利用机器学习提高院际分诊指南的实用性:老年院际创伤分诊评分表的开发
本研究为回顾性预后研究,基于2009—2019年俄克拉荷马州创伤登记处数据,纳入5,913例55岁及以上、从受伤现场首先送至非三级创伤中心(NTC)的老年创伤患者,结局为死亡或需干预操作的严重损伤(ISS≥16)。最终模型曲线下面积75.4%(95%CI 74-76%),最强预测因子及权重为气道干预(10)、交通相关股骨骨折(6)、脊髓损伤(5)、急诊科GCS≤13(5)、血流动力学支持(4);偏倚校正与样本验证AUC分别为74%和72%。风险评分7分时敏感度78%、特异度56%。作者认为该评分可简化老年创伤患者向三级创伤中心(TTC)的二次分诊。
要点
- 风险评分≥7分时,识别需转至TTC的老年创伤患者敏感度78%、特异度56%(最终模型AUC 75.4%,95%CI 74-76%)。
- 评分权重最高的预测因子为气道干预(10分),其次为交通相关股骨骨折(6分)、脊髓损伤(5分)、急诊科GCS≤13(5分)、血流动力学支持(4分)。
- 研究对象为55岁及以上、从受伤现场首先送至NTC的老年创伤患者,结局为死亡或需干预操作的严重损伤(ISS≥16)。
- 偏倚校正和样本验证的AUC分别为74%和72%,作者认为该评分可能简化老年创伤患者的二次分诊。
适用人群与场景
- 55岁及以上受伤成人,从受伤现场首先运送至非三级创伤中心(NTC),随后可能转至三级创伤中心(TTC)。
诊断标准与关键阈值
- 结局定义:死亡或严重损伤(损伤严重度评分ISS≥16)且需要在接收机构进行干预操作。
- 风险评分阈值:7分时敏感度78%、特异度56%。
研究设计与关键结果
- 研究类型:回顾性预后研究,使用2009—2019年俄克拉荷马州创伤登记处数据。
- 样本量:5,913例55岁及以上、首先送至NTC的老年创伤患者;其中32.7%(1,696例)在TTC发生目标结局。
- 最终预后模型AUC 75.4%(95%CI 74-76%),纳入的前四位预测因子及权重:气道干预(10)、交通相关股骨骨折(6)、脊髓损伤(5)、急诊科GCS≤13(5)、血流动力学支持(4)。
- 偏倚校正和样本验证AUC分别为74%和72%。
- 风险评分7分:敏感度78%,特异度56%。
证据类型、等级与局限
- 证据类型:回顾性预后研究;证据等级:预后与流行病学,Level III。
- 局限:原文未报告局限性。
对床旁实践的改变
- 作者认为,使用风险评分可增强老年创伤患者向TTC的二次分诊。
- 作者认为,本研究首次开发了需要转至更高级别医疗机构的老年创伤患者风险分层工具。
提炼说明
- 本条依据论文摘要整理;摘要未涉及「处置流程与药物剂量」「红旗征与不能漏的情况」「指南或共识出处」,故未列出这些栏目;全文请点「阅读原文」。
英文原文(摘要)· 点击展开
BACKGROUND: Undertriage of injured older adults to tertiary trauma centers (TTCs) has been demonstrated by many studies. In predominantly rural regions, a majority of trauma patients are initially transported to nontertiary trauma centers (NTCs). Current interfacility triage guidelines do not highlight the hierarchical importance of risk factors nor do they allow for individual risk prediction. We sought to develop a transfer risk score that may simplify secondary triage of injured older adults to TTCs.
METHODS: This was a retrospective prognostic study of injured adults 55 years or older initially transported to an NTC from the scene of injury. The study used data reported to the Oklahoma State Trauma Registry between 2009 and 2019. The outcome of interest was either mortality or serious injury (Injury Severity Score, ≥16) requiring an interventional procedure at the receiving facility. In developing the model, machine-learning techniques including random forests were used to reduce the number of candidate variables recorded at the initial facility.
RESULTS: Of the 5,913 injured older adults initially transported to an NTC before subsequent transfer to a TTC, 32.7% (1,696) had the outcome of interest at the TTC. The final prognostic model (area under the curve, 75.4%; 95% confidence interval, 74-76%) included the following top four predictors and weighted scores: airway intervention (10), traffic-related femur fracture (6), spinal cord injury (5), emergency department Glasgow Coma Scale score of ≤13 (5), and hemodynamic support (4). Bias-corrected and sample validation areas under the curve were 74% and 72%, respectively. A risk score of 7 yields a sensitivity of 78% and specificity of 56%.
CONCLUSION: Secondary triage of injured older adults to TTCs could be enhanced by use of a risk score. Our study is the first to develop a risk stratification tool for injured older adults requiring transfer to a higher level of care.
LEVEL OF EVIDENCE: Prognostic and Epidemiolgical; Level III.
参考来源
- Garwe T, Newgard CD, Stewart K, et al. Enhancing utility of interfacility triage guidelines using machine learning: Development of the Geriatric Interfacility Trauma Triage score. J Trauma Acute Care Surg. 2022. PMID: 36404409. doi:10.1097/TA.0000000000003846
供执业医师抢救与诊疗参考,不替代临床判断;剂量请结合体重、肝肾功能和本院药品规格复核,以指南原文为准。