住院患者多重耐药菌肺部感染的危险因素分析及列线图模型的构建

Risk factors analysis and nomogram model construction for multidrug-resistant organisms pulmonary infection in hospitalized patients

目的探讨住院患者肺部多重耐药菌(multidrug-resistant organisms,MDROs)感染风险预测模型的构建及模型预测效能的评价。方法收集肺部感染住院患者380例作为建模组(MDROs组139例、非MDROs组241例),150例作为外部验证组(MDROs组47例、非MDROs组103例)。以多因素Logistic回归分析,筛选住院患者MDROs发生的独立危险因素,将有统计学意义的因素纳入列线图预测模型的构建并对其进行多中心验证。结果Logistic回归分析结果显示,长期卧床、呼吸衰竭、重症肺炎、住院时长≥2周、吸痰、气管插管、留置导尿管、留置引流管是住院患者MDROs肺部感染的独立危险因素。基于独立危险因素构建列线图模型,建模组和验证组AUC值分别为0.840、0.897;且Hosmer-Lemeshow检验表明模型拟合差异无统计学意义(P>0.05),校准曲线斜率接近1,校准曲线的预测概率与实际概率相近,说明预测模型有较好的校准能力。结论构建住院患者肺部MDROs感染风险预测模型具有一定的临床意义,能够针对可控因素进行早期个体化干预。

Objective To explore the construction of a risk prediction model for multidrug-resistant organisms(MDROs)pulmonary infection in hospitalized patients and evaluate the predictive efficiency of the model.Methods A total of 380 hospitalized patients with pulmonary infections were collected as the modeling group(139 in the MDROs group and 241 in the non-MDROs group),and 150 patients were included as the external validation group(47 in the MDROs group and 103 in the non-MDROs group).Multivariate Logistic regression analysis was conducted to screen for independent risk factors for MDROs in hospitalized patients,statistically significant factors were included in the construction of a nomogram chart prediction model and subjected to multicenter validation.Results Logistic regression analysis revealed that long-term bed rest,respiratory failure,severe pneumonia,hospitalization duration≥2 weeks,sputum aspiration,tracheal intubation,indwelling urinary catheter,and indwelling drainage tube were independent risk factors for MDROs pulmonary infection in hospitalized patients.A nomogram model was constructed based on these independent risk factors,with AUC values of 0.840 and 0.897 for the modeling and validation groups,respectively.The Hosmer-Lemeshow test showed no statistically significant difference in model fit(P>0.05).The slope of the calibration curve was close to 1,and the predicted probabilities on the calibration curve were similar to the actual probabilities,indicating good calibration ability of the prediction model.Conclusion The construction of a risk prediction model for MDROs pulmonary infection in hospitalized patients has certain clinical significance and can facilitate early individualized interventions targeting controllable factors.