Comparison of the prognostic performance of SOFA-1 and SOFA-2 scores in older critically ill patients: a cohort study

The Sequential Organ Failure Assessment (SOFA) score is one of the most widely used tools for evaluating multiple organ dysfunction or failure in critically ill patients. In October 2025, an updated version (SOFA-2) was introduced to better align with contemporary critical care practice. However, the comparative prognostic performance and clinical interpretation of SOFA-2 relative to the original SOFA score (referred to as SOFA-1 in this study) remain uncertain in older critically ill populations. This study aimed to compare the predictive performance of SOFA-1 and SOFA-2 for intensive care unit (ICU) mortality, in-hospital mortality, and 28-day mortality among older critically ill patients. Older critically ill patients from the Medical Information Mart for Intensive Care IV version 3.1 database were included as the cohort for primary analysis, and those from the eICU Collaborative Research Database were included as an independent validation cohort. The prognostic performance of SOFA-1 and SOFA-2 was evaluated separately in each cohort. All ICU patients aged 65–89 years were included, as exact age information was unavailable for patients older than 89 years because age values above 89 years were top-coded in both databases. However, patients with top-coded ages above 89 years were included in sensitivity analyses. Receiver operating characteristic (ROC) curves were constructed to evaluate the predictive performance of the SOFA-1 and SOFA-2 scores for ICU, in-hospital, and 28-day mortality. Multivariable logistic regression was applied to assess the associations between SOFA scores and mortality outcomes. Discrimination of SOFA-1 and SOFA-2 was evaluated using the area under the ROC curve (AUROC), while discrimination of the adjusted multivariable logistic models was assessed using the concordance index (C-index). Differences in AUROC and C-index were compared using the DeLong test. A total of 26,104 and 53,697 older critically ill patients from the MIMIC-IV and eICU-CRD databases were included, respectively. In the MIMIC-IV database, the total SOFA-2 score showed a statistically higher AUROC than SOFA-1 for ICU mortality, although the absolute difference was modest (AUROC: 0.749 vs. 0.730, P -value < 0.001). Similar modest differences were observed for in-hospital mortality and 28-day mortality, and the overall patterns were largely replicated in the eICU-CRD database. SOFA-2 generally showed higher sensitivity and negative predictive value, whereas SOFA-1 showed higher specificity, positive predictive value, and positive likelihood ratios. In fully adjusted multivariable logistic regression analysis across both databases, inclusion of SOFA-2 yielded significantly higher C-indices than inclusion of SOFA-1 for the total score, the respiratory domain, and the hemostasis domain in predicting ICU and in-hospital mortality. For fully adjusted multivariable logistic regression analysis of 28-day mortality, inclusion of SOFA-2 also showed significantly higher C-indices than inclusion of SOFA-1 for the total score as well as the respiratory and hemostasis domains. SOFA-2 demonstrated broadly comparable prognostic performance to SOFA-1 in older critically ill patients, with modestly higher discrimination for some outcomes and domains, particularly the total score and respiratory component. Its higher sensitivity and negative predictive value suggest potential utility for early risk stratification, but this should be interpreted alongside lower specificity and positive predictive value.