Scheetz Linda J, Zhang Juan, Kolassa John
New York University College of Nursing, 246 Greene Street, New York, NY 10003, USA.
Artif Intell Med. 2009 Jan;45(1):1-10. doi: 10.1016/j.artmed.2008.11.002. Epub 2008 Dec 16.
Motor vehicle crashes are a leading cause of mortality and morbidity worldwide. Even though trauma centers provide the gold standard of care for motor vehicle crash patients with life- or limb-threatening injuries, many whose lives might be saved by trauma center care are treated instead at non-trauma center hospitals. Triage algorithms, designed to identify patients with life- or limb-threatening injuries who should be transported to a trauma center, lack appropriate sensitivity to many of these injuries. The challenge to the trauma community is differentiating patients with life- or limb-threatening injuries from those with less severe injuries at the crash scene so that the patients can be transported to the most appropriate level of care. The purpose of this study was to use crash scene data available to emergency responders to classify adults with moderate and severe injuries. These classifiers might be useful to guide triage decision making.
Records of 74,626 adults, age 18-64 years, from the National Automotive Sampling System Crashworthiness Data Systems database were analyzed using classification and regression trees (CART) analysis. Both CART models (moderate injury and severe injury) included 13 predictor variables. The response variables were the targeted injury severity score cut points for moderate and severe injury. Two final classification trees were developed: one that classified occupants based on moderate injury and the other on severe injury. Misclassification costs were manipulated to achieve the best model fit for each tree.
The moderate injury classification tree had three splitters: police-estimated injury severity, restraint use, and number of persons injured. The severe injury classification tree had four splitters: police-estimated injury severity, manner of collision, number of persons injured in the crash, and age. Sensitivity and specificity of the classification trees were 93.70%, 77.53% (moderate) and 99.18%, 73.96% (severe), respectively.
CART analysis can be used to classify injury severity using crash scene information that is available to emergency responders. This procedure offers an opportunity to examine alternative methods of identifying injury severity that might assist emergency responders to differentiate more accurately persons who should receive trauma center care from those who can be treated safely at a non-trauma center hospital.
机动车碰撞是全球死亡率和发病率的主要原因。尽管创伤中心为有生命或肢体威胁损伤的机动车碰撞患者提供了护理的金标准，但许多本可通过创伤中心护理挽救生命的患者却在非创伤中心医院接受治疗。旨在识别应被送往创伤中心的有生命或肢体威胁损伤患者的分诊算法，对许多此类损伤缺乏适当的敏感性。创伤领域面临的挑战是在事故现场区分有生命或肢体威胁损伤的患者与伤势较轻的患者，以便将患者送往最合适的护理级别。本研究的目的是利用应急响应人员可获取的事故现场数据对中度和重度损伤的成年人进行分类。这些分类器可能有助于指导分诊决策。
使用分类与回归树（CART）分析对来自国家汽车抽样系统碰撞安全性数据系统数据库的74626名18至64岁成年人的记录进行分析。两个CART模型（中度损伤和重度损伤）均包括13个预测变量。响应变量是中度和重度损伤的目标损伤严重程度评分切点。开发了两棵最终分类树：一棵基于中度损伤对驾乘人员进行分类，另一棵基于重度损伤进行分类。对误分类成本进行了调整，以实现每棵树的最佳模型拟合。
中度损伤分类树有三个分割点：警方估计的损伤严重程度、安全带使用情况和受伤人数。重度损伤分类树有四个分割点：警方估计的损伤严重程度、碰撞方式、事故中受伤人数和年龄。分类树的敏感性和特异性分别为93.70%、77.53%（中度）和99.18%、73.96%（重度）。
CART分析可用于利用应急响应人员可获取的事故现场信息对损伤严重程度进行分类。该程序提供了一个机会来研究识别损伤严重程度的替代方法，这可能有助于应急响应人员更准确地区分应接受创伤中心护理的人员与可在非创伤中心医院安全治疗的人员。
