Construction and validation of a sepsis prediction model using inflammatory and hemodynamic markers in the emergency department.

OBJECTIVE

To develop an early diagnosis prediction model for sepsis in the emergency department by integrating inflammatory, hemodynamic, and nursing assessment indicators.

METHODS

A retrospective cohort of 320 patients with suspected infection admitted to our hospital was enrolled. Participants were randomly allocated into a training set and a validation set at a 7:3 ratio. Key variables were selected using least absolute shrinkage and selection operator (LASSO) regression. Subsequently, logistic regression, random forest, and support vector machine models were constructed. Model performance and interpretability were evaluated using the receiver operating characteristic curve, calibration curve, decision curve analysis, and SHapley Additive exPlanations (SHAP) values.

RESULTS

LASSO regression identified three core variables: procalcitonin (PCT), neutrophil-to-lymphocyte ratio (NLR), and Modified Early Warning Score (MEWS). Multivariate analysis revealed that all three were independent risk factors (p < 0.05). The random forest model demonstrated an area under the curve (AUC) of 0.737 in the training set and 0.714 in the internal random-split validation set. At the optimal cutoff, the validation performance yielded a sensitivity of 72.7%, specificity of 76.2%, PPV of 61.5%, and NPV of 84.2%. It showed good calibration and provided clinical net benefit. SHAP analysis indicated that MEWS contributed the most to the model's predictions, followed by NLR and PCT.

CONCLUSION

The integrated model developed in this study exhibits moderate to satisfactory predictive performance and clinical utility. It may act as a candidate bedside risk-stratification tool pending further external multicenter validation for emergency nursing sepsis screening.