SOFA-2 versus SOFA for mortality risk assessment in ICU patients with acute cholangitis: a retrospective cohort study using MIMIC-IV and eICU

Technical Summary: SOFA-2 versus SOFA for Mortality Risk Assessment in ICU Patients with Acute Cholangitis

Problem Statement

Acute cholangitis (AC) is a life-threatening biliary infection that can rapidly progress to organ failure and death. While the Tokyo Guidelines 2018 (TG18) provide a framework for diagnosing and grading AC severity, they do not generate continuous mortality-risk estimates or directly compare organ-dysfunction scores. The Sequential Organ Failure Assessment (SOFA) score is the standard for quantifying organ dysfunction in sepsis and critical care. However, a revised version, SOFA-2, has been developed using modern critical-care data to update organ-dysfunction scoring. It remains unknown whether SOFA-2 offers disease-specific prognostic value beyond the established SOFA score specifically for patients with acute cholangitis. This study addresses the gap in comparative evaluation of these two scores within the specific context of biliary sepsis.

Methodology

The authors conducted a retrospective cohort study utilizing two large electronic health record databases: MIMIC-IV (v3.1) as the primary dataset and the eICU Collaborative Research Database for cross-database assessment.

• Cohort Selection: Adult patients (≥18 years) with acute cholangitis meeting operationalized TG18 diagnostic criteria were included. In MIMIC-IV, criteria were defined using diagnosis codes combined with structured evidence of systemic inflammation (abnormal WBC) and hepatobiliary abnormalities (elevated bilirubin, aminotransferases, or alkaline phosphatase). The primary analysis cohort consisted of 644 patients with at least one ICU stay. The eICU cohort included 289 patients.
• Score Reconstruction: Day-1 SOFA-2 and SOFA scores were reconstructed based on the most abnormal values recorded within the first 24 hours of ICU admission. Unobserved components were assigned 0 points, consistent with standard database reconstruction practices.
• Outcomes: The primary outcome was ICU mortality. Secondary outcomes included in-hospital mortality and 28-day all-cause mortality.
• Statistical Analysis: Discrimination was evaluated using the Area Under the Receiver Operating Characteristic Curve (AUROC) with 5,000 bootstrap samples. The study compared SOFA-2 and SOFA as paired scores. Additional analyses included calibration curves, Brier scores, decision-curve analysis (DCA), and reclassification metrics. Subgroup and sensitivity analyses were performed to test robustness across different phenotypes and data completeness.

Key Results

• Cohort Characteristics: In the MIMIC-IV cohort, ICU mortality was 11.0%, in-hospital mortality was 15.8%, and 28-day mortality was 20.7%. Non-survivors were more likely to have TG18 Grade III disease and required more intensive organ support (ventilation, vasopressors, renal replacement therapy).
• Discrimination Performance:
  • ICU Mortality: SOFA-2 demonstrated good discrimination (AUROC 0.834; 95% CI, 0.777–0.883). However, its performance was statistically indistinguishable from SOFA (AUROC 0.838; 95% CI, 0.784–0.886). The difference was −0.005 (95% CI, −0.020 to 0.010; P = 0.514).
  • Secondary Outcomes: For in-hospital mortality, AUROCs were 0.786 (SOFA-2) vs. 0.784 (SOFA). For 28-day mortality, AUROCs were 0.760 (SOFA-2) vs. 0.756 (SOFA).
• Risk Stratification: Both scores showed a clear gradient of increasing mortality risk with higher scores. ICU mortality rose from 1.9% at SOFA-2 scores of 0–2 to 27.9% at scores ≥9. A maximally selected rank statistic identified SOFA-2 ≥10 as an optimal threshold for survival separation.
• Cross-Database Validation: In the eICU cohort, SOFA-2 and SOFA again showed nearly identical discrimination (AUROC 0.833 vs. 0.835), and the mortality gradient across score strata was reproduced.
• Calibration and Utility: Calibration curves, Brier scores, and decision-curve analyses did not reveal a clinically meaningful advantage for SOFA-2 over SOFA. Reclassification analyses confirmed no incremental value.

Key Contributions

1. Disease-Specific Evaluation: This is the first study to directly compare SOFA-2 and SOFA specifically within the population of ICU patients with acute cholangitis, a condition where biliary obstruction and infection interact closely with organ dysfunction.
2. Comparative Rigor: The study utilized a large, multi-center dataset (MIMIC-IV) for primary analysis and an independent database (eICU) for cross-database assessment, employing robust statistical methods including bootstrap validation and decision-curve analysis.
3. Clarification of SOFA-2 Utility: The findings provide evidence that while SOFA-2 is a feasible measure for risk stratification in AC, it does not currently offer superior prognostic accuracy compared to the established SOFA score.

Significance and Claims

The paper concludes that while SOFA-2 successfully stratifies mortality risk in ICU patients with acute cholangitis, it does not outperform the traditional SOFA score in terms of discrimination, calibration, or clinical utility.

The authors state that these findings support SOFA-2 as a feasible research-compatible organ-dysfunction score for acute cholangitis but explicitly do not support replacing SOFA with SOFA-2 for routine clinical risk assessment in this specific population without further independent validation. The study emphasizes that the similarity in performance is clinically plausible, as the specific updates in SOFA-2 (primarily respiratory and cardiovascular components) may not significantly alter risk ranking in a condition heavily influenced by cholestasis, renal injury, and source control. The authors caution that risk scores should complement, not replace, clinical judgment regarding source control and procedural feasibility in acute cholangitis management.