Limited discriminatory value of C-reactive protein for viral coinfection and respiratory severity in children evaluated in the emergency department for acute respiratory tract illness

Background C-reactive protein is often measured in children who present to an acute care facility with acute respiratory tract illness (ARTI), and multiplex polymerase chain reaction (PCR) testing on respiratory samples frequently identifies more than one pathogen per encounter in children with ARTI. Whether CRP differentiates single-virus from multiple virus infections or improves prediction of clinical course of ARTI is uncertain. Objective To assess whether age-stratified CRP differs by viral coinfection status, and whether CRP adds discriminatory value beyond clinical variables for predicting advanced respiratory support, after accounting for selection bias, probable bacterial coinfection, repeated encounters and overadjustment. Methods This was a retrospective encounter-level cohort study of 17,089 encounters with ARTI in which multiplex respiratory polymerase chain reaction (PCR) testing was performed in the emergency department (ED) between Jan 1, 2019, and Dec 31, 2025. Encounters were not restricted to pathogen-positive results on the PCR. The analytic cohort comprised encounters with a measured CRP ( n = 4,946) drawn from 17,089 total ED encounters. We compared CRP-tested vs. untested encounters using standardized mean differences. Single-virus and multi-virus encounters were compared overall, by virus, by age stratum (< 6months, 6–23 months, 2–5 years, 6–12 years, and 13–17 years.) and by hospital disposition. Sensitivity analyses (a) restricted to encounters without probable bacterial coinfection (no sepsis, pneumonia, or antibiotics use), (b) restricted to one encounter per patient, and (c) re-estimated logistic models without potential mediators. Multivariable logistic regression for advanced respiratory support used a parsimonious specification (age, sex, prematurity, initial SpO2, CRP, virus features); a model including severity-of-illness was reported only as a sensitivity exhibit because severity is partly defined by outcome. Discrimination was summarized by AUC with bootstraps 95% CIs, and CRP thresholds were evaluated for sensitivity, specificity, and positive likelihood ratio (LR). The Benjamini-Hochberg FDR was used to control for multiple testing. Results Compared with untested encounters, tested patients were sicker (pediatric ward admission 74.0% vs. 36.3% [SMD 0.82]; pneumonia 23.4% vs. 9.6% [SMD 0.38]; antibiotic use 59.5% vs. 20.9% [SMD 0.86%]; PICU admission 45.1% vs. 28.5% [SMD 0.35]. RV/EV positivity was lower among tested encounters (30.7% vs. 41.4%, SMD – 0.23]. Within the CRP cohort, 3,037 (61.4%) met criteria for probably bacterial coinfection. After excluding probably bacterial coinfection, only RSV retained as significant single vs. coinfection CRP difference (3.0 vs. 5.2 mg/L, q = 0.005). Significant age x coinfection interactions on log-CRP were detected for RSV, common coronaviruses, adenovirus, and SARS-Cov-2 (all q , 0.05). The coinfection-CRP signal was concentrated in ED-only patients ( p = 0.008) and was absent among pediatric ward ( p = 0.30) or PICU admissions ( p = 0.58). CRP alone did not show discrimination for advanced respiratory support (AU 0.483); a clinical model achieved AUC 0.75, with no improvement after CRP (AUC 0.75). The CRP per-10-mg/L coefficient had aOR 1.013 (1.001–1.026, p = 0.036) in the parsimonious model and aOR 1.000 (0.983–1.018, p = 0.96) when restricted to one-encounter-per-patient. Conclusions CRP testing is preferentially measured in patients who already appear sicker. After accounting for probably bacterial infection, CRP does not reliably distinguish single from multi-virus ARTI except for RSV in younger children, and it does not provide clinically useful incremental discrimination beyond simple clinical variables for advanced respiratory support.