Identification of hemodynamically stable patients with acute pulmonary embolism at high risk for death: external validation of different models

Background: The optimal strategy for identification of hemodynamically stable patients with acute pulmonary embolism (PE) at risk for death and clinical deterioration remains undefined. Objectives: We aimed to assess the performances of currently available models/scores for identifying hemodynamically stable patients with acute, symptomatic PE at risk of death and clinical deterioration. Methods: This was a prospective multicenter cohort study including patients with acute PE (NCT03631810). Primary study outcome was in-hospital death within 30 days or clinical deterioration. Other outcomes were in-hospital death, death, and PE-related death, all at 30 days. We calculated positive and negative predictive values, c-statistics of European Society of Cardiology (ESC)-2014, ESC-2019, Pulmonary Embolism Thrombolysis (PEITHO), Bova, Thrombo-embolism lactate outcome study (TELOS), fatty acid binding protein, syncope and tachicardia (FAST), and National Early Warning Scale 2 (NEWS2) for the study outcomes. Results: In 5036 hemodynamically stable patients with acute PE, positive predictive values for the evaluated models/scores were all below 10%, except for TELOS and NEWS2; negative predictive values were above 98% for all the models/scores, except for FAST and NEWS2. ESC-2014 and TELOS had good performances for in-hospital death or clinical deterioration (c-statistic of 0.700 and 0.722, respectively), in-hospital death (c-statistic of 0.713 and 0.723, respectively), and PE-related death (c-statistic of 0.712 and 0.777, respectively); PEITHO, Bova, and NEWS2 also had good performances for PE-related death (c-statistic of 0.738, 0.741, and 0.742, respectively). Conclusion: In hemodynamically stable patients with acute PE, the accuracy for identification of hemodynamically stable patients at risk for death and clinical deterioration varies across the available models/scores; TELOS seems to have the best performance. These data can inform management studies and clinical practice.