颅脑损伤院内获得性肺炎预测:一项基于机器学习的PCA-Logistic回归分析模型

Prediction of hospital-acquired pneumonia after traumatic brain injury based on the PCA-Logistic regression analysis model

目的探讨PCA-Logistic回归分析模型在颅脑损伤病人院内获得性肺炎(HAP)预测建模中的应用效果。方法收集2011年12月至2017年11月开颅手术治疗的108例颅脑损伤的相关临床数据,建立PCA-Logistic回归分析模型,利用受试者工作特征(ROC)曲线评估模型预测效果。结果 PCA-Logistic回归分析模型发现影响病人HAP发生的重要临床指标,经ROC曲线评估PCA-Logistic回归分析模型预测HAP结局具有较高的预测效能,灵敏度为83.9%,特异度为94.8%,曲线下面积为0.949。结论 PCA-Logistic回归分析模型可以有效的挖掘颅脑损伤后的临床变量,可建立HAP的预测模型,不规范的肠外营养营养支持可能是影响HAP发生的重要临床因素。

Objective To explore the application of principal components analysis (PCA)- Logistic regression analysis to prediction of the hospital-acquired pneumonia (HAP) in the patients with traumatic brain injury (TBI), and to find the pathophysiological patterns and important risk factors related to their clinical prognoses. Methods Original dataset was constituted by the data of 108 patients with TBI derived from the Database of the Trauma Center of Sichuan Provincial People's Hospital from 2011 to 2017. The primary outcome was HAP. After the dataset evaluation and cleaning, the PCA-Logistic regression model was built to identify the risk factors related to HAP. Receiver Operating Characteristic (ROC) curve was used to evaluate the PCA-Logistic regression model. Results The PCA-Logistic regression model analysis found the important clinical indicators affecting the patients HAP, and the PCA-Logistic regression model was evaluated by ROC curve. The HAP outcome model had good predictive power (sensitivity, 83.9%; specificity, 94.8%; AUC, 0.949). Conclusions PCA-Logistic regression analysis can effectively mine the clinical variables of the patients with TBI and establish a clinical prognosis prediction model. The abnormal parenteral nutrition support after severe TBI may be an important clinical factor affecting the occurrence of HAP in the patients with TBI.