Background: The identification of serious bacterial infection (SBI) in children with fever without source remains a challenge. A risk score called Lab-score, based on C-reactive protein, procalcitonin and urinary dipstick results was derived to predict SBI. However, all biomarkers were initially dichotomized, leading to weak statistical reliability and lack of transportability across diverse settings. We aimed to refine and validate this risk-score algorithm.
Methods: The Lab-score was refined using a secondary analysis of a multicenter cohort study of children with fever without source via multilevel regression modeling. The external validation was conducted on data from a Canadian cohort study.
Results: Eight hundred seventy-seven children (24% SBI) were included for the derivation study, and 347 (16% SBI) for validation. Only C-reactive protein, procalcitonin, age and urinary dipstick remained independently associated with SBI. The model achieved an area under the receiver operating characteristic (ROC) curve of 0.94 (95% confidence interval [CI]: 0.93-0.96), which was significantly higher than any other isolated biomarker (P < 0.0001), and the original Lab-score (P < 0.0001). According to a decision curve analysis, the model yielded a better strategy than those based on independently considered biomarkers, or on the original Lab-score. The threshold analysis led to a cutoff that yielded 96% (95% CI: 92-98) sensitivity and 73% (95% CI: 70-77) specificity. The external validation found similar predictive abilities: 0.96 area under the ROC curve (95% CI: 0.93-0.99), 95% sensitivity (95% CI: 85-99) and 87% specificity (95% CI: 83-91).
Conclusion: The refined Lab-score demonstrated higher prediction ability for SBI than the original Lab-score, with promising wider applicability across settings. These results require validation in additional populations.