目的 分析患者感染耐碳青霉烯类肺炎克雷伯菌（carbapenem-resistant Klebsiella pneumoniae，CRKP）的独立危险因素，在此基础上构建一种可用于预测个体化风险的列线图模型。 方法 回顾性收集2023年1月至2025年6月在安徽医科大学附属巢湖医院共422例感染肺炎克雷伯菌患者的临床资料，使用R软件按7∶3的比例随机拆分为训练集295例和验证集127例，基于LASSO和logistic回归确立危险因素，构建CRKP感染风险预测模型列线图并分析模型效能。 结果 多因素分析显示，入住ICU（OR = 4.883，95%CI：1.842 ~ 12.946）、接受气管切开/插管（OR = 4.784，95%CI：1.763 ~ 12.981）、吸痰（OR = 2.309，95%CI：0.767 ~ 6.954）、使用碳青霉烯类药物（OR = 4.832，95%CI：1.944 ~ 12.009）以及联合使用抗菌药物（OR = 2.239，95%CI：0.938 ~ 5.341）是患者发生CRKP感染的5项独立危险因素。根据以上训练集中筛选出的5种变量构建预测模型，Logit（P） = -3.632 + ［入住ICU（是=1，否=0）］× 1.586 + ［使用碳青霉烯类药物（是= 1，否= 0）］× 1.575 + ［联合使用抗生素（是= 1，否= 0）］× 0.806 + ［吸痰（是= 1，否= 0）］× 0.837 + ［气管切开/插管（是= 1，否= 0）］× 1.565，训练集ROC曲线下面积（AUC）为0.921（95%CI：0.880 ~ 0.961），验证集AUC为0.914（95%CI：0.846 ~ 0.983）。训练集与验证集校准曲线结果均展示出模型预测概率和实际感染概率具有较好的一致性，Hosmer-Lemeshow（H-L）检验结果显示模型的校准度处于较好水平。DCA曲线进一步证实该模型具有较好的临床实用性。 结论 本研究所构建的风险预测模型能够良好预测感染CRKP的高危患者，可作为判断住院患者CRKP感染风险的工具。

Objective To analyze the independent risk factors for carbapenem-resistant Klebsiella pneumoniae （carbapenem-resistant Klebsiella pneumoniae，CRKP） infection in clinical patients， and further establish a nomogram model applicable for individualized risk prediction. Methods Clinical data of 422 patients with Klebsiella pneumoniae infection admitted to Chaohu Hospital Affiliated to Anhui Medical University from January 2023 to June 2025 were collected retrospectively. Using R software， the data were randomly split into a training set （295 cases） and a validation set （127 cases） at a ratio of 7∶3. LASSO and Logistic regression were used to identify risk factors， and a nomogram for the CRKP infection risk prediction model was further constructed with its performance analyzed. Results Multivariate analysis revealed five independent risk factors for CRKP infection in patients， namely ICU admission （OR = 4.883， 95%CI： 1.842 ~ 12.946）， tracheotomy/tracheal intubation （OR = 4.784， 95%CI： 1.763 ~ 12.981）， sputum aspiration （OR = 2.309， 95%CI： 0.767 ~ 6.954）， carbapenem administration （OR = 4.832， 95%CI： 1.944 ~ 12.009）， and combined antibacterial agent use （OR = 2.239， 95%CI： 0.938 ~ 5.341）. Subsequently， a prediction model was established based on these five variables screened from the training set， with the formula expressed as： Logit（P） = -3.632 + ［ICU admission （yes = 1， no = 0）］× 1.586 + ［Carbapenem use （yes = 1， no = 0）］× 1.575 + ［combined use of antibiotics （yes = 1， no = 0）］× 0.806 + ［sputum suction （yes = 1， no = 0）］× 0.837 + ［tracheotomy/tracheal intubation （yes = 1， n o =0）］× 1.565. The area under the receiver operating characteristic curve （AUC） was 0.921 （95% confidence interval ［CI］： 0.880~0.961） in the training set and 0.914 （95% CI： 0.846 ~ 0.983） in the validation set. The calibration curves generated for both the training set and validation set in this study demonstrated favorable consistency between the probability of CRKP infection predicted by the established model and the actual CRKP infection probability observed in clinical practice. The Hosmer-Lemeshow （H-L） test indicated that the model had a good degree of calibration. Decision curve analysis （DCA） further validated the favorable clinical utility of the constructed model. Conclusion The constructed risk prediction model can well predict high-risk patients with CRKP infection and can be used as a tool to assess the risk of CRKP infection in hospitalized patients.