Systematic review and validation of diagnostic prediction models in patients suspected of meningitis

Objectives: Diagnostic prediction models have been developed to assess the likelihood of bacterial meningitis (BM) in patients presented with suspected central nervous system (CNS) infection. External validation in patients suspected of meningitis is essential to determine the diagnostic accuracy of these models. Methods: We prospectively included patients who underwent a lumbar puncture for suspected CNS infection. After a systematic review of the literature, we applied identified models for BM to our cohort. We calculated sensitivity, specificity, predictive values, area under the curve (AUC) and, if possible, we evaluated the calibration of the models. Results: From 2012-2015 we included 363 episodes. In 89 (24%) episodes, the patient received a final diagnosis of a CNS infection, of whom 27 had BM. Seventeen prediction models for BM were identified. Sensitivity of these models ranged from 37% to 100%. Specificity of these models ranged from 44% to 99%. The cerebrospinal fluid model of Oostenbrink reached the highest AUC of 0.95 (95% CI 0.91–0.997). Calibration showed over- or underestimation in all models. Conclusion: None of the existing models performed well enough to recommend as routine use in individual patient management. Future research should focus on differences between diagnostic accuracy of the prediction models and physician's therapeutic decisions