ABSTRACT
This systematic review evaluates the evidence for accuracy of automated analyzers that estimate cerebrospinal fluid (CSF) white blood cell counts (WBC) compared to manual microscopy. Inclusion criteria of original research articles included human subjects, English language, and manual microscopy comparator. PUBMED, EMBASE and Cochrane Review databases were searched through 2019 and QUADAS-2 Tool was used for assessment of bias. Data were pooled and analyzed by comparison method, using random effects estimation. Among 652 titles, 554 abstracts screened, 104 full-text review, 111 comparisons from 41 studies were included. Pooled estimates of sensitivity and specificity (n = 7) were 95% (95%-CI 93%-97%) and 84% (95%-CI: 64%-96%), respectively. Pooled R2 estimates (n = 29) were 0.95 (95%-CI: 0.95-0.96); Pooled spearman rho correlation (n = 27) estimates were 0.95 (95% CI 0.95-0.96). Among those comparisons using Bland-Altman analysis (n = 11) pooled mean difference was estimated at 0.98 (95% CI-0.54-2.5). Among comparisons using Passing-Bablok regressions (n = 14) the pooled slope was estimated to be 1.05 (95% CI 1.03-1.07). Q tests of homogeneity were all significant with the exception of the Bland-Altman comparisons (I2 10%, p value 0.35). There is good overall accuracy for CSF WBC by automated hematologic analyzers. These findings are limited by the small sample sizes and inconsistent validation methodology in the reviewed studies.
