Integrating dynamic SOFA changes and age to predict 28-day mortality in ICU patients: a nomogram and machine learning...

Area of Science:

• Critical Care Medicine
• Prognostic Modeling
• Health Informatics

Background:

• The Sequential Organ Failure Assessment (SOFA) score is a standard prognostic tool for critically ill patients.
• The prognostic value of dynamic SOFA score changes (ΔSOFA) and their integration into prediction models requires further investigation.

Purpose of the Study:

• To evaluate the prognostic value of dynamic SOFA changes (ΔSOFA 3-1) combined with baseline SOFA scores (SOFA 1) and age.
• To develop and validate predictive models (nomogram and XGBoost) for patient outcomes using these parameters.

Main Methods:

• Retrospective analysis of 665 ICU patients, collecting initial and daily SOFA scores (days 1-3) and demographic data.
• Development of a nomogram integrating SOFA 1, ΔSOFA 3-1, and age, and an XGBoost model using the same predictors.
• Evaluation of model discriminative ability and calibration, with external validation using the MIMIC-IV database.

Main Results:

• Mortality increased with higher SOFA 1 and ΔSOFA 3-1 scores.
• The nomogram demonstrated high discriminative ability (C-index ~0.85), and the XGBoost model achieved an AUC of 0.833 internally and 0.863 in an independent test cohort.
• ΔSOFA 3-1 was identified as the most influential predictor across models and datasets.

Conclusions:

• Dynamic SOFA changes (ΔSOFA 3-1), particularly in patients with moderate baseline SOFA 1 scores (4-11), significantly enhance prognostic accuracy when combined with age.
• The nomogram offers a practical bedside tool for risk stratification, while the XGBoost model highlights machine learning's potential.
• External validation indicated a need for multicenter studies to improve model generalizability.