A new diagnostic model for the early differentiation of tuberculous and bacterial meningitis to guide timely diagnosis and treatment: Model development and internal validation

BACKGROUND: Tuberculous meningitis (TBM) is a life-threatening infection that requires early detection to improve outcomes. We aimed to develop and internally validate a diagnostic model to differentiate TBM from bacterial meningitis (BM) using routine clinical, laboratory, and radiologic findings. METHODS: This ambispective study included 377 patients aged > 15 years with suspected meningitis who were admitted to Saraburi Hospital, Thailand, between 2017 and 2024. Of these, 142 were classified as TBM (81 definite, 47 probable, 14 possible) and 235 as BM. Patients were classified as having TBM or BM according to standardized criteria. Multivariable fractional polynomial logistic regression with backward elimination was used to develop the model based on admission findings. Model performance was assessed using the area under the receiver operating characteristic curve (AuROC) and calibration analysis, with internal validation through 500 bootstrap resamples to adjust for optimism. Risk groups were defined using the Youden index and calibration plots. A web-based calculator was developed for bedside use. RESULTS: The final model included seven predictors: sex, duration of cough, duration of illness, cerebrospinal fluid (CSF) absolute neutrophil count, CSF glucose-to-protein ratio, chest radiograph findings, and evidence of extrapulmonary tuberculosis. The model exhibited excellent discriminatory performance (optimism-adjusted AuROC: 0.978), with calibration plots demonstrating strong agreement between predicted and observed probabilities. It also significantly outperformed existing tools-such as the Thwaites score, Lancet Consensus Scoring System, and Dendane et al.-in terms of net benefit and AuROC (p < 0.001). CONCLUSION: We developed an internally validated diagnostic model with superior accuracy for distinguishing TBM from BM using seven routine clinical variables. The model is publicly available at https://www.sbh.go.th/TBM/. External validation is still needed.