Abstract
BACKGROUND: Sepsis is a life-threatening complication in patients with pyogenic liver abscess (PLA). Early identification of at-risk individuals is critical for improving outcomes but remains challenging. Our objective was to evaluate the prognostic value of integrating the National Early Warning Score 2 (NEWS2) with the procalcitonin-to-albumin ratio (PAR) for predicting sepsis in this population. METHODS: This retrospective cohort study enrolled 341 patients with PLA admitted to a tertiary hospital between 2011 and 2021. Patients were stratified based on Sepsis-3 criteria. Multivariate logistic regression identified independent predictors. Predictive performance was assessed via the area under the receiver operating characteristic curve (AUC), net reclassification improvement (NRI), integrated discrimination improvement (IDI), and decision curve analysis (DCA). RESULTS: Among the 341 patients, 50 patients (14.7%) developed sepsis. These patients had significantly higher NEWS2 scores [4.00 (IQR 2.00–6.25) vs. 2.00 (IQR 1.00–3.00), p < 0.001], SOFA scores [2 (IQR 1.00–4.25) vs. 1 (IQR 0.00–2.00), p < 0.001], and PAR levels [0.47 (IQR 0.10–1.66) vs. 0.06 (IQR 0.01–0.40), p < 0.001]. The PAR was positively correlated with both the NEWS2 score (r = 0.309, p < 0.001) and the SOFA score (r = 0.254, p < 0.001). Multivariate analysis confirmed both the NEWS2 score (OR 1.369, 95% CI 1.143–1.639, p < 0.001) and the PAR (OR 1.004, 95% CI 1.000–1.008, p = 0.048) were independent predictors. The combined model (NEWS2 + PAR) achieved superior discrimination (AUC 0.798, 95% CI 0.726–0.871) compared with NEWS2 (AUC 0.724, 95% CI 0.635–0.813) or PAR (AUC 0.727, 95% CI 0.648–0.805) alone, with significant reclassification improvement (NRI = 0.324, p = 0.012; IDI = 0.045, p = 0.008), and provided the highest net benefit across decision thresholds. CONCLUSIONS: The integration of NEWS2 and PAR at admission may improve early sepsis prediction in PLA patients compared with either marker alone. This readily applicable integrative model could facilitate rapid risk stratification and may support earlier clinical decision‑making, potentially improving patient outcomes. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1186/s12879-026-12984-6.