# Development and validation of a diagnostic model for tuberculous meningitis based on laboratory data

## TL;DR

This study created a low-cost diagnostic model for tuberculous meningitis using common lab tests, offering reliable results in resource-limited areas.

## Contribution

A novel diagnostic scoring system for TBM using four CSF parameters validated for clinical utility.

## Key findings

- The model achieved 76.10% sensitivity and 84.10% specificity for TBM diagnosis.
- The AUC of the model was 0.86, indicating strong diagnostic performance.
- Calibration curves and decision curve analysis confirmed the model's robustness.

## Abstract

We developed and validated a diagnostic scoring system for tuberculous meningitis (TBM) using 13 laboratory parameters, comparing tuberculous meningitis (TBM) and non-tuberculous meningitis (non-TBM).

This study enrolled patients diagnosed with meningitis. We retrospectively collected and analyzed demographic data (gender, age) and cerebrospinal fluid (CSF) parameters, including biochemical profiles and white blood cell counts with differential analysis. Variable selection was performed using least absolute shrinkage and selection operator (LASSO) regression. The dataset was randomly divided into a training set and a validation set. A diagnostic prediction model was developed using logistic regression in the training set, with nomograms constructed to visually demonstrate the diagnostic relationships. Decision curve analysis (DCA) was employed to assess the clinical utility of the model. Finally, the diagnostic performance of the model was evaluated in the validation set.

A total of 254 patients with meningitis were included in this study. LASSO regression analysis identified four predictive variables: CSF glucose, CSF chloride, CSF protein and CSF mononuclear cells proportion. These parameters were incorporated into a logistic regression model, with weighted factors generating a diagnostic score. A score of ≥ 3 was suggestive of TBM with a sensitivity of 76.10% and a specificity of 84.10%, and the area under the curve (AUC) values was 0.86 (95% CI 0.81-0.91). Both calibration curves and DCA validated the robust performance of model.

We developed and validated a clinically applicable diagnostic model for TBM using routinely available and low-cost CSF parameters. Our findings demonstrated that this scoring system provided reliable TBM diagnosis, particularly in countries and regions with limited microbial and radiological resources.