Bacterial versus viral meningitis: comparison of the old and the new clinical prediction models

Full text

Introduction

Accurate initial diagnosis is the cornerstone for therapeutic decision making of acute bacterial meningitis (ABM). A previously reported statistical model based on a combination of four parameters (total polymorphonuclear cell count in cerebrospinal fluid (CSF), CSF/blood glucose ratio, age and month of onset) appeared effective in differentiating acute viral meningitis (AVM) from acute bacterial meningitis in western countries. The objectives of this study were to validate this model on a independent sample of patients with acute meningitis seen in Okinawa, a tropical region of Japan, and to build a new model based on this sample.

Methods

Retrospective review was performed for medical records of all persons aged more than 15 years for the management of community-acquired acute meningitis treated at a our hospital between 1985 and 1998. The criterion standard for bacterial meningitis was a positive CSF or blood culture. For viral meningitis, it was a discharge diagnosis of viral meningitis with no other etiology evident. A new prediction model was developed by using stepwise logistic regression analysis.

Results

Forty-five cases of bacterial meningitis and 101 cases of viral meningitis were confirmed. The discriminatory power of the old model as measured by the area under the receiver operating characteristic curve (AUC) was 0.695 (95% Confidence Interval or CI 0.545-0.845). The best representative model selected four slightly different independent variables: age, mental confusion, neck stiffness, and total CSF polymorphonuclear cell count. The area under the ROC curve was 0.986 (95% CI 0.949-1.0), and significantly better than the old (P<0.05).

Conclusion

In differential diagnosis of acute bacterial and viral meningitis, the old prediction model may not be clinically useful when applied to a geographically distinct population in a different ethnicity. Older age, mental confusion, neck stiffness, and higher CSF polymorphonuclear cell count were identified as the most accurate logistic model predicting the likelihood of bacterial meningitis on our patients.