Evaluating the Predictive Value of Clinical Risk Scores for Intensive Care Unit Admission in Low- and Intermediate-risk Acute Pulmonary Embolism: A Retrospective Analysis From a Multicenter Health System

Abstract Introduction: Accurate risk stratification is essential for guiding management decisions in patients with acute pulmonary embolism (PE), particularly regarding Intensive Care Unit (ICU) admission, to ensure optimal care and resource allocation. While several clinical scores exist to assess PE severity, their predictive value for ICU admission remains unclear. Methods: A retrospective chart review was performed of acute PE patients managed by PE Response Team at three urban teaching hospitals in the Mount Sinai Health System from January 2019 to June 2024. We compared low- and intermediate-risk patients (by the European Society of Cardiology [ESC] risk classification) initially admitted to the ICU to those on the medical floors. A multivariable regression model and predictive performance of different scores for ICU admission were conducted, adjusting for Pulmonary Embolism Severity Index (PESI), National Early Warning Score (NEWS), presence of proximal deep vein thrombosis, and active malignancy. Results: From a total of 317 patients, 290 (91%) were admitted to the ICU and 27 (9%) to the floors. Mean age was 64 (interquartile range [IQR] 53-75), 172 were females (54.3%) and other baseline demographics were similar between both groups. ICU patients had higher troponin I levels (median 0.12 vs. 0.04, p=0.018), more saddle PE (21% vs 1%, p=0.038) and right ventricle strain (70.3% vs. 48.1%, p=0.028). Both the Bova (p=0.007) and Composite PE Shock (CPES) Score (p=0.008) were significantly higher in the ICU group, while other shock indices and scores, and outcomes such as mortality and bleeding at 30 days showed no significant differences.In the multivariable regression model, the Bova and CPES demonstrated a statistically significant association with ICU admission (Bova adjusted Odds Ratio [aOR]: 1.412; 95% confidence interval [CI]: 1.10-1.82. CPES aOR: 1.615; 95% CI: 1.14-2.32) and had a fair predictive performance (Bova-AUROC: 0.672; CPES-AUROC 0.677). Other scores did not demonstrate significant associations with ICU admission (p > 0.05) and showed poorer predictive performance. Interestingly, the ESC risk score could not fully predict admission to the ICU, possibly due to a weakened effect after adjusting by PESI and NEWS. Conclusions: Among the scores evaluated, only the Bova and CPES provided meaningful predictive value for ICU admission in this cohort. Incorporating them into clinical assessments could enhance decision-making around ICU admissions in this patient population. Our analysis may be limited by the dataset's inability to fully capture patient severity and clinical trajectory. Further studies are needed to validate these scores across diverse settings.