Vetter Philipp, Niggli Cédric, Hambrecht Jan, Niggli Philipp, Vomela Jindrich, Chaloupka Richard, Pape Hans-Christoph, Mica Ladislav
Department of Trauma Surgery, University Hospital Zurich, 8091 Zurich, Switzerland.
Department of Mathematics, ETH Zurich, 8092 Zurich, Switzerland.
J Surg Res (Houst). 2022;5(4):618-624. doi: 10.26502/jsr.10020268. Epub 2022 Dec 5.
The is an outcome prediction tool invented by the University Hospital of Zurich in collaboration with IBM, representing an artificial intelligence application to predict the most adverse outcome scenarios in polytrauma patients: Systemic Inflammatory Respiratory Syndrome (SIRS), sepsis within 21 days and death within 72 h. The hypothesis was how lactate values woud be associated with the incidence of sepsis. Data from 3653 patients in an internal database, with ongoing implementation, served for analysis. Patients were split in two groups according to sepsis presence, and lactate values were measured at formerly defined time points from admission until 21 days after admission for both groups. Differences between groups were analyzed; time points with lactate as independent predictor for sepsis were identified. The predictive quality of lactate at 2 and 12 h after admission was evaluated. Threshold values between groups at all timepoints were calculated. Lactate levels differed from less than 2 h after admission until the end of the observation period (21 d). Lactate represented an independent predictor for sepsis from 12 to 48 h and 14 d to 21 d after admission relative to ISS levels. AUROC was poor at 2 and 12 h after admission with a slight improvement at the 12 h mark. Lactate levels decreased over time at a range of 2 [mmol/L] for 6-8 h after admission. These insights may allow for time-dependent referencing of lactate levels and anticipation of subsequent sepsis, although further parameters must be considered for a higher predictability.
这是苏黎世大学医院与IBM合作发明的一种结果预测工具，代表了一种人工智能应用，用于预测多发伤患者最不利的结果情况：全身炎症反应综合征（SIRS）、21天内发生败血症以及72小时内死亡。假设是乳酸值如何与败血症的发生率相关。内部数据库中3653名患者的数据（正在持续收集）用于分析。根据是否存在败血症将患者分为两组，两组均在从入院到入院后21天的先前定义时间点测量乳酸值。分析组间差异；确定以乳酸作为败血症独立预测指标的时间点。评估入院后2小时和12小时乳酸的预测质量。计算所有时间点组间的阈值。从入院后不到2小时到观察期结束（21天），乳酸水平存在差异。相对于损伤严重度评分（ISS）水平，乳酸在入院后12至48小时以及14至21天是败血症的独立预测指标。入院后2小时和12小时的曲线下面积（AUROC）较差，在12小时时略有改善。入院后6至8小时内，乳酸水平随时间下降幅度为2[mmol/L]。这些见解可能允许对乳酸水平进行时间依赖性参考，并预测随后的败血症，尽管为了获得更高的可预测性还必须考虑其他参数。
