Development and bootstrap-based internal validation of a dynamic model for predicting acute kidney injury after day-3 reassessment in children with severe community-acquired pneumonia.

BACKGROUND

Children with severe community-acquired pneumonia (SCAP) are at risk of acute kidney injury (AKI), but creatinine-based recognition may be delayed, and existing pediatric AKI models are not SCAP-specific or trajectory-based. We developed a model for day-3 reassessment of subsequent AKI risk.

METHODS

This single-center retrospective prediction model development study included children aged 1 month to 18 years with SCAP, defined by at least 1 major or 2 minor Pediatric Infectious Diseases Society (PIDS)/Infectious Diseases Society of America (IDSA) severity criteria. Eligible patients had baseline and day-3 blood counts, C-reactive protein (CRP), albumin, and serial creatinine measurements. Children with chronic kidney disease or congenital urinary anomalies, hospital-acquired or ventilator-associated pneumonia, transfer after >48 hours of treatment, major immunosuppression or chronic liver disease, missing core variables, or AKI before the day-3 index time were excluded. Incident AKI after day 3 was defined by Kidney Disease: Improving Global Outcomes (KDIGO) serum creatinine criteria. Candidate predictors were prespecified based on clinical relevance, prior evidence, availability, and sample-size constraints, with least absolute shrinkage and selection operator (LASSO) used as an auxiliary stability assessment. Bootstrap-based internal validation used 1,000 resamples.

RESULTS

Among 454 children [median age, 22 months; 56.4% male; 21.4% with Pediatric Critical Illness Score (PCIS) ≤70], 94 (20.7%) developed incident AKI. The 9-parameter model included age, PCIS ≤70, mechanical ventilation, vasoactive support, vancomycin-piperacillin/tazobactam exposure, fluid overload >10%, baseline creatinine, ΔPLR, and CAR clearance rate (CARc). The apparent area under the curve (AUC) was 0.872 (95% confidence interval: 0.834-0.910), the optimism-corrected AUC was 0.857, the corrected calibration slope was 0.940, and the corrected Brier score was 0.095. A 10% threshold yielded 90.4% sensitivity, 55.0% specificity, and 95.7% negative predictive value; a 30% threshold yielded 70.2% sensitivity, 80.3% specificity, and 48.2% positive predictive value.

CONCLUSIONS

Dynamic inflammatory changes improved day-3 AKI risk prediction within the development cohort. The model and proposed thresholds require independent external validation before bedside use.