Receiver operating characteristic curves for comparison of serial neutrophil band forms and C reactive protein in neonates at risk of infection

• Published in: Archives of disease in childhood Vol. 67; no. 7 Spec No; pp. 808 - 812
• Main Authors: Russell, G A, Smyth, A, Cooke, R W
• Format: Journal Article
• Language: English
• Published: London BMJ Publishing Group Ltd and Royal College of Paediatrics and Child Health 01.07.1992; BMJ
• Subjects: Bacteria - isolation & purification; Bacterial diseases; Bacterial Infections - blood; Bacterial Infections - epidemiology; Bacterial Infections - microbiology; Biological and medical sciences; C-Reactive Protein - analysis; Chi-Square Distribution; False Positive Reactions; Human bacterial diseases; Humans; Infant, Newborn; Infectious diseases; Leukocyte Count - methods; Medical sciences; Nephelometry and Turbidimetry; Neutrophils - cytology; Risk Factors; ROC Curve; Performance evaluation; Exploration; In vitro; Infection; Nephelometry; Newborn diseases; Morphology; Bacteriosis; Receiver operating characteristic curves; Early; Diagnosis; Neutrophil; C reactive protein; Blood cell count; Maturity
• ISSN: 0003-9888; 1468-2044
• DOI: 10.1136/adc.67.7_Spec_No.808

The performance of indirect indices of infection in the newborn vary because of differences in techniques, including diagnostic cut off levels. We have compared serial neutrophil band cell counts with C reactive protein measured by rate nephelometry. The 'gold standard' was a positive culture and the performance of the tests was compared by the technique of receiver operating characteristics (ROC) as well as sensitivity and specificity. A total of 172 septic screens were performed in 56 patients. The operational diagnostic cut off values were: C reactive protein greater than 8 mg/l, immature:total neutrophil ratio (I:T ratio) greater than 0.2, and band count greater than 5%. Compared with the sensitivity of C reactive protein (71%), I:T ratio (34%) was significantly different but band count (69%) was not. The specificity of C reactive protein (72%) was better than band count (39%) but no better than I:T ratio (73%). ROC curves were constructed for all possible diagnostic cut off values of the tests and superior performance was demonstrated for C reactive protein compared with band count and I:T ratio. We conclude that C reactive protein is a useful early indicator of infection in neonates and that ROC curves permit comprehensive and graphic comparison between tests and the calculation of optimal diagnostic cut off values.