O. Dasilva et al., ACCURACY OF LEUKOCYTE INDEXES AND C-REACTIVE PROTEIN FOR DIAGNOSIS OFNEONATAL SEPSIS - A CRITICAL-REVIEW, The Pediatric infectious disease journal, 14(5), 1995, pp. 362-366
To evaluate the value of C-reactive protein and leukocyte indices in t
he workup of patients suspected of having infection in a neonatal inte
nsive care setting, a literature search was conducted in all languages
using MEDLINE (1966 to May, 1994), EMBASE (1988 to May, 1994), biblio
graphic lists of primary and review articles and personal files. Citat
ions identified as potentially relevant were reviewed by two independe
nt investigators; only studies meeting preset criteria for population,
diagnostic test and data presentation were included. Two observers in
dependently assessed studies using explicit methodologic criteria. All
data from the articles were extracted by one observer, whereas the se
cond reviewer checked these data for accuracy. Four of the selected st
udies dealt with leukocyte count and ratios. The chi square test for h
omogeneity of proportions revealed significant heterogeneity across st
udies (P = 0.014 for the ratios; P < 0.001 for white blood cell count)
, suggesting that test properties varied widely across studies. Fiftee
n of the selected studies evaluated C-reactive protein; of these six w
ere qualitative using a latex agglutination method, Among these studie
s the chi square test for homogeneity of proportions was highly signif
icant (P < 0.01), reflecting the great heterogeneity across studies, A
mong the nine studies that evaluated five different quantitative metho
ds heterogeneity was again present (P < 0.001). Because of the strikin
g heterogeneity among the studies evaluated, pooling to give a summary
point estimate of the sensitivity and specificity of the various stud
ies was not possible and the results are reported as ranges, Clinician
s faced with a neonate with suspected sepsis cannot rely on either C-r
eactive protein or leukocyte indices alone to make a decision, given t
hat the results vary significantly depending on the methods of measure
ment used and the target population.