COPD稳定期患者急性加重的危险因素及风险预测模型构建

Risk factors of acute exacerbation and construction of risk prediction model in patients with stable COPD

目的分析慢性阻塞性肺疾病(chronic obstructive pulmonary disease,COPD)稳定期患者急性加重的危险因素,并建立风险预测模型,分析模型的临床价值,为临床早期预测及管理提供参考。方法回顾性研究,选取2019年8月—2023年8月在焦作煤业(集团)有限责任公司中央医院收治的COPD稳定期患者372例,按照收治时间将其分为模型组(n=279)和验证组(n=93),根据6个月内是否出现急性加重又将模型组分为加重组(n=121)和非加重组(n=158),收集患者临床资料,单因素和多因素logistic回归分析影响COPD稳定期患者急性加重的危险因素,并根据危险因素构建风险预测模型,受试者工作特征(receiver operating characteristic,ROC)曲线对模型的预测效能进行检测,Hosmer-Lemeshow检验对模型的拟合度予以评估,验证组进行模型验证。结果模型组和验证组临床基线资料和实验室指标对比差异均无统计学意义(P>0.05)。加重组体质量指数、1秒用力呼气容积(forced expiratory volume in one second,FEV1)低于非加重组,近一年内急性加重的次数、近一年内急性加重的持续时间、吸烟史比例、CAT评分及C反应蛋白(C-reactive protein,CRP)均高于非加重组(P<0.05)。多因素logistic回归分析结果显示影响COPD稳定期患者急性加重的独立危险因素为体质量指数、FEV1降低、近一年内急性加重的次数及CRP增加(P<0.05)。Logistic回归分析预测模型预测概率P=1/{1+e^([80.903-4.644×(体质量指数)+4.925×(近一年内急性加重的次数)-2.433×(FEV_(1))+2.647×(CRP)])};ROC曲线检验模型效能曲线下面积(area under the curve,AUC)为0.853,敏感度为80.17%,特异度为89.87%(P<0.05);Hosmer-Lemeshow拟合优度检验χ^(2)=7.966,P=0.437。利用验证组数据进行验证,验证组AUC为0.846,敏感度为81.72%,特异度为81.00%(P<0.05)。结论体质量指数、FEV1、近一年内急性加重的次数及CRP均为影响COPD稳定期患者急性加重的独立危险因素,通过上述危险因素构建logistic回归模型可以有效预测COPD稳定期患者急性加重,并可通过该风险预测模型进行有效干预。

Objective To analyze the risk factors of acute exacerbation in patients with stable chronic obstructive pulmonary disease(COPD),construct the risk prediction model and analyze clinical value of the model in order to provide reference for early clinical prediction and management.Methods In the retrospective study,a total of 372 patients with stable COPD admitted to the Central Hospital of Jiaozuo Coal Industry(Group)Company Limited between August 2019 and August 2023.According to the time of admission,they were divided into model group(n=279)and verification group(n=93).According to presence or absence of acute exacerbation within 6 months,patients in model group were further divided into exacerbation group(n=121)and non-exacerbation group(n=158).The clinical data of patients were collected.Univariate and multivariate logistic regression analysis was conducted to analyze the risk factors affecting acute exacerbation in stable COPD patients,and the risk prediction model was constructed according to the risk factors,and the predictive efficacy of the model was tested by receiver operating characteristic(ROC)curve,and its fit was evaluated by Hosmer-Lemeshow test.The model was verified in verification group.Results There were no significant differences in clinical data or laboratory indexes between model group and verification group(P>0.05).The body mass index and forced expiratory volume in 1 second(FEV_1)in exacerbation group were lower than those in non-exacerbation group,frequency and duration of acute exacerbation in the past year,proportion of smoking history,CAT score and C-reactive protein(CRP)were higher than those in non-exacerbation group(P<0.05).The results of multivariate logistic regression analysis showed that body mass index,FEV_1 reduction,frequency of acute exacerbation in the past year and CRP increase were independent risk factors of acute exacerbation in stable COPD patients(P<0.05).The incidence of acute exacerbation by prediction model was as follow:P=1/{1+e^([80.903-4.644×(body mass index)+4.925×(frequency of acute exacerbation in the past year)-2.433×(FEV_(1))+2.647×(CRP)])}.ROC curve analysis showed that area under the curve(AUC)of the model was 0.853,the sensitivity was 80.17%,and the specificity was 89.87%,respectively(P<0.05).The results of Hosmer-Lemeshow goodness of fit test were as follows:χ^(2)=7.966,P=0.437.According to data from verification group,AUC,sensitivity and specificity in verification group were 0.846,81.72%and 81.00%,respectively(P<0.05).Conclusion The independent risk factors for acute exacerbations in stable COPD patients are decreased body mass index and FEV_1,the number of acute exacerbations in the past year,and increased CRP.Logistic regression model constructed by the above risk factors can effectively predict acute exacerbation,and effective intervention can be taken through the risk prediction model.