ABSTRACT

Background

There is paucity in the literature to predict the occurrence of Ventilator Associated Pneumonia (VAP) in patients with Traumatic Brain Injury (TBI). We aimed to build a C.5. Decision Tree (C.5 DT) machine learning model to predict VAP in patients with moderate to severe TBI.

Methods

This was a retrospective study including all adult patients who were hospitalized with TBI plus head abbreviated injury scale (AIS) ≥ 3 and were mechanically ventilated in a level 1 trauma center between 2014 and 2019.

Results

A total of 772 eligible patients were enrolled, of them 169 had VAP (22%). The C.5 DT model achieved moderate performance with 83.5% accuracy, 80.5% area under the curve, 71% precision, 86% negative predictive value, 43% sensitivity, 95% specificity and 54% F-score. Out of 24 predictors, C.5 DT identified 5 variables predicting occurrence of VAP post-moderate to severe TBI (Time from injury to emergency department arrival, blood transfusion during resuscitation, comorbidities, Injury Severity Score and pneumothorax).

Conclusions

This study could serve as baseline for the quest of predicting VAP in patients with TBI through the utilization of C.5. DT machine learning approach. This model helps provide timely decision support to caregivers to improve patient’s outcomes.

中文翻译：

---

预测创伤性脑损伤患者呼吸机相关肺炎的机器学习模型：C.5 决策树方法

摘要

背景

文献中缺乏预测外伤性脑损伤 (TBI) 患者呼吸机相关肺炎 (VAP) 的发生率。我们的目标是建立一个 C.5。用于预测中重度 TBI 患者 VAP 的决策树 (C.5 DT) 机器学习模型。

方法

这是一项回顾性研究，包括所有在 2014 年至 2019 年期间在 1 级创伤中心接受机械通气的 TBI 加头部缩写损伤量表 (AIS) ≥ 3 住院的成年患者。

结果

共招募了 772 名符合条件的患者，其中 169 名患有 VAP（22%）。C.5 DT 模型实现了中等性能，准确度为 83.5%，曲线下面积为 80.5%，精确度为 71%，阴性预测值为 86%，灵敏度为 43%，特异性为 95%，F 值为 54%。在 24 个预测因子中，C.5 DT 确定了 5 个预测 VAP 发生中度至重度 TBI 后的变量（从受伤到急诊室到达的时间、复苏期间输血、合并症、损伤严重程度评分和气胸）。

结论

该研究可以作为通过利用 C.5 来预测 TBI 患者 VAP 的基线。DT 机器学习方法。该模型有助于为护理人员提供及时的决策支持，以改善患者的治疗效果。