Accuracy of cerebrospinal leucocyte count, protein and culture for the diagnosis of acute bacterial meningitis: a comparative study using Bayesian latent class analysis

Abstract objective To examine the utility of laboratory methods other than bacterial culture in diagnosingacute bacterial meningitis (ABM).methods Bayesian latent class analysis was used to estimate diagnostic precision of cerebrospinalfluid (CSF) culture, leucocyte counts and protein concentrations for ABM in Melanesian children.results With a cut-off of ≥20 leucocytes/mm 3 , the area under the receiver operating characteristiccurve (AUC ROC) was >97.5% for leucocyte counts. A lower (93%) AUC ROC was observed forCSF protein concentrations ≥1 g/l. CSF culture had poor sensitivity and high specificity.conclusion Leucocyte counts provide sufficient diagnostic precision to aid clinical decision-makingin ABM.keywords cerebrospinal fluid, acute bacterial meningitis, Bayesian latent class analysis, Papua NewGuineaIntroductionAn accurate diagnosis is an essential component of theoptimal management of acute bacterial meningitis (ABM)(Peltola 2001; Brouwer et al. 2012). Globally, many ofthe 180 000 deaths and the neurological complicationsresulting from ABM occur in children in epidemiologicalsettings with too few or no diagnostic facilities (van deBeek 2012). Even when there is a functional laboratory,bacterial culture is usually unavailable and diagnostictests for ABM may be limited to the detection of leuco-cytes by microscopy of cerebrospinal fluid (CSF), with orwithout point-of-care semi-quantitative estimates of CSFfluid protein and glucose concentrations (Berkley et al.1999; Peltola 2001; Fuller et al. 2003).Studies of the diagnostic precision of CSF leucocytedensities and protein concentrations have generallyassumed that CSF culture is the appropriate gold stan-dard for determination of sensitivity and specificity.However, antibiotic administration before presentation/lumbar puncture (LP) may lead to negative CSF culturesin patients who have clinical and laboratory features ofABM. Small CSF samples, suboptimal specimen handling,and problems with culture media, incubation and bacte-rial identification can further limit the sensitivity of CSFculture. The use of Bayesian latent class analysis (LCA) isone approach that addresses the limitations inherent inassuming CSF culture as the gold standard for diagnosisof ABM. This technique has been successfully applied toother infectious diseases in developing countries (Ocholaet al. 2006; Limmathurotsakul et al. 2010; Manninget al. 2012) and assumes a latent (but unknown) class(disease or no disease) for each individual.As part of a prospective study conducted in a malariaendemic area of Papua New Guinea (PNG) where ABMcaused by Streptococcus pneumoniae or Haemophilusinfluenzae Type B is common (Laman et al. 2010), weperformed microbiologic cultures, CSF leucocyte countsand semi-quantitative protein concentrations to diagnoseABM in hospitalised children. Rather than assuming CSFculture as the gold standard, we performed BayesianLCA to determine the performance of each diagnosticmodality. The primary aim of the study was to determinethe diagnostic precision of CSF leucocyte count and semi-quantitative protein concentrations as well as the optimal