Impact of the updated SOFA-2 score on sepsis diagnosis and prognosis: a retrospective multicenter cohort study

The Sepsis-3 criteria operationalized organ dysfunction using the original Sequential Organ Failure Assessment (SOFA-1) score, which was updated to SOFA-2 in October 2025 to align with modern intensive care unit (ICU) practices. However, the impact of adopting SOFA-2 for sepsis detection under the Sepsis-3 criteria has not yet been evaluated.

We conducted a retrospective multicenter cohort study using three large-scale ICU databases from the United States and the Netherlands. Adult patients with suspected infection within 72 h of ICU admission were included. Sepsis was independently identified according to Sepsis-3 criteria, utilizing either the SOFA-1 or SOFA-2 score. We systematically compared diagnostic concordance, the timeliness of sepsis detection, clinical outcomes and predictive performance of prognostic models between the two scoring systems. The primary outcome was ICU mortality, while secondary outcomes included hospital mortality and 28-day survival.

The study cohort comprised 74,615 adult patients with suspected infection. The diagnostic concordance of sepsis between SOFA-1 and SOFA-2 reached 89.62%. However, SOFA-1 and SOFA-2 uniquely identified an additional 3.54% and 6.84% of patients as having sepsis, respectively. The diagnostic discrepancies were primarily attributable to updates in respiratory and renal scoring criteria. ICU mortality was highest among the Concordant Positive group (15.63%). Notably, both discordant groups exhibited substantial mortality (SOFA-1 Only: 8.31%; SOFA-2 Only: 9.23%), both of which were significantly higher than those of patients not classified as having sepsis by either score (6.71%; P = .002 and P = 2.87 × 10^-15). Within the Concordant Positive group, 61.24% of patients were diagnosed simultaneously by both criteria. However, SOFA-2 achieved earlier diagnosis in a greater proportion of cases than SOFA-1 (23.12% vs. 15.64%), despite heterogeneity across different databases. Predictive models derived from the SOFA-2 score demonstrated numerically higher area under the receiver operating characteristic curve (AUROC) values in forecasting ICU mortality than those based on SOFA-1 (0.736 vs. 0.728 in internal cross-validation, P = .386; 0.743 vs. 0.720 in external validation, P = .375).

SOFA-1 and SOFA-2 showed high concordance in sepsis detection, yet each identified distinct patient subgroups with significant mortality. A transitional strategy utilizing both SOFA-1 and SOFA-2 is advised until updated expert-validated sepsis criteria are established.