Accuracy of clinical presentation for differentiating bacterial from viral meningitis in adults: a multivariate approach

Abstract

Objective: To determine whether bacterial (BM) and viral (VM) meningitis can be differentiated based on initial clinical presentation.

Design and setting: Retrospective cohort study in a medical emergency department and intensive care unit in a university hospital.

Patients: 144 adults, including 90 with confirmed BM and 54 unpretreated VM.

Measurements and results: Symptoms, examination findings, paraclinical data, and clinical outcome were assessed. Severity was defined by the presence at referral of one of the following criteria: altered consciousness, seizures, focal neurological findings, and shock. After univariate analyses we performed stepwise logistic regression to determine predictors for BM available at referral (except for CSF Gram stain) and logistic regression using previously validated CSF cutoffs. Univariate methods identified the presence of one sign of severity as the most important predictor for BM (sensitivity 0.989, specificity 0.981, positive predictive value 0.989, negative predictive value 0.981, odds ratio 4,770) and showed that CSF results differ in BM and in VM (except for CSF glucose). Logistic regression analysis revealed severity and CSF absolute neutrophil count as the two predictors of BM (R2=0.876). Logistic analysis showed that BM was related to severity (beta=6.46+/-1.27) and a CSF absolute neutrophil count above 1,000/mm3 whereas CSF glucose below 2 mmol/l and CSF protein higher than 2 g/l were not predictive.

Conclusions: The presence of at least one sign of severity at referral and a CSF absolute neutrophil count above 1,000/mm3 mm are predictive of BM.