Malik Arinder, Hui Charles P S, Pennie Ross A, Kirpalani Haresh
Division of Neonatology, Department of Pediatrics, McMaster University Medical School, 1200 Main Street West, Hamilton, Ontario L8N 1C5, Canada.
Arch Pediatr Adolesc Med. 2003 Jun;157(6):511-6. doi: 10.1001/archpedi.157.6.511.
To systematically review the accuracy of modern laboratory tests for the diagnosis of serious bacterial infection in newborns.
The MEDLINE, EMBASE, and Cochrane Library databases were searched using the keywords newborn, infection, sepsis, and diagnosis. We included studies published from 1995 through 2001 that included infants younger than 90 days with proven bacterial growth in a sample from a sterile site. Whenever possible, relevant data were extracted to calculate likelihood ratios (LRs) for whether each test can diagnose a serious bacterial infection. Two independent reviewers selected and reviewed the articles (interobserver agreement, kappa = 0.80). All disagreements were resolved by consensus.
Of the 137 citations we retrieved, 37 articles met the inclusion criteria; 17 studies, evaluating 11 different tests, met the highest methodological criteria. The most commonly evaluated test was interleukin 6 (IL-6) level (n = 7 studies). The remaining tests were each evaluated in no more than 3 studies. Positive LRs ranged from 1.5 to infinity. Six individual tests examined in 8 studies had LRs of more than 10 (range, 12.5- infinity ). Combined tests also had a wide range of LRs (3.4-9.9). All studies were performed in single medical centers and had small sample sizes, making recommendations according to gestational age criteria difficult.
We found few methodologically rigorous studies of the accuracy of laboratory tests for the diagnosis of bacterial infection in newborns; in a significant proportion of studies, the accuracy of the tests could not be independently determined because of a lack of adequate data. There was marked heterogeneity in sample selection and cutoff levels for diagnosis of neonatal sepsis. A few tests showed promising accuracy, but there are insufficient data to support their confident use as clinical tools.
