G192 C-reactive protein for diagnosing late-onset infection in newborn infants: cochrane review of test accuracy

Abstract

Aims To determine the accuracy of elevated serum C-reactive protein (CRP) for diagnosing late-onset neonatal infection. Methods Cochrane systematic review of diagnostic test accuracy. We searched MEDLINE, Embase, and Science Citation Index to September 2017 for cohort and cross sectional studies evaluating the diagnostic accuracy of serum CRP for detecting late-onset infection in newborns: Index test: Serum CRP level (threshold defined by individual studies). Reference standards Invasive infection diagnosed>72 hour after birth, confirmed by culture from a normally sterile site or findings on autopsy examination consistent with invasive microbial infection. We screened titles and abstracts and evaluated the full text of possibly eligible articles. We extracted study characteristics and used the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool to assess quality. One reviewer extracted data for calculation of diagnostic accuracy parameters. These were checked independently by a second reviewer with referral to a third reviewer to resolve discrepancies. We constructed ‘two-by- two’ tables based on data from the reference standard and index test and created forest plots for sensitivity and specificity. We conducted a bivariate random effects meta-analysis of sensitivity and specificity data and used these estimates to construct a summary receiver operating characteristic curve. We estimated post-test probabilities of late-onset neonatal sepsis based on a range of pre-test probabilities. Results We included 20 studies (total number of infants 1,615) with sample sizes ranging between 11 and 184. Most studies were conducted in high income countries, investigating both term and preterm babies. Overall, the methodological quality of the studies was good and the risk of bias low. Data synthesis Pooled sensitivity: 0.58 (95% CI 0.45 to 0.69); Pooled specificity: 0.79 (95% CI 0.69 to 0.86). There was relatively high heterogeneity as reflected in the forest plots and 95% prediction region. Positive likelihood ratio: 2.73 (95% CI 1.95 to 3.84); Negative likelihood ratio 0.54 (95% CI 0.42 to 0.69). Conclusion Meta-analysis shows that diagnostic accuracy of serum CRP level is modest. Serum CRP level in this context is not sufficiently accurate to reliably confirm or exclude a diagnosis of infection.