Abstract:

Objective　To construct a model based on the LASSO (Least Absolute Shrinkage and Selection Operator) -logistic regression method, and to predict the short-term mortality risk in elderly patients with chronic obstructive pulmonary disease (COPD) and severe multidrug-resistant Acinetobacter baumannii pneumonia. Methods　In this study, 74 elderly COPD patients with severe multidrug-resistant Acinetobacter Baumannii pneumonia from the Department of Respiratory and Critical Care Medicine of Xi'an Gaoxin Hospital from January 2022 to December 2023 were selected as the study objects, including 55 males and 19 females, 41 patients <70 years old and 33 patients ≥70 years old. The LASSO-logistic regression algorithm was used to identify clinical variables significantly associated with short-term death risk, and these variables were used to construct a risk prediction model. The 10-fold cross-validation and Bootstrap method were used to validate the model internally, and the performance of the model was evaluated from the aspects of discrimination [area under the curve (AUC)] and calibration (calibration curve). χ² test and independent sample t test were used. Results　The survival of the patients within 30 days was observed. According to the final outcomes, the patients were divided into a death group (21 cases) and a survival group (53 cases), with a mortality rate of 28.37% (21/74). After LASSO cross-validation to determine the optimal parameters, the model selected five variables that were closely related to short-term death from general data, clinicopathological features, and previous treatment information: mechanical ventilation, fiberoptic bronchoscopy, sedation, septic shock, and use of antifungal medications. These variables were included in the logistic regression model, and the regression analysis showed that they were independent influencing factors for death in elderly COPD patients with severe multidrug-resistant Acinetobacter baumannii pneumonia (all P<0.05). The forest plot model constructed based on these predictors demonstrated excellent predictive performance for predicting short-term death in elderly COPD patients with severe multidrug-resistant Acinetobacter baumannii pneumonia, with an AUC of 0.927. In the analysis of the training set, the 1 000 bootstrap resamples and calibration curve analysis showed that the model's prediction results were highly consistent with the actual situation, with a mean absolute error (MAE) of 0.027. The Hosmer-Lemeshow goodness of fit test also confirmed the model's good calibration. Conclusions　The LASSO-logistic regression model constructed in this study can effectively predict the short-term death risk in elderly COPD patients with severe multidrug-resistant Acinetobacter Baumannii pneumonia. This model helps clinicians to better identify high-risk patients during the treatment decision-making process, so that appropriate treatment measures can be taken in time.

Key words:

Chronic obstructive pulmonary disease, Multidrug-resistant Acinetobacter Baumannii pneumonia, Elderly, Short term death warning model, LASSO logistic regression method

摘要：

目的　构建一种基于LASSO（Least Absolute Shrinkage and Selection Operator）-logistic回归算法的模型，用于预测老年慢性阻塞性肺疾病（COPD）合并重症多重耐药鲍曼不动杆菌肺炎患者的短期死亡风险。方法　将2022年1月至2023年12月西安高新医院呼吸与危重症医学科的74例老年COPD合并重症多重耐药鲍曼不动杆菌肺炎患者选为研究对象，其中男55例，女19例，年龄<70岁41例，≥70岁33例。通过LASSO-logistic回归算法筛选出与短期死亡风险显著相关的临床变量，利用这些变量构建风险预测模型，并采用10折交叉验证和Bootstrap方法对模型进行内部验证，从区分度［曲线下面积（AUC）］、校准度（校准曲线）方面评估模型的性能。采用χ²检验、独立样本t检验。结果　观察30 d内患者的生存情况，根据患者转归情况分为死亡组（21例）和生存组（53例），病死率为28.37%（21/74）。LASSO法通过交叉验证确定最优参数后，从一般资料、临床病理特征及既往治疗信息中筛选出5个与短期死亡密切相关的变量：行机械通气、行纤维支气管镜检查、使用镇静药物、合并脓毒症休克和使用抗真菌药物，这些变量被纳入logistic回归模型，回归分析显示它们是老年COPD合并重症多重耐药鲍曼不动杆菌肺炎患者死亡的独立影响因素（均P<0.05）。基于模型构建的用于预测老年COPD合并重症多重耐药鲍曼不动杆菌肺炎的短期死亡的森林图模型展示出良好的预测效能（AUC值为0.927），在训练集的分析中，Bootstrap法进行的1 000次重抽样验证和校准曲线分析显示，模型预测结果与实际情况高度一致，验证曲线的平均绝对误差（MAE）为0.027，Hosmer-Lemeshow拟合优度检验也证实了模型的校准度良好。结论　本研究构建的LASSO-logistic回归模型能够有效预测老年COPD合并重症多重耐药鲍曼不动杆菌肺炎患者的短期死亡风险。该模型有助于临床医生在治疗决策过程中更好地识别高风险患者，从而及时采取适当的治疗措施。

关键词:

慢性阻塞性肺疾病, 多重耐药鲍曼不动杆菌肺炎, 老年, 短期死亡预警模型, LASSO-logistic回归算法