A NEW study demonstrates the potential of machine learning to predict the occurrence and mortality of nonpulmonary sepsis-associated acute respiratory distress syndrome (NPS-ARDS), a severe complication linked to increased intensive care unit (ICU) stays and mortality rates. Using advanced algorithms, the research offers a tool to assist clinicians in identifying high-risk patients early.
Researchers analysed data from 11,409 sepsis patients in the MIMIC-IV database, focusing on those without prior pulmonary conditions and with nonpulmonary infection sites, such as bloodstream infections. Of these, 66.9% developed NPS-ARDS, which was associated with significantly longer ICU stays (6.2 vs. 4.4 days) and higher 28-day mortality rates (19.5% vs. 14.9%).
The study employed multiple machine learning techniques, including K-nearest neighbour, support vector machine, and extreme gradient boosting (XGBoost), to construct predictive models. The XGBoost model outperformed others, achieving an accuracy of 77.5% for predicting NPS-ARDS and 71.8% for mortality in internal validation. External validation confirmed the model’s robustness, with accuracies of 78.0% for NPS-ARDS and 81.4% for mortality.
These findings highlight the promise of machine learning in improving clinical decision-making for sepsis patients. By accurately identifying those at risk of NPS-ARDS and mortality, healthcare providers can prioritise interventions and optimise outcomes.
Reference
Lin J et al. Machine learning-based model for predicting the occurrence and mortality of nonpulmonary sepsis-associated ARDS. Sci Rep. 2024;DOI:10.1038/s41598-024-79899-7.
