预测院际转院前「未见」普通外科患者死亡率:死亡率风险计算器的开发、验证与可行性试验
该研究回顾性纳入2016年1月至2022年8月转至普通外科或结直肠外科的患者,随机按3:1分为训练集和验证集,开发并验证用于院际转院时预测死亡率的工具。4664例转院患者中280例(6.0%)发生死亡,最终采用仅含12个变量的惩罚回归模型,验证集AUC为0.851、灵敏度0.90、特异度0.67、平衡准确度0.79,偏倚校正后Brier评分为0.04。作者认为该计算器预测能力优秀,可用于改善分诊、床位与资源分配并标准化诊疗。
要点
- 转院时使用12个常规采集变量构建的死亡率风险计算器,验证集AUC为0.851。
- 验证集灵敏度0.90、特异度0.67、平衡准确度0.79,偏倚校正后Brier评分0.04。
- 4664例转院患者中280例(6.0%)发生入院相关死亡。
核心结论
- 使用12个变量的风险计算器对「未见」急诊普通外科患者院际转院时的死亡率具有优秀预测能力。
- 在转院时量化患者死亡风险可改善患者分诊、床位与资源分配并标准化诊疗。
研究设计与样本
- 回顾性识别2016年1月至2022年8月转至普通外科或结直肠外科服务的患者,随机按3:1分为训练队列和验证队列。
- 主要结局为入院相关死亡,定义为指数住院期间死亡或出院后30天内死亡。
- 开发并验证多个预测模型,因仅使用12个变量即表现优秀而选择惩罚回归模型。
主要发现
- 4664例转院患者中280例(6.0%)发生死亡。
- 利用19个常规采集变量生成预测模型,最终选用仅含12个变量的惩罚回归模型。
- 验证集受试者工作特征曲线下面积、灵敏度、特异度、平衡准确度分别为0.851、0.90、0.67、0.79。
- 偏倚校正后Brier评分为0.04,提示分配风险与观察到的死亡频率强相关。
临床怎么用
- 作者认为在转院时量化患者死亡风险可改善患者分诊、床位与资源分配并标准化诊疗。
局限性
- 原文未报告局限性。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
OBJECTIVE: Develop and validate a mortality risk calculator that could be utilized at the time of transfer, leveraging routinely collected variables that could be obtained by trained nonclinical transfer personnel. BACKGROUND: There are no objective tools to predict mortality at the time of interhospital transfer for Emergency General Surgery patients that are "unseen" by the accepting system. METHODS: Patients transferred to general or colorectal surgery services from January 2016 to August 2022 were retrospectively identified and randomly divided into training and validation cohorts (3:1 ratio). The primary outcome was admission-related mortality, defined as death during the index admission or within 30 days postdischarge. Multiple predictive models were developed and validated. RESULTS: Among 4664 transferred patients, 280 (6.0%) experienced mortality. Predictive models were generated utilizing 19 routinely collected variables; the penalized regression model was selected over other models due to excellent performance using only 12 variables. The model performance on the validating set resulted in an area under the receiver operating characteristic curve, sensitivity, specificity, and balanced accuracy of 0.851, 0.90, 0.67, and 0.79, respectively. After bias correction, the Brier score was 0.04, indicating a strong association between the assigned risk and the observed frequency of mortality. CONCLUSIONS: A risk calculator using 12 variables has excellent predictive ability for mortality at the time of interhospital transfer among "unseen" Emergency General Surgery patients. Quantifying a patient's mortality risk at the time of transfer could improve patient triage, bed and resource allocation, and standardize care.
引用
- Said SA, Gentle CK, Gross A, et al. Predicting Mortality Before Interhospital Hospital for "Unseen" General Surgery Patients: Development, Validation, and Feasibility Trial of a Mortality Risk Calculator. Ann Surg. 2024. PMID: 38726671. doi:10.1097/SLA.0000000000006334
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。