一种预测儿童扁桃体切除术后不良行为改变的列线图模型

A nomogram model for predicting adverse behavior changes after tonsillectomy in children

目的建立一种儿童扁桃体切除术后不良行为改变的预测模型。方法选取2013年7月至2021年12月在天门市第一人民医院耳鼻咽喉头颈外科就诊的2368例患者作为研究数据,按照住院时间顺序分为建模组和验证组。所有患者均行扁桃体切除术,对患者进行住院后行为问卷进行评估,分为不良行为改变组和无不良行为改变组,对建模组采用Lasso回归和随机森林筛选自变量,构建不同的模型A和模型B,选择赤池信息准则(akaike information criterion,AIC)最小为最优模型,通过构建的模型建立列线图,从区分度、校准度和临床适用度3个方面对模型进行评估,同时应用验证组数据评估模型的预测效能。结果2368例患者一次性接受扁桃体手术,术后不良性格改变发生率约为17.06%(404/2368),模型A的AIC较小为1257.736,最终选择模型A为最终模型,居住地(OR=3.875)、赡养(OR=0.003)、术前合并分泌性中耳炎(OR=22.483)、术前白细胞(OR=1.898)、性格类型(OR=1.985)是扁桃体切除术后不良行为改变发生的危险因素,Y=-1.946+1.355×居住地-5.766×赡养+3.113×术前合并分泌性中耳炎+1.376×术前白细胞+0.685×性格类型。当特异性为75.6%,敏感性为56.7%,此时约登指数最大为0.323,对应的P值为0.173,根据P=1/(1+e^(-y)),可计算出Y=-1.56,根据多因素Logistic回归模型构建列线图,建模组AUC:0.73(95%CI:0.69~0.76),Hosmer-Lemeshow拟合优度检测显示χ^(2)=8.24,P=0.51;验证组AUC:0.71(95%CI:0.66~0.77),Hosmer-Lemeshow拟合优度检测显示χ^(2)=5.65,P=0.77,结果表明建模组和验证组具有良好的临床适用性。结论本模型可为临床医师术前评估儿童扁桃体手术术后不良行为的发生提供一定的参考。

Objective Establishing a predictive model for adverse behavior changes after tonsillectomy in children.Methods A total of 2368 patients who visited the Department of Otorhinolaryngology Head and Neck Surgery at the First People's Hospital of Tianmen City from July 2013 to December 2021 were selected as research sample.They were divided into a modeling group and a validation group in chronological order of hospitalization.All patients underwent tonsillectomy,and their behavior questionnaire was evaluated after hospitalization.The patients were divided into two groups:a group with adverse behavior changes and a group without adverse behavior changes.Lasso regression and random forest screening were used to select independent variables for the modeling group to construct different models A and B.The minimum Akashi information criterion was selected as the optimal model to establish a column chart through the constructed model.The model from three aspects of discrimination,calibration,and clinical applicability were evaluated.At the same time,data from the verification group were used to evaluate the prediction efficiency of the model.Results 2368 patients underwent a one-time tonsillectomy and the incidence of adverse personality changes after surgery was approximately 17.06%(404/2368).Model A had a small Akashi information criterion(AIC=1257.736),and Model A was ultimately selected as the final model,with residence(OR=3.875),maintenance(OR=0.003),preoperative concomitant secretory otitis media(OR=22.483),preoperative white blood cells(OR=1.898),personality type(OR=1.985)were the risk factors for adverse behavior changes after tonsillectomy,Y=-1.946+1.355xResidence-5.766xMaintenance+3.113xPreoperative concomitant secretory otitis media+1.376xPreoperative white blood cells+0.685xPersonality type.When the specificity was 75.6%and the sensitivity was 56.7%,the maximum Youden index was 0.323,corresponding to a P value of 0.173.Based on P=1/(1+e^(-y)),Y=-1.56 can be calculated.Based on a multivariate Logistic regression model,a column chart was constructed.In the modeling group:the AUC was 0.73(95%CI:0.69-0.76),and the Hosmer Limeshow goodness of fit test showed that χ^(2)=8.24,P=0.51;In the validation group,the AUC was 0.71(95%CI:0.66-0.77),and the Hosmer Limeshow goodness of fit test showed χ^(2)=5.65,P=0.77.The modeling and validation groups had good clinical applicability.Conclusions This model can provide a reference for clinician to evaluate the occurrence of adverse behavior after tonsillectomy in children before surgery.