Thirty-three studies (with a total of 1,885 participants) were included in the review. The mean QUADAS score was 10.4 (range 8 to 14) and many aspects of study quality were poorly reported. Only nine studies described the selection of participants adequately and collected cerebrospinal fluid samples consecutively.
The hierarchical summary receiver operating characteristic curve gave a pooled estimate of sensitivity of 93% (95% CI 89% to 96%) and a pooled estimate of specificity of 96% (95% CI 93% to 98%). Likelihood ratio positive was 22.9 (95% CI 12.6 to 41.9) and likelihood ratio negative was 0.07 (95% CI 0.05 to 0.12). The diagnostic odds ratio was 313 (95% CI 141 to 698). Pooling using the random-effects model gave similar results.
Subgroup analyses indicated that sensitivity was lower in participants who were pre-treated with antibiotics (49%; 95% CI 23% to 75%; nine studies) compared with those not receiving pre-treatment (98%; 95% CI 96% to 100%; 17 studies).
Meta-analysis indicated that the optimal cut-off value to distinguish bacterial meningitis from aseptic meningitis was around 35 mg/dL.
Begg's test suggested possible publication bias but the authors stated that this test does not provide a valid method of assessing publication bias in reviews of test accuracy studies.