Abstract: 【Abstract】 Objective To analyze differences in the expression of routine laboratory parameters and cerebrospinal fluid （CSF） examination indicators between patients with non-neurosyphilis （syphilis without nervous system involvement） and those with neurosyphilis, to screen for key predictive factors, and to construct a predictive model for neurosyphilis. Methods A retrospective analysis was conducted on the clinical data from patients with syphilis at the Fifth People's Hospital of Suzhou from 2019 to 2024. Patients with neurosyphilis and non-neurosyphilis who were hospitalized from November 2019 to June 2022 were included in the model cohort, and those hospitalized from January 2024 to October 2024 were included in the validation cohort. The patients' basic information and laboratory test indicators （including routine blood tests, CSF biochemical analysis, and syphilitic antibody tests） were collected. Statistical analysis was performed using the GraphPad software. The receiver operating characteristic （ROC） curve and the binary logistic regression method were used to analyze the predictive performance of key indicators in patients from the model cohort with SPSS software, and a predictive model for neurosyphilis was constructed. The performance of the neurosyphilis predictive model for neurosyphilis was validated based on relevant indicators from the validation cohort. Results The model cohort included 99 patients with non-neurosyphilis （including 49 males and 50 females）, and they were aged between 19 and 85 years, with an average age of 47 years; 69 patients with neurosyphilis were also included in the model cohort, including 58 males and 11 females, and they were aged between 26 and 73 years, with an average age of 51 years. The neurosyphilis group showed a significant increase in the median levels of CSF adenosine deaminase（1 U/L）and microprotein （711 mg/L）, white blood cell counts （0.009 × 10?/L）, as well as in the proportion of positive Pandy tests （35/69, 50.7%） compared with the non-neurosyphilis group （0 U/L, 309 mg/L, 0.002 × 10?/L, 2 /99 ［2.0%］, respectively, all P < 0.001）. Based on the ROC curve analysis, the CSF microprotein and white blood cell count had relatively high discriminative ability （area under the ROC curve ［AUC］ > 0.85）, while adenosine deaminase and the Pandy test showed moderate discriminative ability （0.7 < AUC < 0.85）. According to the above four indicators, the logistic regression analysis showed that CSF microprotein combined with CSF white blood cell counts could construct the best predictive model for neurosyphilis, with a prediction accuracy rate of 0.980, a sensitivity of 98.5%, and a specificity of 89.9%. The prediction formula was logit（p） = -9.926 + 0.015 × microprotein + 362.33 × CSF white blood cell count, with a cutoff value of ≥ -0.867. The validation cohort enrolled 72 patients with non-neurosyphilis and 51 with neurosyphilis, and there were significant differences in CSF microprotein levels and white blood cell counts between the two groups （both P < 0.001）. In the validation cohort, the predictive model demonstrated an accuracy of 86.2%, with a sensitivity of 83.6% and a specificity of 91.1% for predicting neurosyphilis. Conclusion The predictive model for neurosyphilis constructed by combining CSF microprotein and CSF white blood cell count may contribute to the early differential diagnosis of neurosyphilis.

Key words: Neurosyphilis, Syphilis, Microprotein, White blood cell count of cerebrospinal fluid, Prediction model