Table 1_Predicting sepsis prognosis using deep learning with routine biomarkers.xlsx

ObjectiveSepsis, a critical illness with extremely high mortality, has its patient prognosis early predicted as crucial to improving therapeutic outcomes. Early and accurate risk stratification is particularly vital for emergency sepsis patients, who often present with rapid disease progression and complex pathophysiological changes. This study aimed to construct an artificial intelligence (AI) model integrating red cell distribution width (RDW) and estimated plasma volume status (ePVS) to predict the 28-day mortality of patients with emergency sepsis.
MethodsBased on the data of 73 sepsis patients in the emergency department (42 in the survival group and 31 in the death group), a multilayer perceptron (MLP) deep learning neural network model was constructed. Internal and external validations were conducted using independent MIMIC-III and MIMIC-IV datasets, and the model performance was evaluated in terms of accuracy, sensitivity, specificity, and the area under the receiver operating characteristic (ROC) curve (AUC).
ResultsIn internal validation, the deep learning model achieved an accuracy of 87.0%; during external validation, its accuracy improved to 92.0%, with a sensitivity of 84.0% and specificity of 83.0%. Patients in the deceased group exhibited significantly elevated RDW and ePVS (p < 0.05). The deep learning model combining RDW-CV and ePVS demonstrated the best. These findings suggest that elevated RDW and ePVS may serve as potential warning signs for adverse clinical outcomes in sepsis. Predictive performance, achieving an AUC of 0.812, which was significantly superior to any single indicator (DeLong test, p < 0.01).
ConclusionThis study successfully developed a deep learning-based prognostic prediction model, confirming that a deep learning framework integrating RDW and ePVS can significantly enhance the predictive accuracy for the prognosis of emergency department sepsis patients. This novel predictive strategy compensates for the limitations of traditional single biomarkers and clinical scoring systems. This approach aids in the early identification of high-risk patients and provides an intelligent support tool for clinical decision-making.