Cytokines in cerebrospinal fluid combined with machine learning improve the diagnostic accuracy and predict the progression of neurosyphilis

BackgroundSyphilis is a sexually transmitted disease (STD) caused by Treponema pallidum (TP) which could progress to neurosyphilis affecting the nervous system. However, the pathogenesis of neurosyphilis remain unknown.

In addition, current diagnosis of neurosyphilis is mainly based on serological tests and cerebrospinal fluid (CSF) analysis, with limitations in terms of sensitivity and specificity.MethodsFrom April 2021 to April 2023, 129 patients with syphilis were enrolled and divided into the non-neurosyphilis group (common syphilis, CS, n=19), asymptomatic neurosyphilis group (AN, n=77), symptomatic neurosyphilis group (NS, n=33) and a control group (n=15). Forty-eight cytokines/chemokines in the CSF samples were measured and analysed in combination with clinical indices.

Feature selections were further analysed for the using different machine learning model for the diagnosis and progression predicting of neurosyphilis.ResultsMachine learning-based feature selection identified a three-cytokine panel (RANTES, MIP-1β, and IL-3) that effectively distinguished syphilis patients from controls (AUC = 0.869, 95% CI: 0.821–0.917). For prediction of symptomatic neurosyphilis progression, IL-2Rα emerged as the optimal standalone biomarker, achieving an AUROC of 0.843 (95% CI: 0.784–0.902) for discriminating asymptomatic from symptomatic disease.

Multivariable analysis confirmed IL-2Rα retained independent discriminatory power after adjusting for age (multivariable AUC = 0.876, 95% CI: 0.827–0.925). Age-stratified ROC analysis revealed IL-2Rα maintains robust diagnostic utility across all adult age groups, with optimal Youden-derived cut-offs of 21.82 U/mL (sensitivity 100.0%, specificity 85.7%, AUC 0.939) for 35–45 years; 25.48 U/mL (sensitivity 83.3%, specificity 88.9%, AUC 0.889) for 45–55 years; and 24.54 U/mL (sensitivity 73.7%, specificity 84.6%, AUC 0.789) for >55 years.

The combination of IL-2Rα and IP-10 significantly improved predictive accuracy within specific age strata, achieving AUCs of 0.949 (35–45 years), 1.000 (45–55 years), and 0.814 (>55 years) when distinguishing symptomatic from asymptomatic neurosyphilis.ConclusionIn conclusion, IL-2Rα and IP-10 represent promising predictive biomarkers for neurosyphilis progression, with their combined application achieving high diagnostic accuracy for distinguishing symptomatic from asymptomatic disease. These findings establish a foundation for CSF cytokine-guided risk stratification in syphilis patients, although prospective validation in larger, age-matched cohorts is warranted to optimize clinical cut-offs and confirm generalizability across diverse populations.