Discrepant Treponemal Test Results: Identification of Associated Risk Factors Through Machine Learning Technology in 18-Year Electronic Medical Records and National Claims Data.

Background: Syphilis is a prevalent disease diagnosed primarily through serological tests. Although one confirmatory treponemal tests (TT), including Treponema pallidum particle agglutination (TPPA) or fluorescent treponema antibody absorption (FTA-Abs), is required for syphilis diagnosis, multiple TTs are commonly administered throughout the disease course. Discrepant TT results can cause confusion and delay treatment. In this study, we identified the clinical characteristics of patients with discrepant TT results and developed a machine learning tool to evaluate the risk of TT discrepancies.

Materials and methods: In this retrospective cohort study, electronic health records were linked to national claims records collected from 2001 to 2018. Variables of interest in risk factor identification and machine learning model development included medical histories and demographic characteristics. The association between syphilis treatment and discrepant TT results was assessed.

Results: Among 5,780 eligible patients tested for syphilis, 133 (2.30%) had discrepant TT results. HIV and AIDS were identified as prominent risk factors associated with discrepant TT results (adjusted odds ratio = 2.6, 95% confidence interval = 1.4-4.7). Patients with a top 5% risk probability in the LightGBM model were 10 times more likely than others to have discrepant TT results. TPPA was more likely than FTA-Abs to become negative after treatment among patients with discrepant TT results (odds ratio = 9.18, 95% confidence interval = 2.05-41.07).

Conclusions: Risk factor identification and machine learning model development can support the interpretation of serological tests for syphilis, enabling accurate diagnosis and clinical decision-making.