Triage Performance and Physiological Parameters Predicting Serious Illness in Children Presenting With Fever to the Pediatric Emergency Department: A Multicenter Prospective Cohort Study

Design: Prospective, multicenter, observational cohort study. Setting: 10 pediatric emergency departments in Turkey; Erciyes University is the coordinating center. Population: consecutive children aged 0-18 years presenting with fever (>=38.0 C at triage or history of fever within the preceding 24 hours). Trauma presentations, elective visits, records with missing triage data, and re-attendances within 72 hours for the same episode are excluded (only the first visit is included). Physiological parameters are recorded on a standardized triage form and are NOT used to change the triage decision. Heart rate is modeled as a continuous variable with age-based z-scores (Fleming reference ranges), with body temperature always included as a covariate to distinguish fever-related from pathological tachycardia. Sample size: assuming a serious-outcome (composite) event rate of ~3%, approximately 3000 children (~90 events) are targeted (~300 per center over 3 months), following Riley et al. (BMJ 2020) criteria (target R2>=0.15, calibration slope>=0.90) in addition to the EPV>=10 rule. A pilot evaluation at the end of the first month recalculates the true event rate. Analysis: Triage performance is summarized with sensitivity, specificity, PPV, NPV, LR+ and LR- with 95% CIs. A pre-specified 3-step mixed-effects logistic regression (centers as random effect) compares Model 1 (triage category only), Model 2 (+ physiological parameters and temperature), and Model 3 (+ laboratory variables in the tested subgroup). Discrimination (ROC/AUC, DeLong test, NRI, IDI) and calibration (calibration curve and slope, O/E ratio) are reported. Missing data are handled by multiple imputation (m=20).