A time-incorporated SOFA score-based machine learning model for predicting mortality in critically ill patients: A...

Area of Science:

• Critical care medicine
• Machine learning in healthcare
• Predictive analytics for patient outcomes

Background:

• Current organ dysfunction (OD) assessment scores in ICUs lack a time-dimension component.
• Existing scores quantify OD severity and number but not the duration of organ injury.
• Accurate prediction of mortality in critically ill patients is crucial for timely intervention.

Purpose of the Study:

• To develop and validate a machine learning model for predicting in-ICU mortality.
• To enhance the Sequential Organ Failure Assessment (SOFA) score by incorporating a time dimension (T-SOFA).
• To evaluate the predictive performance of the T-SOFA model against the original SOFA score.

Main Methods:

• Utilized data from eICU Collaborative Research Database and MIMIC-III for model development.
• External validation performed using MIMIC-IV and Nanjing Jinling Hospital Surgical ICU data.
• Developed a modified SOFA model (T-SOFA) using extreme gradient boosting (XGBoost) and other algorithms, incorporating time-dimensional features from consecutive SOFA scores.
• Assessed predictive performance using area under the receiver operating characteristic curves (AUROC) and calibration plots.

Main Results:

• Included 82,132 patients; 9.12% experienced in-ICU mortality.
• The T-SOFA model (M3, incorporating time-dimension features and age, using XGBoost) significantly outperformed the original SOFA score (AUROC 0.800 vs. 0.693, p < 0.01).
• Consistent high predictive performance observed in external validation sets A and B (AUROC 0.803 and 0.830, respectively).

Conclusions:

• The time-incorporated T-SOFA model significantly improves mortality prediction compared to the original SOFA score.
• The T-SOFA model demonstrates potential for identifying high-risk critically ill patients in clinical settings.
• Incorporating the time dimension enhances the utility of OD assessment scores for predicting patient outcomes.