Abstract:

Objective　To construct and verify the prediction model of intestinal preparation failure for colonoscopy in elderly patients based on logistic regression and Bayes. Methods　From January 2023 to December 2023, 249 patients with failed bowel preparation for colonoscopy were enrolled from Guangzhou First People's Hospital. Using a 1:1.88 ratio, 467 patients without bowel preparation failure during the same period were selected. They were divided into a test group (501 cases) and a validation group (215 cases) according to the different treatment durations. In the test group, there were 174 cases of failed intestinal preparation, including 104 males and 70 females, with an age of (69.49±7.405) years. Intestinal preparation failed in 327 cases, including 139 males and 188 females, with an age of (68.54±5.779) years. In the verification group, there were 75 cases of failed intestinal preparation, including 36 males and 39 females, with an age of (70.32±8.076) years. Intestinal preparation failed in 140 cases, including 70 males and 70 females who were (67.94±6.115) years old.In this study, the data were collected using the General information Questionnaire and the Boston Bowel Readiness Assessment Scale. The differences in categorical variables and continuous variables between different groups were compared respectively through the χ² test and the mean t test. Subsequently, a multivariate Logistic regression model and a Bayesian model were constructed in the test set and verified using the validation set. Finally, the predictive efficacy of the two models was evaluated through the receiver operating characteristic curve(ROC). Results　From January 2023 to December 2023, a total of 716 patients were admitted to Guangzhou First People's Hospital. Among them, 249 patients failed in colonoscopy and intestinal preparation, with an incidence rate of 34.78%. In the test group, the following indicators in the failed bowel preparation group were significantly lower than those in the qualified bowel preparation group: the sequence of this colonoscopy, the grade of hypertension, the history of abdominal surgery, the detection rate of adenoma, the time of the first defecation after taking laxatives, and the compliance with a low-residue diet the day before the examination (all P<0.05). A total of 716 patients were included in this study, among whom 349 were male (48.7%) and 367 were female (51.3%). The age range was 60 to 95 years old, with an average of 69. Through the multivariate logistic regression analysis, it found that the influencing factors of failed bowel preparation included: the number of colonoscopy examinations this time, the grade of hypertension, the history of abdominal surgery, adenoma, the time of the first defecation after taking laxatives, and whether there was a low-residue diet the day before the examination. The comparison of model performance showed that the AUC of the logistic regression model was 0.908 (95%CI: 0.881-0.934), which was significantly higher than 0.767 (95%CI: 0.701-0.833) of the Bayesian model, indicating that the logistic regression model has better classification performance. Conclusions　The logistic regression model can provide parameter details among various variables, offer detailed scores for the prediction of failed bowel preparation, and can be used for clinical promotion. The Bayesian model is faster when the amount of data is large, and can be used as a clinical assistant.

Key words: Old age,  , Intestinal preparation failure,  , logistic regression,  , Bayes,  , Prediction model

摘要：

目的　logistic回归与贝叶斯方法构建老年患者结肠镜肠道准备失败预测模型及验证研究。方法　选取2023年1月至2023年12月广州市第一人民医院收治的249例肠镜肠道准备失败患者，按1∶1.88的比例选取同期入院治疗的467例肠镜肠道准备未失败患者。根据治疗时间的不同将其分为测试组（501例）和验证组（215例）。测试组肠道准备失败174例，男性104例，女性70例，年龄（69.49±7.405）岁；肠道准备未失败327例，男性139例，女性188例，年龄（68.54±5.779）岁。验证组肠道准备失败75例，男性36例，女性39例，年龄（70.32±8.076）岁；肠道准备未失败140例，男性70例，女性70例；年龄（67.94±6.115）岁。本研究采用一般资料调查表和波士顿肠道准备评价量表收集数据，通过χ2检验和均值t检验分别比较分类变量和连续变量在不同组间的差异；随后在测试集中构建多元Logistic回归模型和贝叶斯模型，并利用验证集进行验证，最后通过受试者操作特征曲线（ROC）评估两种模型的预测效能。结果　2023年1月至2023年12月年广州市第一人民医院共收治716例患者，其中249例为肠镜肠道准备失败患者，发生率为34.78%。在测试组中，肠道准备失败组的以下指标均显著低于肠道准备合格组：本次肠镜检查的序次、高血压等级、腹部手术史、腺瘤检出率、服用泻药后首次排便时间以及检查前一天少渣饮食的依从性（均P<0.05）。本研究共纳入716例患者，其中男性349例（48.7%），女性367例（51.3%），年龄范围为60~95岁，平均年龄69岁。通过多元logistic回归分析发现，肠道准备失败的影响因素包括：本次肠镜检查的序次、高血压等级、腹部手术史、腺瘤、服用泻药后首次排便时间及检查前一天是否少渣饮食。模型性能比较显示，logistic回归模型的AUC值为0.908（95%CI：0.881~0.934），显著高于贝叶斯模型的0.767（95%CI：0.701~0.833），表明logistic回归模型具有更好的分类性能。结论　logistic回归模型能提供各变量间的参数细节，为肠道准备失败预测提供详细的评分，可用于临床推广。贝叶斯模型在数据量大时算法更快捷，可作为临床辅助。

关键词: 老年, 肠道准备失败, logistics回归, 贝叶斯, 预测模型