Abstract Background Timely and accurate risk stratification in the emergency department (ED) for patients with suspected infection and sepsis is crucial for improving outcomes and optimizing resource utilization. The MeMed Severity™ (MMSev) test integrates 3 blood biomarkers to generate a risk score for progression to severe outcomes. Here we evaluate the performance of MMSev and its potential to augment standard-of-care (SOC) tools. Methods The study population included retrospective and prospective cohorts of adults with suspected infection, recruited from 35 sites across the United States, Europe, and Israel. Severe outcomes were defined as a composite endpoint of invasive mechanical ventilation, vasopressor support, or renal replacement therapy within 72 hours, or mortality within 14 days. The MMSev score, derived from this cohort using multivariate logistic regression and validated via cross-validation, stratifies patients into 5 score bins by the likelihood of deterioration. MMSev performance was compared to SOC tools and emerging sepsis markers. The Area Under the Receiver Operating Characteristic Curve (AUC) was calculated for each comparator based on its data availability. Net reclassification improvement (NRI) was calculated using a cutoff between bins 2 and 3 for MMSev, while a cutoff of two was applied for SIRS and qSOFA. Results The cohort included 1,939 adult patients (median age: 64 years, 47% female), with 15.8% experiencing severe outcomes, of whom 60.1% deteriorated at least one day after blood draw. Recruitment was 56% retrospective and 44% prospective, with 74.4% from the ED. Among the cohort, 53.3% met =2 Systemic Inflammatory Response Syndrome (SIRS) criteria at the time of presentation, 8.7% were admitted to the ICU, 8.7% required vasopressor support, 5.5% required mechanical ventilation, and 7.7% died. MMSev achieved an AUC of 0.84 on the derivation cohort, outperforming comparators (AUC: 0.64–0.76), and score bins were positively associated with an increased likelihood of severe outcomes (Figure). Rule-in bins had positive predictive values of 57% and 30% with specificities of 95% and 81% at the higher and lower rule-in cutoffs, respectively, while rule-out bins showed negative predictive values of 99% and 93% with sensitivities of 97% and 91% at the lower and higher rule-out cutoffs, respectively. Cross-validation AUC was 0.84, consistent with the training set AUC, demonstrating no evidence of overfitting and supporting model generalizability. The NRI of MMSev compared to qSOFA and SIRS was 0.27 and 0.20 respectively, highlighting its added value over routinely used scores. Conclusion MMSev shows potential for improving risk stratification in ED patients with suspected infection and sepsis. This predictive capability serves as a complementary tool to SOC methods, offering an additional layer of risk assessment. Further studies in independent cohorts are required to validate its clinical utility and generalizability.
Angel et al. (Wed,) studied this question.