The new SOFA for sepsis identification and mortality prediction: a multicenter cohort study Authors: Lina Zhao, Qian Cui, Jie Liu, Shuoyan Dong, Lei Zong, Fei Yang, Hui Shi, Yunying Wang, Xiaoli Zhao, Haibo Li, Yipeng Fang, Ying Gao, Hanjun Pei, Fuhong Su, Haibo Zhang, Daniel De Backer, Yun Li, Keliang Xie Published: 2026 ABSTRACT BackgroundThe Sequential Organ Failure Assessment (SOFA) score is central to Sepsis-3 criteria. SOFA-2 updates thresholds and incorporates contemporary organ support practices, but its impact on sepsis identification and outcome prediction remains uncertain. This study aimed to compare the diagnostic yield and prognostic performance of SOFA-2 versus SOFA-1 for sepsis identification in critically ill adults.MethodsWe conducted a retrospective multicenter cohort study of 11,669 ICU admissions from three tertiary hospitals in China (January 2022-October 2025) and externally validated findings in 29,811 ICU patients from MIMIC-IV. Sepsis was defined by Sepsis-3 using either SOFA-1 or SOFA-2. We evaluated diagnostic agreement, organ dysfunction profiles, and discrimination for ICU mortality.ResultsSOFA-2 identified a significantly larger sepsis population than SOFA-1 (49.0% vs. 45.5%, P < 0.001), while maintaining substantial diagnostic agreement with SOFA-1 (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k$$\end{document}k=0.891–0.908). Compared with SOFA-1, SOFA-2 assigned higher total scores in the mid-to-high range and captured more organ dysfunction. Despite these reclassifications, ICU and hospital length of stay and mortality did not differ significantly between sepsis-SOFA-1 and sepsis-SOFA-2 groups (all P > 0.05). SOFA-2 showed slightly but consistently higher discrimination for ICU mortality than SOFA-1 (internal AUC 0.720 vs. 0.703; external AUC 0.735 vs. 0.722) and clearer mortality gradients in moderate-to-severe dysfunction categories (P < 0.001).ConclusionThe updated SOFA-2 score may identify a moderately larger sepsis population with more advanced organ dysfunction and may provide modestly improved mortality risk stratification compared with SOFA-1, while maintaining substantial diagnostic agreement. These findings suggest that SOFA-2 could be considered for clinical use in ICU sepsis assessment, and the Sepsis-3 SOFA ≥ 2 threshold may not require recalibration when using SOFA-2.Supplementary InformationThe online version contains supplementary material available at 10.1186/s40560-026-00891-6. Introduction Sepsis, defined as life-threatening organ dysfunction caused by a dysregulated host response to infection, remains a leading cause of ICU mortality worldwide [1]. The Third International Consensus Definition for Sepsis and Septic shock (Sepsis-3) proposed that organ dysfunction can be recognized as an acute increase of ≥ 2 points in the Sequential Organ Failure Assessment (SOFA) score [1]. The original SOFA score (SOFA-1) has been widely used to quantify organ dysfunction and predict outcomes in sepsis and other critically ill populations [2, 3]. However, SOFA-1 was developed almost three decades ago in a very different therapeutic landscape, and emerging data suggest that some of its thresholds and variables may no longer optimally reflect contemporary ICU practice [4–7]. Recently, an updated SOFA-2 score was proposed to address these limitations [8]. SOFA-2 modifies several organ-system components, incorporates organ modalities, and revises thresholds to better align with current clinical practice. For neurological assessment, SOFA-2 integrates the impact of sedation and delirium alongside the Glasgow Coma Scale (GCS). For the respiratory system, oxygenation thresholds were recalibrated and non-invasive ventilation (NIV), high-flow nasal cannula (HFNC), and extracorporeal membrane oxygenation (ECMO) were explicitly incorporated. The renal component explicitly codes continuous renal replacement therapy (CRRT) and updates creatinine thresholds, while the cardiovascular support (e.g., intra-aortic balloon pump [IABP], ECMO). Thresholds for hepatic dysfunction and coagulation were also updated [8]. Initial analyses suggest that SOFA-2 has good predictive validity for mortality in heterogeneous ICU populations [8], but its impact on sepsis identification, particularly when Sepsis-3 thresholds continue to be expressed in terms of “SOFA ≥ 2,” is largely unknown. We hypothesized that the updated SOFA-2 score would (1) identify a larger sepsis population than SOFA-1 by incorporating contemporary organ support modalities (including non-invasive ventilation, high-flow nasal cannula, continuous renal replacement therapy, intra-aortic balloon pump, and ECMO), revising physiological thresholds, and integrating sedation and delirium assessment, thereby enabling more sensitive detection of early or evolving multiorgan dysfunction that may be underrecognized by SOFA-1; and (2) provide more accurate mortality risk stratification, particularly in patients with intermediate disease severity, because SOFA-2 enhances granularity and discriminatory resolution within the mid‑range of organ dysfunction where SOFA‑1 often provides limited risk differentiation, which is a critical but under‑stratified population in clinical practice. To test these hypotheses, we conducted a multicenter retrospective cohort study in three Chinese ICUs with external validation using the MIMIC-IV database. We compared diagnostic agreement, organ-system profiles, and prognostic performance of SOFA-1 and SOFA-2 in adult ICU patients with suspected or confirmed infection. Methods Data sources This was a retrospective multicenter cohort study. The internal validation cohort included 11,669 consecutive adult ICU admissions to three tertiary hospitals (Tianjin Medical University General Hospital, The First Affiliated Hospital of Baotou Medical College, and Chifeng Municipal Hospital) between January 2022 and December 2025. The study protocol was approved by the institutional review boards of all participating centers. A waiver of informed consent was granted because only de-identified routinely collected data were analyzed. All procedures complied with the Declaration of Helsinki and local regulations. An external validation cohort was derived from 29,811 ICU patients in the publicly available Medical Information Mart for Intensive Care IV (MIMIC-IV) [9]. We followed STROBE guidelines for reporting observational cohort studies to ensure methodological transparency and reproducibility. Case definitions We evaluated sepsis within the first 24 h following ICU admission. Patients were eligible if there were ≥ 18 years old, had suspected or confirmed infection, and had sufficient data to compute both SOFA-1 and SOFA-2 scores. Suspected infection required a documented antibiotic prescription and microbiological culture sampling. Microbiological samples were collected within 24 h of first antibiotic administration, and antibiotics were initiated within 72 h of the first microbiological sampling. Sepsis was defined as an increase in SOFA of ≥ 2 points from baseline according to Sepsis-3 [1], calculated separately using SOFA-1 and SOFA-2. In the present study, baseline SOFA score was defined as 0 for all eligible patients upon ICU admission. The operationalization of SOFA-2 components was defined a priori for each organ system, based on the original SOFA-2 technical specifications (eTable 1) [8]. We excluded patients younger than 18 years, those readmitted to the ICU during the same hospitalization, those with missing critical variables required for SOFA calculation, and those with ICU survival time < 24 h and palliative care in the internal cohort. For external validation cohort, we included only the first ICU admission during a hospitalization and applied the same exclusion criteria. To ensure comparability between cohorts, all SOFA component variables were extracted using the same 24-h window following ICU admission, consistent with Sepsis-3 operationalization. Data collection For the internal cohort, clinical data were extracted from electronic health records (EHRs) during the first 24 h of ICU stay. For the external cohort, standardized MIMIC-IV queries were used to extract analogous variables. Collected variables included demographics, comorbidities and Charlson comorbidity index (CCI), infection source and microbiology, physiological and laboratory variables for SOFA computation, and ICU interventions (invasive ventilation, non-invasive ventilation [NIV], high-flow nasal cannula [HFNC], continuous renal replacement therapy [CRRT], extracorporeal membrane oxygenation [ECMO], intra-aortic balloon pump [IABP], and cardiopulmonary resuscitation). Severity scores (Acute Physiology and Chronic Health Evaluation II, APACHE II and Simplified Acute Physiology Score II, SAPS II) were collected to characterize baseline illness severity across cohorts, rather than as covariates in prognostic modeling, the choice of APACHE II (internal cohort) and SAPS II (external cohort) reflects regional clinical practices—APACHE II is more widely used in Chinese ICUs, while SAPS II is standard in MIMIC-IV, ensuring clinically relevant comparisons in each setting. Analysis Continuous variables are summarized as medians with interquartile ranges (IQRs) and were compared using the Mann–Whitney U test. Categorical variables are presented as counts and percentages and were compared using χ2 or Fisher’s exact test, as appropriate. We described the distribution of total SOFA-1 and SOFA-2 scores across patient categories and evaluated the reclassification between scoring systems using heatmaps and Sankey diagrams. Diagnostic agreement for sepsis identification (SOFA ≥ 2) was assessed using weighted Cohen's kappa statistics. The incidence of organ dysfunction across SOFA categories was systematically compared between the two scoring systems. ICU mortality proportions with 95% confidence intervals were estimated for each SOFA category. To assess prognostic performance, we developed multivariable logistic regression models incorporating either SOFA-1 or SOFA-2 along with baseline severity indicators (APACHE II, SAPS II), the Charlson Comorbidity Index, lactate levels, hematologic parameters, and age. These models generated adjusted odds ratios (ORs) with 95% CIs. We quantified discrimination using area under the receiver operating characteristic curve (AUC), determining optimal cut-offs through Youden index maximization and comparing AUCs via DeLong test. Non-linear relationships (restricted cubic spline, RCS) between individual SOFA components and ICU mortality were explored using restricted cubic spline models, stratified by sepsis classification according to each scoring system. The statistical analyses focused on evaluating the discrimination and classification performance of the SOFA score, with corresponding sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) derived from contingency tables to assess its diagnostic accuracy. Missing data were handled using a domain-specific approach. For binary organ support variables (vasopressors, mechanical ventilation, CRRT, ECMO, IABP), missing values were conservatively scored as 0. Continuous SOFA variables (PaO2·FiO2⁻1, bilirubin, creatinine, platelet count, GCS) required complete data for SOFA computation; patients missing these values were excluded (eTable 2). For other variables with > 10% missingness, multiple imputation with chained equations was performed (eFigure 1). Outliers were managed using predefined clinical ranges for SOFA-related variables (eTable 1) and Tukey’s fences (1.5 × IQR) for other continuous variables. A two-sided P value < 0.05 was considered statistically significant. Analyses were performed using R (version 4.4.3). Result Descriptive comparison The internal cohort included 11,669 ICU admissions after exclusions, and the external validation cohort included 29,811 ICU admissions from MIMIC-IV. In the internal cohort, 5,313 (45.5%) met SOFA-1-defined sepsis and 5,717 (49.0%) met SOFA-2-defined sepsis (Fig. 1). Baseline characteristics were broadly similar between sepsis-SOFA-1 and sepsis-SOFA-2 groups (Table 1). Sepsis patients defined by either score had significantly higher illness severity and worse outcomes compared with non-sepsis patients (eTables 3–4).Fig. 1Study flow diagram and cohort derivation. A Flow diagram of the internal validation cohort. Among 16,523 ICU admissions, patients were excluded for ICU readmission during the same hospitalization, missing SOFA components, or ICU survival < 24 h, leaving 11,669 unique ICU patients. Sepsis was defined as infection plus an acute increase ≥ 2 points in SOFA-1 or SOFA-2 within 24 h of ICU admission. Numbers of patients classified as septic by SOFA-1 only, SOFA-2 only, or by both scores are shown. The Venn diagram depicts overlap between sepsis-SOFA-1 and sepsis-SOFA-2 within the internal cohort. B Flow diagram of the external validation cohort (MIMIC-IV). Among 65,366 ICU stays, exclusions for short ICU stays and missing SOFA components yielded 29,811 first ICU admissions. The numbers of sepsis patients by SOFA-1 and SOFA-2 criteria and their overlap are presented, with a Venn diagram illustrating unique and shared sepsis classificationsTable 1Baseline data and outcome indicators of sepsis patientsVariableSepsis-SOFA1 (n = 5313)Sepsis-SOFA2(n = 5717)PAge, years,(median[IQR])66.80 [55.40, 77.20]66.80 [55.30, 77.20]0.837Gender, male, n (%)3173 (59.7)3414 (59.7)1.000BMI, (median[IQR])28.10 [24.10, 33.20]28.10 [24.10, 33.20]0.987Comorbidity, n (%) Charlson, (median[IQR])6.00 [4.00, 8.00]6.00 [4.00, 8.00]0.431 Hypertension1840 (34.6)1970 (34.5)0.857 Diabetes1762 (33.2)1960 (34.3)0.219 Myocardial infarct1121 (21.1)1232 (21.5)0.577 Congestive heart failure2053 (38.6)2228 (39.0)0.725 Mild liver disease1364 (25.7)1424 (24.9)0.357 Severe liver disease816 (15.4)839 (14.7)0.324 Chronic kidney disease1519 (28.6)1698 (29.7)0.201 Chronic pulmonary disease1427 (26.9)1529 (26.7)0.897 Cerebrovascular disease698 (13.1)771 (13.5)0.594 Immunosuppression1218 (22.9)1321 (23.1)0.839Illness severity scores, (median[IQR]) SOFA1 score4.00 [2.00, 7.00]4.00 [3.00, 7.00]< 0.001 SOFA2 score5.00 [3.00, 8.00]5.00 [3.00, 8.00]0.619 APACHE II score20.00 [14.00, 28.00]20.00 [14.00, 28.00]0.766Infection source, n (%) Lung infection2316 (43.6)2659 (46.5)0.002 Abdominal infection403 (7.6)468 (8.2)0.244 Urinary infection448 (8.4)478 (8.4)0.918 Bloodstream infection238 (4.5)254 (4.4)0.927 Skin and soft tissue infection416 (7.8)412 (7.2)0.219Pathogenic microorganisms, n (%) Klebsiella pneumoniae1201 (22.6)1410 (24.7)0.011 Acinetobacter baumannii996 (18.7)1169 (20.4)0.026 Pseudomonas aeruginosa603 (11.3)656 (11.5)0.857 Staphylococcus aureus1145 (21.6)1214 (21.2)0.693 Escherichia coli919 (17.3)1041 (18.2)0.213Outcome Septic shock, n (%)1727 (32.5)1816 (31.8)0.414 Vasoactive drugs, n (%)2962 (55.8)2992 (52.3)< 0.001 InvasiveVent, n (%)4072 (76.6)4265 (74.6)0.013 CPR, n (%)114 (2.1)117 (2.0)0.740 CRRT, n (%)826 (15.5)1034 (18.1)< 0.001 RRT, n (%)1050 (19.8)1351 (23.6)< 0.001 Days in Hospital, (median[IQR])12.60 [7.40, 21.80]12.80 [7.60, 22.10]0.247 Days in ICU, (days), (median[IQR])5.00 [3.10, 9.30]5.10 [3.10, 9.60]0.524 ICU mortality, n (%)1232 (23.2)1312 (22.9)0.769 Hospital mortality, n (%)1582 (29.8)1671 (29.2)0.531BMI: Body Mass Index, SOFA: Sequential Organ Failure Assessment, APACHE II: Acute Physiology and Chronic Health Evaluation II, InvasiveVent: Invasive Ventilation, CPR: Cardiopulmonary Resuscitation, CRRT: Continuous Renal Replacement Therapy, RRT: Renal Replacement Therapy, IQR: Interquartile Range, P < 0.05 was considered statistically significant Study flow diagram and cohort derivation. A Flow diagram of the internal validation cohort. Among 16,523 ICU admissions, patients were excluded for ICU readmission during the same hospitalization, missing SOFA components, or ICU survival < 24 h, leaving 11,669 unique ICU patients. Sepsis was defined as infection plus an acute increase ≥ 2 points in SOFA-1 or SOFA-2 within 24 h of ICU admission. Numbers of patients classified as septic by SOFA-1 only, SOFA-2 only, or by both scores are shown. The Venn diagram depicts overlap between sepsis-SOFA-1 and sepsis-SOFA-2 within the internal cohort. B Flow diagram of the external validation cohort (MIMIC-IV). Among 65,366 ICU stays, exclusions for short ICU stays and missing SOFA components yielded 29,811 first ICU admissions. The numbers of sepsis patients by SOFA-1 and SOFA-2 criteria and their overlap are presented, with a Venn diagram illustrating unique and shared sepsis classifications Table 1Baseline data and outcome indicators of sepsis patientsVariableSepsis-SOFA1 (n = 5313)Sepsis-SOFA2(n = 5717)PAge, years,(median[IQR])66.80 [55.40, 77.20]66.80 [55.30, 77.20]0.837Gender, male, n (%)3173 (59.7)3414 (59.7)1.000BMI, (median[IQR])28.10 [24.10, 33.20]28.10 [24.10, 33.20]0.987Comorbidity, n (%) Charlson, (median[IQR])6.00 [4.00, 8.00]6.00 [4.00, 8.00]0.431 Hypertension1840 (34.6)1970 (34.5)0.857 Diabetes1762 (33.2)1960 (34.3)0.219 Myocardial infarct1121 (21.1)1232 (21.5)0.577 Congestive heart failure2053 (38.6)2228 (39.0)0.725 Mild liver disease1364 (25.7)1424 (24.9)0.357 Severe liver disease816 (15.4)839 (14.7)0.324 Chronic kidney disease1519 (28.6)1698 (29.7)0.201 Chronic pulmonary disease1427 (26.9)1529 (26.7)0.897 Cerebrovascular disease698 (13.1)771 (13.5)0.594 Immunosuppression1218 (22.9)1321 (23.1)0.839Illness severity scores, (median[IQR]) SOFA1 score4.00 [2.00, 7.00]4.00 [3.00, 7.00]< 0.001 SOFA2 score5.00 [3.00, 8.00]5.00 [3.00, 8.00]0.619 APACHE II score20.00 [14.00, 28.00]20.00 [14.00, 28.00]0.766Infection source, n (%) Lung infection2316 (43.6)2659 (46.5)0.002 Abdominal infection403 (7.6)468 (8.2)0.244 Urinary infection448 (8.4)478 (8.4)0.918 Bloodstream infection238 (4.5)254 (4.4)0.927 Skin and soft tissue infection416 (7.8)412 (7.2)0.219Pathogenic microorganisms, n (%) Klebsiella pneumoniae1201 (22.6)1410 (24.7)0.011 Acinetobacter baumannii996 (18.7)1169 (20.4)0.026 Pseudomonas aeruginosa603 (11.3)656 (11.5)0.857 Staphylococcus aureus1145 (21.6)1214 (21.2)0.693 Escherichia coli919 (17.3)1041 (18.2)0.213Outcome Septic shock, n (%)1727 (32.5)1816 (31.8)0.414 Vasoactive drugs, n (%)2962 (55.8)2992 (52.3)< 0.001 InvasiveVent, n (%)4072 (76.6)4265 (74.6)0.013 CPR, n (%)114 (2.1)117 (2.0)0.740 CRRT, n (%)826 (15.5)1034 (18.1)< 0.001 RRT, n (%)1050 (19.8)1351 (23.6)< 0.001 Days in Hospital, (median[IQR])12.60 [7.40, 21.80]12.80 [7.60, 22.10]0.247 Days in ICU, (days), (median[IQR])5.00 [3.10, 9.30]5.10 [3.10, 9.60]0.524 ICU mortality, n (%)1232 (23.2)1312 (22.9)0.769 Hospital mortality, n (%)1582 (29.8)1671 (29.2)0.531BMI: Body Mass Index, SOFA: Sequential Organ Failure Assessment, APACHE II: Acute Physiology and Chronic Health Evaluation II, InvasiveVent: Invasive Ventilation, CPR: Cardiopulmonary Resuscitation, CRRT: Continuous Renal Replacement Therapy, RRT: Renal Replacement Therapy, IQR: Interquartile Range, P < 0.05 was considered statistically significant Baseline data and outcome indicators of sepsis patients BMI: Body Mass Index, SOFA: Sequential Organ Failure Assessment, APACHE II: Acute Physiology and Chronic Health Evaluation II, InvasiveVent: Invasive Ventilation, CPR: Cardiopulmonary Resuscitation, CRRT: Continuous Renal Replacement Therapy, RRT: Renal Replacement Therapy, IQR: Interquartile Range, P < 0.05 was considered statistically significant Trends in incidence Comparison of total SOFA distributions revealed that SOFA-2 assigned higher scores across the mid-to-high range (4–21 points), particularly between 4–9 and 13–21 points (Fig. 2A). These patterns were reproduced in the external cohort (eFigure 2). Organ-system analyses across all patients revealed both overlapping and non-overlapping components between SOFA-1 and SOFA-2 scores: while SOFA-2 consistently assigned higher neurologic, renal, and hemostasis scores in both internal and external validation cohorts, SOFA-1 conversely assigned higher respiratory scores, highlighting distinct diagnostic contributions from each scoring system alongside their shared components in sepsis diagnosis (Fig. 2B and eFigure 3). Similar organ-system patterns were observed within overlapping SOFA-1, SOFA-2 sepsis cases (eFigure 4A-B).Fig. 2Distribution of total SOFA-1 and SOFA-2 scores and organ dysfunction contributions. A Histogram of total SOFA-1 and SOFA-2 scores among sepsis patients in the internal cohort, expressed as the proportion of patients within each score category. B Corresponding mean organ-system scores for SOFA-1 vs. SOFA-2 in the non-overlapping and overlapping internal cohort. C Corresponding mean organ-system scores for SOFA-1 vs. SOFA-2 in the non-overlapping internal cohort Distribution of total SOFA-1 and SOFA-2 scores and organ dysfunction contributions. A Histogram of total SOFA-1 and SOFA-2 scores among sepsis patients in the internal cohort, expressed as the proportion of patients within each score category. B Corresponding mean organ-system scores for SOFA-1 vs. SOFA-2 in the non-overlapping and overlapping internal cohort. C Corresponding mean organ-system scores for SOFA-1 vs. SOFA-2 in the non-overlapping internal cohort In the internal cohort, 115 patients met sepsis criteria exclusively by SOFA-1 and 519 exclusively by SOFA-2 (eTable 5). SOFA-2-exclusive cases had higher illness severity and worse clinical outcomes than SOFA-1-exclusive cases (all P < 0.005). In the internal validation cohort and external validation cohort (Fig. 2C and eFigure 5), SOFA-2 score-exclusive patients with sepsis had higher organ dysfunction scores of brains, kidney, and cardiovascular (all, P < 0.001). Diagnostic agreement between SOFA-1 and SOFA-2 was substantial in both cohorts (internal κ = 0.891; external κ = 0.908), as shown in the Sankey and Venn diagrams (eFigure 6A–B). Despite this strong agreement, SOFA-2 newly classified 9.1% of internal-cohort patients and 5.9% external cohort patients as septic, whereas 2.2% and 0.6% patients, respectively, were classified as septic only by SOFA-1. Trends in, and risk factors for, acute ICU mortality In both cohorts, ICU mortality increased progressively with rising SOFA-1 and SOFA-2 total scores. In the internal validation cohort, among patients with a SOFA score of 20 points, ICU mortality was 66.7% in the SOFA‑2 group and 50.0% in the SOFA‑1 group (P = 1.000); among those with a SOFA score of 21 points, ICU mortality was 83.3% in the SOFA‑2 group and 50.0% in the SOFA‑1 group (P = 0.4667) (Fig. 3A, B). In the external validation cohort, among patients with SOFA scores of 23–24 points, ICU mortality in the SOFA‑2 group was 66.7% and 100%, respectively; no patients with SOFA‑1 scores of 23–24 points were observed (eFigure 7A-B). SOFA-2 assigned significantly higher risk for moderate-to-severe dysfunction (scores 3–4) in cardiovascular, respiratory, and other systems (eFigure 8A-B).Fig. 3ICU mortality according to total SOFA-1 and SOFA-2 scores in the internal cohort. A ICU mortality proportion (with 95% confidence intervals) across total SOFA score categories for SOFA-1 and SOFA-2 in the internal cohort. B Smoothed mortality curves for ICU mortality as a function of total SOFA-1 and SOFA-2 scores in the internal cohort, with patient counts displayed below the x-axis ICU mortality according to total SOFA-1 and SOFA-2 scores in the internal cohort. A ICU mortality proportion (with 95% confidence intervals) across total SOFA score categories for SOFA-1 and SOFA-2 in the internal cohort. B Smoothed mortality curves for ICU mortality as a function of total SOFA-1 and SOFA-2 scores in the internal cohort, with patient counts displayed below the x-axis RCS analyses with 4 knots (based on quantiles) revealed significant non-linear relationships between organ dysfunction components and mortality (all P values for non-linearity < 0.01 except for mean blood pressure. In the SOFA-2 cohort compared with SOFA-1, differing mortality patterns were observed: neurological dysfunction (GCS score) showed a higher baseline risk, respiratory failure (PaO2 FiO2−1) demonstrated greater hypoxemia-associated risk, and hepatic dysfunction (bilirubin) exhibited a slightly attenuated association at high levels. Renal dysfunction (creatinine) had a marginally stronger association in SOFA-2, while mean blood pressure showed no significant differences (Fig. 4). These patterns suggest SOFA-2 may provide refined mortality prediction for certain organ failures while maintaining comparable performance for others (eTable 6). These findings were consistent in the external validation cohort (eFigure 9, eTable 7).Fig. 4Restricted cubic spline (RCS) analyses of SOFA component variables and ICU mortality: internal cohort. Restricted cubic spline models illustrating the exploratory non-linear associations between key SOFA component variables and ICU mortality in the internal cohort, shown separately for patients meeting Sepsis-3 criteria using SOFA-1 vs. SOFA-2. Patients display the estimated adjusted log-odds (or odds ratio, as labeled) of ICU mortality across the observed range of each variable with 95% confidence intervals, using 4 knots placed at prespecified quantiles. Component variables include Glasgow Coma Scale (GCS), PaO2/FiO2, total bilirubin, serum creatinine, platelet count, and mean arterial pressure (MAP) Restricted cubic spline (RCS) analyses of SOFA component variables and ICU mortality: internal cohort. Restricted cubic spline models illustrating the exploratory non-linear associations between key SOFA component variables and ICU mortality in the internal cohort, shown separately for patients meeting Sepsis-3 criteria using SOFA-1 vs. SOFA-2. Patients display the estimated adjusted log-odds (or odds ratio, as labeled) of ICU mortality across the observed range of each variable with 95% confidence intervals, using 4 knots placed at prespecified quantiles. Component variables include Glasgow Coma Scale (GCS), PaO2/FiO2, total bilirubin, serum creatinine, platelet count, and mean arterial pressure (MAP) Assessment of predictive validity In the internal cohort, non-survivors had significantly higher SOFA-1, SOFA-2, and APACHE II scores; higher lactate; and more frequent use of vasoactive drugs, mechanical ventilation, and CRRT than survivors (all P < 0.001; eTable 8). Multivariable logistic regression (adjusted for baseline variables, clinical frailty scale, and laboratory parameters; eTable 8) with backward stepwise selection identified SOFA-1 and SOFA-2 as independent predictors of in-ICU mortality in sepsis patients (all P < 0.05). Both internal and external validation confirmed SOFA-2 as an independent risk factor for ICU mortality in critically ill patients (eTable 9). The ROC analyses demonstrated that SOFA-2 achieved moderately higher AUCs than SOFA-1 (internal 0.720 vs. 0.700; external 0.730 vs. 0.720) and outperformed APACHE II and SAPS II (eFigure 10 A–B). In the internal cohort, SOFA-2 showed significantly better discriminatory performance than SOFA-1 (DeLong’s test, P = 0.0083). This was confirmed in the external validation cohort (P = 0.00028). At optimal cut-offs, SOFA-2 offered higher sensitivity with similar specificity, resulting in improved negative predictive value for ruling out ICU mortality (eTables 10–11). Discussion In this multicenter retrospective observational cohort study with external validation, SOFA-2 identified a moderately larger sepsis population than SOFA-1 while maintaining substantial diagnostic agreement. The additional cases captured by SOFA-2 exhibited higher illness severity and worse clinical outcomes than SOFA-1-exclusive cases, which may support the potential clinical relevance of SOFA-2 reclassification. SOFA-2 also appeared to demonstrate modestly improved discrimination for ICU mortality and clearer risk gradients in moderate-to-severe organ dysfunction categories. The broader sepsis population identified with SOFA-2, also observed in recent reports such as Ranzani et al. [8] may reflect changes in sepsis phenotypes over the last two decades. SOFA-2 incorporates non-invasive ventilation, HFNC, IABP, ECMO, updated vasoactive thresholds, and explicit coding for CRRT. These revisions may enhance the score’s sensitivity to clinically relevant neurologic, renal, cardiovascular, and respiratory dysfunctions. Our results suggest that these differences are not merely classificatory: SOFA-2 may more consistently identify patients with demonstrably higher severity, suggesting that SOFA-1 may underrecognize evolving multiorgan failure until more advanced dysfunction or invasive support emerges. This may provide a mechanistic rationale for the expansion in SOFA-2-defined sepsis incidence while maintaining clinical coherence. While SOFA-2’s broader capture of organ dysfunction may improve sensitivity, its use in ICUs with aggressive application of organ support therapies (such as early CRRT initiation, escalating vasopressor strategies, HFNC, or ECMO) raises valid concerns regarding potential overclassification. In such settings, support modalities may reflect proactive clinical management rather than true progression of sepsis-related organ failure. For instance, short-term HFNC for transient hypoxemia could potentially elevate SOFA-2 scores in patients without a dysregulated host response. However, the pattern observed in our cohorts suggests that this risk may be limited. The patients newly identified by SOFA-2 consistently demonstrated higher illness severity, more advanced organ dysfunction, indicating that SOFA-2 tends to uncover clinically relevant pathology rather than inflate scores due to device use alone. Thus, while cautious interpretation may be warranted in highly protocolized ICUs, our data suggest that SOFA-2’s expanded scope may primarily identify patients with genuine physiological deterioration rather than those receiving support for non-septic indications. From a prognostic perspective, SOFA-2 may provide modestly improved mortality discrimination compared with SOFA-1. Because SOFA-1 and SOFA-2 are point‑based severity scores rather than fitted prediction models, we assessed discrimination using unadjusted AUROC based on rank‑ordering of patients by total score. In this context, the consistent AUROC advantage observed for SOFA-2 across both cohorts may support incremental improvement in bedside risk stratification, even though absolute differences were small. SOFA-1 places substantial weight on invasive ventilation-based thresholds and static cut-offs for organ dysfunction, which may delay recognition of progressive injury in patients receiving modern supportive therapies. SOFA-2’s recalibrated thresholds and broader inclusion of respiratory and renal support modalities may enhance recognition of clinically meaningful impairment. The associations observed in non-linear (RCS) analyses for SOFA-2 across neurologic, oxygenation, bilirubin, creatinine, and platelet domains suggest that SOFA-2 may more closely capture the inflection points where physiologic deterioration translates to mortality risk. By contrast, SOFA-1’s reliance on creatinine alone and absence of CRRT coding may obscure severe kidney failure, potentially explaining why SOFA-1-defined sepsis paradoxically demonstrated higher mortality in the creatinine curves despite lower overall severity [10, 11]. Exploratory RCS analyses characterized non-linear associations between selected physiologic variables and ICU mortality, requiring cautious interpretation. Since variables like GCS or creatinine contribute to both SOFA versions, these analyses aimed not to compare score superiority but to reveal clinically plausible relationships in sepsis patients. The SOFA-2-exclusive group exhibited consistently higher illness severity, worse organ dysfunction compared with SOFA-1-exclusive cases, suggesting that SOFA-2 may uncover a high-risk segment of the sepsis population that SOFA-1 may miss. Conversely, although SOFA-1-exclusive cases represented a small proportion of patients, their non-trivial mortality may suggest the presence of early biochemical or respiratory abnormalities that do not yet trigger SOFA-2 thresholds. This subgroup may be more relevant in pre-ICU settings (ED, wards), where organ support therapies are not yet initiated. Therefore, while SOFA-2 may provide improved alignment with ICU-level physiology, clinicians should remain aware of its potential to delay detection of early organ dysfunction outside of ICU environments. Whether the Sepsis‑3 criterion of SOFA ≥ 2 may require recalibration when SOFA-2 replaces SOFA-1 is an important question. Our findings suggest that no adjustment of this threshold may be necessary. Although SOFA-2 identified 9.1% more sepsis cases than SOFA-1, these additional patients consistently demonstrated higher illness severity, greater organ‑support requirements, and higher mortality, suggesting that SOFA-2 may be detecting clinically meaningful organ dysfunction rather than broadening the definition to include milder illness. Importantly, overall outcomes among SOFA-2-defined and SOFA-1-defined sepsis groups remained comparable, suggesting that SOFA-2 may expand detection without diluting risk. At higher severity thresholds, organ-specific risk stratification may diverge significantly. SOFA-2 may demonstrate superior discrimination of mortality risk in moderate-to-severe organ dysfunction, potentially better reflecting the prognostic burden of advanced organ failure—a finding that may align with modern critical care priorities. In contrast, SOFA-1 may show broader applicability in less severe illness, consistent with its derivation from heterogeneous critical illness populations, and may remain robust for general risk stratification. These results may underscore the importance of precision in SOFA selection, particularly for sepsis management targeting high-risk phenotypes. Our findings may extend prior evidence of SOFA-1's broad prognostic utility by delineating SOFA-2's added value in severe organ dysfunction [12–18], reinforcing SOFA scores' established role in sepsis outcomes [19, 20]. Limitations This study has several limitations. First, the retrospective design and reliance on ICU databases may limit generalizability to non-ICU settings, where early sepsis recognition is crucial, as well as to non-critically ill populations. Second, we excluded patients who died within 24 h of ICU admission to enable SOFA calculation; we fully acknowledge that this exclusion may have systematically removed the most critically ill patients, and that this may limit the validity of mortality comparisons at extremely high SOFA score levels. In future prospective studies, we will collect real‑time, high‑density physiological data to enable valid SOFA scoring even in patients with early death, allowing a more comprehensive assessment of SOFA‑2 performance in the most critically ill populations. Third, while SOFA-2 improves alignment with current organ support practices, it remains a physiology-based score and does not incorporate immune, endothelial, or gastrointestinal dysfunctions, domains increasingly recognized as central to sepsis pathophysiology. Fourth, while our analyses demonstrated statistically significant differences between SOFA-1 and SOFA-2 in sepsis identification and mortality prediction, the absolute improvements in prognostic performance (e.g., AUC increments of 0.017–0.013) were modest. This underscores that SOFA-2’s clinical relevance may lie primarily in its updated alignment with modern organ support practices rather than transformative prognostic gains. Fifth, furthermore, despite adjusting for key confounders in multivariable models, the multiple comparisons undertaken across organ systems and subgroups (e.g., SOFA-1–exclusive vs. SOFA-2–exclusive sepsis) increase the risk of type I errors. We mitigated this by prespecifying primary endpoints and using conservative statistical methods (e.g., DeLong’s test for AUC comparisons), but cautious interpretation of secondary findings is warranted. Sixth, patients diagnosed with sepsis prior to ICU admission may have been misclassified as non-septic in our dataset if early SOFA changes were not captured, potentially attenuating differences between scoring systems. [12, 14, 17–19]. Finally, we acknowledge that ceiling-of-treatment or limitation-of-life-sustaining-therapy decisions may influence organ support usage and SOFA scoring. Data regarding these decisions were not available in this study, which may introduce potential bias and limit the interpretability of the results. In future prospective studies, we will systematically collect detailed data on ceiling-of-treatment decisions and palliative care integration to further clarify their potential impact on SOFA‑2 scoring and prognostic performance. Conclusion SOFA-2 represents a potentially valuable update to the SOFA framework. Compared with SOFA-1, SOFA-2 may identify a moderately broader sepsis population with more advanced organ dysfunction and may offer modestly improved ICU mortality discrimination. These findings support the potential consideration of SOFA-2 for ICU sepsis assessment and suggest that Sepsis-3 diagnostic thresholds may be applied without recalibration. Supplementary Information Supplementary material 1. eFigure 1 Correlation matrix of key variables after multiple imputations. Heatmap of pairwise correlations among major clinical, laboratory, and severity-score variables in the internal cohort after multiple imputation. Colors indicate the strength and direction of Pearson correlation coefficients, demonstrating plausible relationships and supporting the adequacy of the imputation mode. eFigure 2 Distribution of SOFA scores in sepsis by SOFA-1 and SOFA-2 with the external validation cohort. eFigure 3 Mean organ-system scores for cardiovascular, brain, hemostasis, liver, kidney, and respiratory components in sepsis patients classified by SOFA-1 and SOFA-2 in the external cohort. eFigure 4 Comparison of organ-system SOFA-1 and SOFA-2 scores in overlapping sepsis groups. eFigure4A: Internal cohort: mean system-specific SOFA-1 and SOFA-2 scores among patients meeting both sepsis-SOFA-1 and sepsis-SOFA-2 criteria. eFigure 4B: External cohort: analogous comparison for respiratory, hemostasis, liver, cardiovascular, brain, and kidney components. eFigure 5 Comparison of organ-system SOFA-1 and SOFA-2 scores in non-overlapping sepsis groups in external cohort. eFigure 6 Agreement and reclassification of sepsis between SOFA-1 and SOFA-2. Sankey diagram for the internal cohort showing concordance and reclassification of sepsis status between SOFA-1 and SOFA-2. eFigure 6 A: Flows depict transitions from non-sepsis and sepsis by SOFA-1 to corresponding categories by SOFA-2; the overall weighted kappa is 0.891, with 8.2% of patients reclassified. eFigure 6 B: Venn diagram for the external cohort displaying the number of sepsis patients classified by SOFA-1 only, SOFA-2 only, and both scores. Agreement is high, with 0.6% SOFA-1-exclusive and 2.2% SOFA-2-exclusive sepsis patients. eFigure 7 ICU mortality according to SOFA-1 and SOFA-2 total and organ-system scores in the external validation cohort. eFigure 7A: ICU mortality by total SOFA-1 and SOFA-2 scores in the internal cohort. Bars show mortality proportion for each score category. eFigure 7B: Smoothed mortality curves for ICU mortality vs. total SOFA-1 and SOFA-2 scores. eFigure 8 Heatmap of ICU mortality by organ-system scorefor SOFA-1 and SOFA-2 in the internal cohort and the external validation cohort. eFigure 9 Association between SOFA-related clinical parameters and ICU mortality in the external validation cohort Restricted cubic spline models in the external cohort mirroring eFigure 9A: Glasgow Coma Scale; eFigure 9B: PaO2/FiO2 ratio; eFigure 9 C: total bilirubin; eFigure 9 D: creatinine; eFigure 9E: platelet count; eFigure 9 F: mean arterial pressure. Curves compare ICU mortality probabilities for patients classified as sepsis-SOFA-1 vs. sepsis-SOFA-2, confirming the internal-cohort patterns. eFigure 10 Discriminative Performance of SOFA-1 vs. SOFA-2 in Predicting ICU Mortality. eFigure 10A: ROC curves for SOFA-1, SOFA-2, and APACHE II in the internal cohort. eFigure 10B: ROC curves for SOFA-1, SOFA-2, and SAPS II in the external cohort. SOFA-2 consistently yields the highest AUC and greater sensitivity at optimal cut-offs. eTable 1 Operational Definitions and Coding Criteria for SOFA-2 Components by Organ System. eTable 2 Missing Variables in Internal and External Validation Cohorts. eTable 3 Baseline character according to SOFA-1 score diagnosis of sepsis and non-sepsis. eTable 4 Baseline character according to SOFA-2 score diagnosis of sepsis and non-sepsis. eTable 5 Baseline data of non-overlapping parts of sepsis diagnosed by SOFA-1 and SOFA-2 scores. eTable 6 Restricted Cubic Spline Analysis of Clinical Variables and ICU Mortality in Internal Validation Cohort. eTable 7 Restricted Cubic Spline Analysis of Clinical Variables and ICU Mortality in External Validation Cohort. eTable 8 Baseline characteristics of sepsis patients with different SOFA scores in survival and non-survival groups. eTable 9 Multivariable Logistic Regression Analysis of ICU Mortality in Patients with Sepsis. eTable 10 Assessment of the Prognostic Value of Disease Severity Scores in internal validation cohort of patients with sepsis. eTable 11 Assessment of the Prognostic Value of Disease Severity Scores in external validation cohort of patients with sepsis. Supplementary material 1. eFigure 1 Correlation matrix of key variables after multiple imputations. Heatmap of pairwise correlations among major clinical, laboratory, and severity-score variables in the internal cohort after multiple imputation. Colors indicate the strength and direction of Pearson correlation coefficients, demonstrating plausible relationships and supporting the adequacy of the imputation mode. eFigure 2 Distribution of SOFA scores in sepsis by SOFA-1 and SOFA-2 with the external validation cohort. eFigure 3 Mean organ-system scores for cardiovascular, brain, hemostasis, liver, kidney, and respiratory components in sepsis patients classified by SOFA-1 and SOFA-2 in the external cohort. eFigure 4 Comparison of organ-system SOFA-1 and SOFA-2 scores in overlapping sepsis groups. eFigure4A: Internal cohort: mean system-specific SOFA-1 and SOFA-2 scores among patients meeting both sepsis-SOFA-1 and sepsis-SOFA-2 criteria. eFigure 4B: External cohort: analogous comparison for respiratory, hemostasis, liver, cardiovascular, brain, and kidney components. eFigure 5 Comparison of organ-system SOFA-1 and SOFA-2 scores in non-overlapping sepsis groups in external cohort. eFigure 6 Agreement and reclassification of sepsis between SOFA-1 and SOFA-2. Sankey diagram for the internal cohort showing concordance and reclassification of sepsis status between SOFA-1 and SOFA-2. eFigure 6 A: Flows depict transitions from non-sepsis and sepsis by SOFA-1 to corresponding categories by SOFA-2; the overall weighted kappa is 0.891, with 8.2% of patients reclassified. eFigure 6 B: Venn diagram for the external cohort displaying the number of sepsis patients classified by SOFA-1 only, SOFA-2 only, and both scores. Agreement is high, with 0.6% SOFA-1-exclusive and 2.2% SOFA-2-exclusive sepsis patients. eFigure 7 ICU mortality according to SOFA-1 and SOFA-2 total and organ-system scores in the external validation cohort. eFigure 7A: ICU mortality by total SOFA-1 and SOFA-2 scores in the internal cohort. Bars show mortality proportion for each score category. eFigure 7B: Smoothed mortality curves for ICU mortality vs. total SOFA-1 and SOFA-2 scores. eFigure 8 Heatmap of ICU mortality by organ-system scorefor SOFA-1 and SOFA-2 in the internal cohort and the external validation cohort. eFigure 9 Association between SOFA-related clinical parameters and ICU mortality in the external validation cohort Restricted cubic spline models in the external cohort mirroring eFigure 9A: Glasgow Coma Scale; eFigure 9B: PaO2/FiO2 ratio; eFigure 9 C: total bilirubin; eFigure 9 D: creatinine; eFigure 9E: platelet count; eFigure 9 F: mean arterial pressure. Curves compare ICU mortality probabilities for patients classified as sepsis-SOFA-1 vs. sepsis-SOFA-2, confirming the internal-cohort patterns. eFigure 10 Discriminative Performance of SOFA-1 vs. SOFA-2 in Predicting ICU Mortality. eFigure 10A: ROC curves for SOFA-1, SOFA-2, and APACHE II in the internal cohort. eFigure 10B: ROC curves for SOFA-1, SOFA-2, and SAPS II in the external cohort. SOFA-2 consistently yields the highest AUC and greater sensitivity at optimal cut-offs. eTable 1 Operational Definitions and Coding Criteria for SOFA-2 Components by Organ System. eTable 2 Missing Variables in Internal and External Validation Cohorts. eTable 3 Baseline character according to SOFA-1 score diagnosis of sepsis and non-sepsis. eTable 4 Baseline character according to SOFA-2 score diagnosis of sepsis and non-sepsis. eTable 5 Baseline data of non-overlapping parts of sepsis diagnosed by SOFA-1 and SOFA-2 scores. eTable 6 Restricted Cubic Spline Analysis of Clinical Variables and ICU Mortality in Internal Validation Cohort. eTable 7 Restricted Cubic Spline Analysis of Clinical Variables and ICU Mortality in External Validation Cohort. eTable 8 Baseline characteristics of sepsis patients with different SOFA scores in survival and non-survival groups. eTable 9 Multivariable Logistic Regression Analysis of ICU Mortality in Patients with Sepsis. eTable 10 Assessment of the Prognostic Value of Disease Severity Scores in internal validation cohort of patients with sepsis. eTable 11 Assessment of the Prognostic Value of Disease Severity Scores in external validation cohort of patients with sepsis.