A Host-Clinical Integrated Prediction Model for Early Sepsis Risk in Diabetic Patients with Acute Infections: Development and Internal Validation.

OBJECTIVE

Diabetic patients exhibit increased susceptibility to infections and a higher risk of progression to sepsis. This study aimed to develop a sepsis risk prediction model specifically for diabetic patients with acute infections.

METHODS

In this retrospective, single-center study, we enrolled 604 hospitalized diabetic patients with acute infections. A propensity score-matched cohort (103 sepsis cases vs 412 non-sepsis controls) was constructed. A predictive model integrating clinical manifestations and laboratory parameters was developed by first pre-selecting variables via a bootstrap-based penalized regression process, followed by Firth penalized logistic regression. The model underwent internal validation and subgroup analysis, with its performance compared against the National Early Warning Score 2 (NEWS2), Systemic Inflammatory Response Syndrome (SIRS), and Quick Sequential Organ Failure Assessment (qSOFA) scoring systems in terms of discrimination, calibration, clinical utility, and cost-effectiveness.

RESULTS

The final model incorporated five predictors: temperature dysregulation, nausea or vomiting, fatigue, albumin, and procalcitonin. It demonstrated excellent predictive performance, with a corrected AUC of 0.915 and a calibration slope of 1.022. Subgroup analyses confirmed consistent performance across all subgroups, and the model significantly outperformed NEWS2, SIRS, and qSOFA in predictive accuracy, clinical applicability, and cost-effectiveness.

CONCLUSION

The host-clinical integrated model developed in this study could provide a novel tool for early sepsis risk warning in diabetic patients with acute infections. This approach of integrating host characteristics with the pathophysiology of sepsis may offer new insights for sepsis risk early warning in specific populations. However, external validation in independent cohorts is needed before clinical implementation.