Respiratory support requirement among children hospitalized with pneumonia is a key marker of disease severity and resource needs, yet scalable risk stratification tools for routine hospital settings in Southern Vietnam remain limited. Background: This study aimed to develop and evaluate clinical and laboratory-based multivariable models to predict respiratory support requirement in children under five hospitalized with pneumonia, using a routine care dataset. Methods: We conducted a retrospective cohort study conducted at a tertiary pediatric hospital in Southern Vietnam (July 2024–November 2025), children aged 2–59 months hospitalized with pneumonia were included after predefined exclusions. The outcome was the maximum (worst) level of respiratory support required during hospitalization (oxygen therapy, CPAP, or invasive mechanical ventilation), analyzed as a binary endpoint (any support vs. none) for model development. Candidate predictors included bedside clinical variables (age < 12 months, malnutrition, recurrent pneumonia, cyanosis, tachypnea, chest indrawing) and complete blood count-derived inflammatory indices. Univariable logistic regression was used for crude associations. Two multivariable logistic regression models were built: Model 1 (clinical-only) and Model 2 (clinical + neutrophil-to-lymphocyte ratio [NLR]; primary). Discrimination was assessed using area under the ROC curve (AUC), and calibration was evaluated using the Hosmer–Lemeshow test and observed-to-expected (O:E) ratio. Results: A total of 1797 children were included; 154 (8.6%) required respiratory support. In the primary model, independent predictors were age < 12 months (aOR 2.57, 95% CI 1.69–3.92), malnutrition (aOR 4.33, 2.56–7.33), recurrent pneumonia (aOR 1.82, 1.18–2.81), cyanosis (aOR 24.02, 7.41–77.87), chest indrawing (aOR 4.19, 2.73–6.43), and higher NLR (per 1 unit: aOR 1.49, 1.38–1.60), while tachypnea was not independently associated after adjustment. Discrimination improved from Model 1 (AUC 0.754) to Model 2 (AUC 0.840; 95% CI 0.806–0.874). At the optimal probability cut-off (0.122), Model 2 achieved sensitivity 66.2%, specificity 86.2%, PPV 31.1%, NPV 96.5%, and accuracy 84.5%. Calibration was acceptable (Hosmer–Lemeshow p = 0.662; O:E = 1.00). Conclusions: A simple clinical model strengthened by NLR provided good discrimination and calibration for predicting respiratory support requirement among children under-five hospitalized with pneumonia in Southern Vietnam. This approach may support early triage, prioritization of monitoring intensity, and escalation readiness in resource-constrained settings, although external validation is warranted.

Background: Pediatric sepsis remains a critical global health concern due to its high morbidity and mortality. Several clinical scoring systems have been employed to assess disease severity and predict outcomes in patients with sepsis. The Phoenix Sepsis Score is a recently proposed pediatric-specific tool for risk stratification, but evidence on its performance in Southeast Asia remains limited. Objective: To evaluate clinical outcomes and assess the prognostic performance of the Phoenix Sepsis Score in predicting mortality among pediatric patients with sepsis. Materials and Methods: A cross-sectional analytical study was conducted on 94 pediatric patients (aged 2 months to 16 years) with sepsis admitted to a tertiary referral hospital in the Mekong Delta, Vietnam. Clinical, laboratory, and scoring variables were recorded within the first 6 hours after the diagnosis of sepsis. Prognostic performance was assessed using receiver operating characteristic curve analysis. Results: Among 94 patients, 67% were under 5 years of age. Lower Glasgow Coma Scale scores (10.96 ± 3.61 vs. 14.19 ± 3.41; p<0.001) and higher rates of septic shock (61.5% vs. 20.6%; p<0.001) were observed among non-survivors compared with survivors. The use of vasoactive agents (96.2% vs. 26.5%; p<0.001), mechanical ventilation (92.3% vs. 38.2%; p<0.001), and ≥2 antibiotic modifications (80.8% vs. 55.9%; p=0.026) were also associated with mortality. The Phoenix Sepsis Score demonstrated high discriminative ability (area under the curve [AUC], 0.947; 95% confidence interval [CI], 0.90–0.99), similar to that of the pediatric Sequential Organ Failure Assessment (pSOFA) score (AUC, 0.943; 95% CI, 0.898–0.988). At optimal cutoffs of 3.5 and 8.5 points, respectively, the Phoenix Sepsis Score showed higher sensitivity (92.3% vs. 76.9%), whereas pSOFA showed higher specificity (91.2% vs. 83.8%). Conclusion: The Phoenix Sepsis Score demonstrated prognostic performance similar to that of pSOFA in this cohort, with higher sensitivity but lower specificity. These findings suggest that the Phoenix Sepsis Score may be useful as a supportive tool for early risk stratification in pediatric sepsis, particularly in resource-limited settings. Further multicenter studies are required to validate these findings.