Introduction: The diagnosis of secondary bacterial meningitis in intensive care units and surgical departments is usually complicated by frequent false-negative cerebrospinal fluid (CSF) culture results because of previous antibacterial therapy or false-positive CSF culture results because of sample contamination. Some nonspecific clinical and laboratory criteria of CSF content can be used, but they do not provide 100% sensitivity and selectivity. The search for new diagnostic approaches is a relevant direction.
Objectives: The aim of this study was to construct various multivariate models based on biomarkers and metabolites in the cerebrospinal fluid for the diagnosis of secondary bacterial meningitis in patients with acute or chronical critical illness.
Methods: Patients with acute (n = 17, CSF samples = 19) or chronical critical illness (n = 35, CSF samples = 77) were divided into CSF sample groups without (n = 63, group I) and with secondary bacterial meningitis (n=33, group II). CSF samples were analyzed by UPLC-MS/MS to determine aromatic metabolites, and by electrochemiluminescence to determine interleukin-6 (IL-6), NSE, and S100 protein.
Results: Median values of CSF parameters (leukocytes, relative content of neutrophils, protein, IL-6, S100, 4-hydroxyphenyllactic, phenyllactic, indole-3-lactic acids) were statistically higher in group II compared to group I (p-value < 0.001), with CSF glucose higher in group I compared to group II (p-value < 0.001).
ROC-analysis of univariate models based on these CSF parameters did not reveal any model with 100% sensitivity and selectivity, while ROC-analysis of various multivariate models based on CSF biomarkers and metabolites demonstrated excellent prognostic ability for CSF sample stratification into two groups: the best model was constructed using Catboost [https://doi.org/10.48550/arXiv.1706.09516] with AUC-ROC=1.00 with 100% sensitivity and selectivity.
Conclusion: Multivariate models based on various CSF biomarkers and metabolites demonstrated better characteristics compared to univariate prognostic models for CSF sample stratification into groups with or without secondary bacterial meningitis.