Integrating Social Determinants of Health with SOFA Scoring to Enhance Mortality Prediction in Septic Patients: A Multidimensional Prognostic Model

Abstract

Background: The Sequential Organ Failure Assessment (SOFA) score is an established tool for monitoring organ failure and defining sepsis. However, its predictive power for sepsis mortality may not account for the full spectrum of influential factors. Recent literature highlights the potential impact of socioeconomic and demographic factors on sepsis outcomes. Objective: This study assessed the prognostic value of SOFA scores relative to demographic and social health determinants in predicting sepsis mortality, and evaluated whether a combined model enhances predictive accuracy. Methods: We utilized the Medical Information Mart for Intensive Care (MIMIC)-IV database for retrospective data and the Penn State Health (PSH) cohort for prospective external validation. SOFA scores, social/demographic data, and the Charlson Comorbidity Index were used to train a Random Forest model using the MIMIC-IV dataset, and then to externally validate it using the PSH dataset. Findings: Of 32,970 sepsis patients in the MIMIC-IV dataset, 6,824 (20.7%) died within 30 days. The model incorporating demographic, socioeconomic, and comorbidity data with SOFA scores showed improved predictive accuracy over SOFA parameters alone. Day 2 SOFA components were highly predictive, with additional factors like age, weight, and comorbidity enhancing prognostic precision. External validation demonstrated consistency in the model's performance, with delta SOFA between days 1 and 3 emerging as a strong mortality predictor. Conclusion: Integrating patient-specific information with clinical measures significantly enhances the predictive accuracy for sepsis mortality. Our findings suggest the need for a multidimensional prognostic framework, considering both clinical and non-clinical patient information for a more accurate sepsis outcome prediction.