conducted by Chi-square test or Mann Whitney U test.According to the admission date, they were divided into the training set (n=588) and test set (n=157).The training set was used to construct the random forest prediction model, and the test set was utilized to validate the prediction model.The prediction performance of this model was assessed by the area under the ROC curve (AUC) and confusion matrix.

Results

In this study, non-neoplastic gallbladder polyps patients accounted for 87.2%(650/745), including 518 cases of cholesterol polyps, 55 cases of inflammatory polyps and 47 cases of adenomatous hyperplasia.The proportion of neoplastic gallbladder polyps was 12.8%(95/745), including 83 cases of gallbladder adenomas and 12 cases of T1 gallbladder carcinomas.Univariate analysis showed that the number of polyp, maximum and minimum diameter of polyp, polyp short diameter, polyp basal status, polyp location and echo intensity were correlated with the incidence of neoplastic gallbladder polyps (χ²=20.675, Z=-4.694, Z=-2.595, χ²=6.692, Z=3.935, Z=-2.690; P＜0.05).Based on the risk factors of neoplastic gallbladder polyps and the ranking of importance, a random forest prediction model was constructed.The AUC of the training and test sets was 0.79 and 0.69, with a sensitivity of 0.74 and 0.63 and a specificity of 0.75 and 0.68, respectively.Based on the random forest prediction model and confusion matrix analysis of neoplastic gallbladder polyps, the accuracy of the training and test sets was 75% and 68%,respectively.

Conclusions

The incidence of neoplastic gallbladder polyps is significantly correlated with the number of polyp, polyp size, polyp basal status, polyp location and echo intensity, etc.The prediction model based on random forest algorithm contributes to identifying neoplastic gallbladder polyps and providing decision support for surgical diagnosis, treatment and follow-up strategy for patients with gallbladder polyps.