A Combined Model of Vital Signs and Serum Biomarkers Outperforms Shock Index in the Prediction of Hemorrhage Control...

BackgroundDistinguishing surgical intensive care unit (ICU) patients with ongoing bleeding who require hemorrhage control interventions (HCI) can be challenging. Guidelines recommend risk-stratification with clinical variables and prediction tools, however supporting evidence remains mixed.MethodsThis retrospective study evaluated adult patients admitted to the surgical ICU with concern for ongoing hemorrhage under our institution's "Hemorrhage Watch" (HW) protocol and aimed to derive a clinical prediction model identifying those needing HCI with serial vital signs (VS) and serum biomarkers. The HW protocol included ICU admission followed by a 3-h observation period with VS monitoring every 15 min and hourly biomarkers. The primary outcome was the need for HCI (operative and endovascular interventions) within nine hours of ICU arrival. Secondary outcomes included in-hospital mortality, blood transfusions, and ICU and hospital length-of-stay. A clinical prediction model was developed by utilizing the variables most associated with HCI in a best subsets regression, which was subsequently internally validated using a Bootstrap algorithm.Results305 patients were identified for inclusion and 18 (5.9%) required HCI (3 operative, 15 endovascular). The median age was 70 years (IQR 54, 83), 60% had traumatic injuries, and 73% were enrolled from the emergency department. Blood product transfusion and mortality were similar between the HCI and no-HCI groups. Our analysis demonstrated that a model based on the minimum hemoglobin (9.9 vs 8.1 g/dL), minimum diastolic (57 vs 53 mm Hg) and systolic blood pressures (105 vs 90 mm Hg), and minimum respiratory rate (15 vs 18) could predict HCI with an area under the Receiver Operating Characteristics curve (AUROC) of 0.87, outperforming the Shock Index (SI) (AUROC = 0.64).ConclusionsIn this study of surgical ICU patients with concern for ongoing bleeding, a prediction model using serial VS and biomarkers outperformed the SI and may help identify those requiring HCI.