基于Logistic回归的机械通气伴发高氧血症预测模型的构建及验证

李晓婕.基于Logistic回归的机械通气伴发高氧血症预测模型的构建及验证[J].内科急危重症杂志,2025,31(5):429-434

DOI：10.11768/nkjwzzzz20250507

英文关键词:

基金项目:

作者	单位	E-mail
李晓婕	南方医科大学第三附属医院	yhicu_nysy@163.com

摘要点击次数: 17

全文下载次数: 21

中文摘要:

摘要目的: 构建机械通气患者伴发高氧血症的预测模型，并验证其预测价值。方法:回顾性分析2020年1月至2021年3月南方医科大学第三附属医院重症医学科147例接受机械通气患者的临床资料，根据机械通气72 h最高动脉氧分压是否≥120 mmHg分为高氧血症组（94例）和非高氧血症组（53例）。收集入选患者性别、年龄、体重指数、急性生理学与慢性健康状况评估（APACHEⅡ）评分、序贯器官衰竭评估（ SOFA）评分、合并慢性疾病、收治原因、机械通气原因、机械通气参数、循环指标及实验室检查结果。对2组患者的临床数据进行单因素、Logistic多因素分析，通过独立危险因素的回归系数构建预测模型；采用受试者工作特征（ ROC）曲线对比预测模型与各原始协变量ROC曲线下面积（ AUC），以约登指数最大时确定最佳临界值，计算敏感度、特异性及预测准确性等工作性能参数。将预测模型应用于2022年1月至6月53例机械通气患者，验证其预测效能。结果:单因素分析结果显示，机械通气伴发高氧血症与SOFA评分、因内科疾患转入、肺部病变、因意识状态改变上机、持续正压气道通气（CPAP）、同步间歇指令通气、血乳酸及血红蛋白水平相关（P均<0.05）。Logistic多因素分析显示年龄（X1）、SOFA评分（X2）、因内科疾患转入（X3）、因意识状态改变上机（X4）、使用CPAP（X5）是机械通气患者发生高氧血症的独立危险因素（P均 < 0.05）。预测模型Y= 4.317-0.036X1-0.183X2+1.699X3-2.045X4-1.864X5。Hosmer-Lemeshow拟合优度检验χ2 =10.202，P=0.177。绘制ROC曲线，预测模型AUC为 0.817。预测模型的最佳临界值为0.404。预测模型预测机械通气伴发高氧血症的AUC高于年龄、SOFA评分、因内科疾患转入、因意识状态改变上机、使用CPAP模式单独预测时的AUC（0.817 vs. 0.574、0.651、0.609、0.554、0.641）。将预测模型和最佳临界值应用于53例机械通气患者作为验证，预测正确率为79.2%，敏感度80.6%，特异性77.3%，阳性预测值83.3%，阴性预测值73.9%，阳性似然比3.551，阴性似然比0.251。结论: 使用Logistic回归构建的预测模型对机械通气伴发高氧血症有较好的预测价值。

英文摘要:

Abstract Objective:To construct a predictive model of hyperoxemia in patients with mechanical ventilation and to verify its predictive value. Methods: This is a retrospective analysis of clinical data from 147 patients who underwent mechanical ventilation in the Intensive Care Unit of the Third Affiliated Hospital of Southern Medical University from January 2020 to March 2021, and patients were categorized into hyperoxemia group (94 cases) and non-hyperoxemia group (53 cases) according to whether the maximum arterial partial pressure of oxygen was ≥120mmHg during 72h of mechanical ventilation. The sex, age and body mass index (BMI) of the enrolled patients were recorded, along with the acute physiological assessment and chronic health evaluation Ⅱ (APACHEⅡ) score, sequential organ failure assessment (SOFA) score, chronic complications, reasons for admission, reasons for mechanical ventilation, ventilator parameters, hemodynamic data and laboratory test results. Univariate analysis was performed on the clinical variables of the two groups. Multivariate binary regression analysis of risk factors was conducted to screen independent risk factors affecting hyperoxemia in patients with mechanical ventilation. Then the combined predictor was constructed and the receiver operating characteristic (ROC) curve for the predictive model was built. The area under the ROC curve (AUC) for both the new indicator and original ones were compared. The best cut-off value was obtained where the Youden index reached the maximum value. Parameters such as sensitivity, specificity and predictive accuracy were also calculated for comparison. Finally, individual data of 53 patients with mechanical ventilation, who were admitted from January to June in 2022, were substituted into the equation to test the performance of the predictive model in predicting hyperoxemia. Results: Univariate analysis showed that SOFA score, admitting due to medical diseases, ventilator applied for lung lesions, ventilator applied for change of consciousness, using continuous positive airway pressure (CPAP), using synchronized intermittent mandatory ventilation (SIMV), lactate value and Hb value were risk factors for hyperoxemia in patients with mechanical ventilation (all P< 0.05). Multivariate analysis showed that age (X1), SOFA score (X2), admitting due to medical diseases (X3), ventilator applied for change of consciousness (X4) and using CPAP mode (X5) were independent risk factors for hyperoxemia in patients with mechanical ventilation (all P< 0.05). The combined predictor model was as follows: Y=4.317-0.036X1-0.183X2+1.699X3-2.045X4-1.864X 5. Hosmer and lemeshow test showed good fitting effect (χ2=10.202, P= 0.177). ROC curve analysis showed that the AUC for predicting hyperoxemia in patients with mechanical ventilation was 0.817. The best cut-off value of the predictive model was 0.404 points. The AUC for the new predictive model was higher than that of any other indicator, including age, SOFA score, admitting due to medical diseases, ventilator applied for change of consciousness and using CPAP mode (0.817 vs. 0.574, 0.651, 0.609, 0.554, 0.641). The clinical data of 53 patients with mechanical ventilation were substituted in the probability equation of prediction. Its predictive accuracy was 79.2% and its sensitivity was 80.6%, specificity was 77.3%, positive predictive value was 83.3%, negative predictive value was 73.9%, positive likelihood ratio was 3.551 and negative likelihood ratio was 0.251. Conclusion: The predictive model based on logistic regression analysis has good discrimination and can be used as a reference and assessment tool for prediction of hyperoxemia in patients with mechanical ventilation.