Objectives: To determine which individual items in published croup scores or combination of items predict return Emergency department visit or hospital admission (bad croup).

Design/Methods: Prospective cohort study of children >3 mos to <5 years with 24 hrs of respiratory symptoms presenting to urban pediatric emergency (E) department in Halifax, NS, Canada, from 10/95-11/96. Croup scoring by Triage nurse (TN), E nurse (EN) and E doctor (ED) using 9 items[air entry (AE), stridor (S), retraction (R), cough (C), color (CL), level of consciousness (LOC), respiratory rate (RR), heart rate (HR)].

Results: 314 children were seen with croup; 201 met entry criteria. Interobserver agreement (IOA) between TN and EN using weighted Kappa was 0.4-0.8 for S, R, AE, RR; between 0.1-0.4 for HR, C. EN-ED IOA was <0.4 for all items except S and R (0.4-0.8). TN scores were used in the analysis. In univariate analysis of the ability of individual items to predict bad croup, only R was statistically significant (SS); odds ratios (OR) were: LOC (5.4), R(1.8), S (1.6), AE (1.4), HR (1.3), RR (0.81), C (0.64). In univariate logistic regression to predict bad croup the following croup indices, created by adding or subtracting item scores, were SS: LOC + R, OR 1.9, 95% confidence interval (CI) (1.1, 3.3) receiver operating curve area (ROC) 0.63, correct classification (CC) 71%; AE + LOC + R + S, OR 1.31 CI (1.0, 1.7), ROC 0.65, CC 67%; S + R + AE + C +LOC, OR 1.3, CI (1.0, 1.7) CI, ROC 0.65, CC 67%; AE + LOC+R + S - C, OR 1.3, CI (1.0, 1.7), ROC.66, CC 62%.

Conclusions: Some items in croup scores do not provide clinically useful information in this ambulatory population (CL, RR, HR). Cough is inversely correlated with croup severity. One previously published score and 3 new scores are useful in predicting return ED visit or admission.