# Optimizing metagenomic next-generation sequencing in CNS infections: a diagnostic model based on CSF parameters

## TL;DR

This study creates a model to predict when metagenomic sequencing is likely to detect CNS infections based on cerebrospinal fluid parameters.

## Contribution

A predictive model using CSF cell count and protein concentration to optimize metagenomic sequencing in CNS infections.

## Key findings

- CSF cell count and protein concentration are strong predictors of mNGS positivity.
- The model showed good diagnostic performance with an AUC of 0.782 in internal validation.
- External validation confirmed the model's clinical utility with 77.8% sensitivity and 67.7% specificity.

## Abstract

This study aimed to assess the association between routine cerebrospinal fluid (CSF) biochemical parameters and metagenomic next-generation sequencing (mNGS) results, and to develop a predictive model to optimize mNGS testing strategies in patients with suspected central nervous system (CNS) infections.

We retrospectively enrolled 110 patients with suspected CNS infections between December 2019 and January 2024. All underwent both CSF analysis and mNGS testing. Patients were divided into mNGS-positive (n = 62) and negative (n = 48) groups. Logistic regression identified independent predictors, and a nomogram was constructed based on CSF cell count and protein concentration. Model performance was assessed via receiver operating characteristic (ROC) curves, calibration plots, and decision curve analysis (DCA). Internal validation included 10-fold cross-validation and 1000-sample bootstrap. An external validation was performed using a cohort of 40 patients enrolled from another hospital campus (May–October 2024). The derivation cohort was retrospectively collected, whereas the external validation cohort was prospectively enrolled.

mNGS positivity rate was 56.36%, significantly higher than CSF culture (6.36%), with an overall diagnostic concordance of 79.09%. Compared to the mNGS-negative group, positive patients had significantly higher CSF cell counts, protein levels, turbidity, ICU admission (ICUA), antimicrobial regimen adjustment (AAR), and mortality, while glucose was significantly lower (P < 0.05). Logistic regression confirmed CSF cell count binary variables (BV) and protein-BV as independent predictors (P < 0.05). The areas under curve (AUCs) for the cell-count, protein-only, and combined models were 0.827, 0.813, and 0.782, respectively. Internal validation showed stable results: 10-fold CV AUC = 0.773 ± 0.184 (95% CI: 0.641–0.904), bootstrap AUC = 0.770 ± 0.064 (95% CI: 0.766–0.774). External validation yielded an AUC of 0.763 (95% CI: 0.554–0.918), with sensitivity and specificity of 77.8% and 67.7%. Calibration and DCA demonstrated good agreement and clinical utility.

CSF cell count and protein are reliable predictors of mNGS positivity. The model for practice showed consistent diagnostic performance and may aid in guiding precision mNGS testing, particularly in resource-constrained settings.