Patient-level reclassification and discordance between SOFA and SOFA-2: a multicenter ICU study

BACKGROUND: The Sequential Organ Failure Assessment (SOFA) score is widely used to quantify organ dysfunction in critically ill patients. A revised version, SOFA-2, was proposed to better reflect contemporary intensive care practice, but patient-level reclassification and the implications of differences between SOFA and SOFA-2 remain unclear. METHODS: We conducted a retrospective multicenter cohort study using the OneICU database, a Japanese ICU registry including 15 hospitals. Adult ICU admissions between April 2012 and December 2025 were analyzed. Day-1 SOFA and SOFA-2 scores were calculated. Reclassification patterns were visualized using Sankey diagrams, and mortality patterns were assessed using heat maps. Score discordance was defined as SOFA deviation (dSOFA = SOFA-2 - SOFA) and categorized as Negative ( ≤ - 2), Minimal ( > - 2 and < 2), or Positive (≥ 2). Two primary exact-matching analyses were performed across dSOFA categories, separately matching patients on Day-1 SOFA-2 and conventional SOFA total scores. RESULTS: Among 134,528 ICU admissions, SOFA-2 resulted in systematic patient-level reclassification. Upward shifts were most frequent in the brain, cardiovascular, and respiratory subscores, whereas the kidney subscore showed bidirectional changes. Mortality increased with higher SOFA-2 subscores at fixed SOFA and with higher SOFA subscores at fixed SOFA-2. After exact matching on SOFA-2, mortality was highest in the Negative dSOFA group, in which SOFA was higher than SOFA-2. Conversely, after exact matching on conventional SOFA, mortality was highest in the Positive dSOFA group, in which SOFA-2 was higher than SOFA. CONCLUSIONS: SOFA-2 introduces systematic reclassification at the patient level. Reciprocal exact-matching analyses demonstrated that patients assigned the same severity by one scoring system remained heterogeneous in mortality according to the other scoring system. SOFA and SOFA-2 may therefore provide overlapping but complementary information for assessing organ dysfunction severity.