Related Experiment Video
Updated: Jul 14, 2026
A Thin-skull Window Technique for Chronic Two-photon In vivo Imaging of Murine Microglia in Models of Neuroinflammation
Published on: September 19, 2010
Establishment of non-invasive diagnostic prediction model for HIV-positive asymptomatic neurosyphilis patients
Hongjing Guan1, Chunmiao Zou1, Xiaoyun Di2
1Department of Infectious Diseases, The School of Public Health of Nanjing Medical University, The Second Hospital of Nanjing, Nanjing, 210003, China.
Background:
Although guidelines strongly recommended universal screening for neurosyphilis (NS) in people living with HIV (PLWH), the invasiveness of lumbar puncture (LP) made comprehensive screening difficult, particularly among asymptomatic individuals, increasing the risk of missed diagnoses. Because of the common transmission routes, the co-infection rate of PLWH and syphilis was extremely high. Establishing effective diagnostic and predictive models for NS was essential to improve diagnostic accuracy and timeliness. This study aimed to develop a non-invasive method for detecting asymptomatic neurosyphilis (ANS) in PLWH.
Methods:
This study included HIV patients hospitalized at the Second Hospital of Nanjing from January 2013 to September 2024 who tested positive for syphilis antibodies and underwent their first LP to screen for NS. Missing data were addressed using multiple imputations, and the dataset was randomly divided into training and testing groups at a 7:3 ratio. Independent risk factors were identified through logistic regression and LASSO regression analyses. Model performance was evaluated using ROC, sensitivity, specificity, calibration curve, decision curve analysis (DCA), and clinical impact curve (CIC). Reliability was further assessed in subgroups defined by the time interval between HIV and syphilis detection, syphilis treatment, and HIV treatment.
Results:
A total of 544 participants were included. Logistic regression and LASSO regression analyses identified variables for the diagnostic prediction model, including age, sero-TRUST, viral load (VL), CD4/CD8 ratio, CD4%, and MRI-lacunar infarction. Both the test model (AUC: 0.75) and the training model (AUC: 0.80) demonstrated acceptable discrimination and good robustness across subgroups. Calibration curves and Hosmer-Lemeshow tests indicated good calibration. DCA and CIC showed clear clinical benefit of the model.
Conclusion:
The diagnostic model achieved an AUC greater than 70 for detecting ANS in PLWH, with good robustness confirmed by calibration and DCA, as well as subgroup analyses. This model can assist clinicians in identifying ANS in PLWH without requiring LP.
