Discrepant treponemal test results: Identification of associated risk factors through machine learning technology in 18-year electronic medical records and national claims data - PMC Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice Biomed J . 2025 Jul 19;49(2):100890. doi: 10.1016/j.bj.2025.100890 Search in PMC Search in PubMed View in NLM Catalog Add to search Discrepant treponemal test results: Identification of associated risk factors through machine learning technology in 18-year electronic medical records and national claims data Hsin-Yao Wang Hsin-Yao Wang a School of Medicine, National Tsing Hua University, Hsinchu, Taiwan b Department of Laboratory Medicine, Chang Gung Memorial Hospital at Linkou, Taoyuan, Taiwan Find articles by Hsin-Yao Wang a, b , Ru-Fang Hu Ru-Fang Hu c Department of Information Management, Chang Gung University, Taoyuan, Taiwan Find articles by Ru-Fang Hu c , Ting-Wei Lin Ting-Wei Lin b Department of Laboratory Medicine, Chang Gung Memorial Hospital at Linkou, Taoyuan, Taiwan Find articles by Ting-Wei Lin b , Wan-Ying Lin Wan-Ying Lin d Department of Medicine, University at Buffalo-Catholic Health System, Buffalo, NY, USA Find articles by Wan-Ying Lin d , Yu-Chiang Wang Yu-Chiang Wang e John A Burn School of Medicine, University of Hawaii, HI, USA f Department of Cardiology, Queens Medical Center, HI, USA Find articles by Yu-Chiang Wang e, f , Jang-Jih Lu Jang-Jih Lu g Division of Clinical Pathology, Taipei Tzu Chi Hospital, Buddhist Tzu Chi Medical Foundation, New Taipei City, Taiwan Find articles by Jang-Jih Lu g , Yi-Ju Tseng Yi-Ju Tseng h Department of Computer Science, National Yang Ming Chiao Tung University, Hsinchu, Taiwan i Computational Health Informatics Program, Boston Children's Hospital, Boston, MA, USA Find articles by Yi-Ju Tseng h, i, ⁎ Author information Article notes Copyright and License information a School of Medicine, National Tsing Hua University, Hsinchu, Taiwan b Department of Laboratory Medicine, Chang Gung Memorial Hospital at Linkou, Taoyuan, Taiwan c Department of Information Management, Chang Gung University, Taoyuan, Taiwan d Department of Medicine, University at Buffalo-Catholic Health System, Buffalo, NY, USA e John A Burn School of Medicine, University of Hawaii, HI, USA f Department of Cardiology, Queens Medical Center, HI, USA g Division of Clinical Pathology, Taipei Tzu Chi Hospital, Buddhist Tzu Chi Medical Foundation, New Taipei City, Taiwan h Department of Computer Science, National Yang Ming Chiao Tung University, Hsinchu, Taiwan i Computational Health Informatics Program, Boston Children's Hospital, Boston, MA, USA ⁎ Corresponding author. Department of Computer Science, National Yang Ming Chiao Tung University, No. 1001, Daxue Rd. East Dist., Hsinchu City 300093, Taiwan. [email protected] Received 2024 Apr 22; Revised 2025 May 30; Accepted 2025 Jul 16; Issue date 2026 Apr. © 2025 The Authors This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). PMC Copyright notice PMCID: PMC13091108 PMID: 40691897 Abstract 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 further assessed. Results Among 5780 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 = 14.7, 95 % confidence interval = 1.9–115.4). 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. Keywords: Syphilis, Serologic test, Treponemal test, Discrepant results, Risk factor Highlights • Discrepant treponemal tests delay syphilis diagnosis and management. • With our model, top 5 % risk patients were 10x more likely to have discrepant results. • The study findings can guide accurate diagnosis and improve patient care. • Negative TPPA is more strongly associated with syphilis treatment than FTA-Abs. 1. Introduction Syphilis, known as “the great imitator,” is a globally widespread sexually transmitted disease [ 1 ] Transmission routes include sexual intercourse and blood transfusion from an infected person. Syphilis remains a major public health concern worldwide [ [2] , [3] , [4] , [5] ]. In 2012, the World Health Organization estimated that 6 million new cases of syphilis develop annually among individuals aged 15–49 years [ 6 ]. Between 2014 and 2018, the number of reported syphilis cases in the United States increased by approximately 80 % [ 7 ]. Since the 2000s, a growing number of isolated syphilis outbreaks have emerged in North America and Europe [ 8 ]. Consequently, the World Health Organization has set a goal to reduce the incidence of syphilis by 2030 [ 6 ]. In Taiwan, the incidence of syphilis has increased over the years, and clinical personnel are currently required to report syphilis cases to health authorities [ 9 ]. Serological tests are the typical standard for syphilis diagnosis. The conventional testing algorithm comprises a nontreponemal test (NTT) for screening followed by a confirmatory treponemal test (TT). Treponema pallidum particle agglutination (TPPA) and fluorescent treponema antibody absorption (FTA-Abs) are two widely used TTs [ 10 ]. These tests are crucial in confirming syphilis infection and can be applied simultaneously to improve diagnostic accuracy. A meta-analysis examining the simultaneous use of TPPA and FTA-Abs for syphilis diagnosis revealed greater sensitivity and specificity compared with the separate administration of these tests [ 11 ]. Although syphilis diagnosis requires only one TT, multiple TTs are commonly administered throughout the disease course. Patients may undergo different TTs at various health-care institutes. Different TTs may also be ordered simultaneously, particularly when physicians are less familiar with TTs or in areas where TTs are relatively inexpensive (e.g., Taiwan). However, certain factors can interfere with TT results and complicate syphilis diagnosis [ [12] , [13] , [14] ]: advanced age, autoimmune diseases (e.g., systemic lupus, scleroderma), intravenous drug use, and pregnancy [ 13 ]. Discrepant TT results can cause confusion and delay treatment. Therefore, large-scale studies incorporating real-world data and machine learning models are required to identify the risk of and risk factors associated with discrepant TT results. In this study, we linked large-scale electronic health records from a tertiary medical center in Taiwan with nationwide claims data to identify the risk factors associated with discrepant TT results. We also developed a machine learning model to identify high-risk patients. The risk factors identified provide valuable information for the interpretation of TT results, facilitating early diagnosis and treatment and improving follow-up care for patients with discrepant TT results. 2. Materials and methods 2.1. Study scheme We examined the risk factors associated with discrepant TT results and developed a machine learning model capable of evaluating an individual's risk of TT discrepancy to support clinical decision-making. Additionally, we explored the association between TT type and the risk of TT discrepancy after syphilis treatment. The study scheme is illustrated in [ Fig. 1 ]. Fig. 1. Open in a new tab Study scheme. First, the associations between clinical features and discrepant TT results were examined. Risk factors linked to discrepant TT results were identified, and an online calculator was developed to estimate the risk of discrepant TT results. Second, the association between TT type and treatment in patients with discrepant TT results was assessed. TPPA: Treponema pallidum particle agglutination; FTA-Abs: fluorescent treponema antibody absorption. 2.2. Data sources In this retrospective cohort study, records from the National Health Insurance Research Database (NHIRD) were linked to those from the Chang Gung Research Database (CGRD) [ 15 ]. The National Health Insurance (NHI) program covers more than 99 % of Taiwan's population, and the NHIRD prospectively records standardized health-care service data submitted to the NHI program [ 16 ]. Diagnoses made before 2016 are registered using codes from the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM), and subsequent diagnoses are registered using codes from the International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM). NHIRD data are routinely validated by the NHI Bureau [ 17 ]. The CGRD includes deidentified electronic medical records (EMRs) with laboratory test results from the largest group of health-care providers in Taiwan [ 15 ]. To obtain comprehensive medical histories, we linked EMRs from the CGRD with those from the NHIRD by using encrypted personal identification numbers. Four NHIRD data sets were retrieved, including ambulatory care expenditures by visit, detailed ambulatory care orders, inpatient expenditures by admission, and detailed inpatient orders. Before analysis, data were anonymized and deidentified to protect patient privacy [ 18 , 19 ]. Data on the patients’ date of birth, sex, laboratory test results, and treatment modalities were extracted from the CGRD, and data on comorbidities were extracted from the NHIRD. The Institutional Review Board of Chang Gung Medical Foundation approved this study (approval no. 201801524B0) and waived the requirement for patient consent. 2.3. Participant identification This study included Taiwanese patients who underwent at least two confirmatory TTs, identified using laboratory test results recorded in the CGRD, at the Linkou branch of Chang Gung Memorial Hospital (CGMH) from January 2001 to September 2018. TTs included semiquantitative TPPA, semiquantitative T. pallidum hemagglutination, and qualitative FTA-ABS. TTs performed on individuals aged under 1 year were excluded because of test result instability. Weak, equivocal, or borderline test results and tests of cerebrospinal fluid specimens were also excluded. Patients not from Taiwan and those with no outpatient or inpatient visit records in the NHIRD were excluded. 2.4. Definition of discrepant test results Participants were divided into two groups depending on their TT results. The discrepancy group included those with a positive TT result followed by a negative TT result. All other patients, such as those with all positive (consistent) or all negative (noninfected) TT results or those with a negative followed by a positive TT result (newly infected), were defined as the nondiscrepancy group. The index date for the discrepancy group was defined as the date at which the TT results changed from positive to negative as a result of either false-positive results for the earlier test or false-negative results for the later test. The index date for the nondiscrepancy group was defined as the date of the last TT. Age was calculated from the date of birth and the index date. 2.5. Comorbidity assessment Comorbidities were identified on the basis of each patient's diagnostic history, extracted from diagnostic codes in inpatient, outpatient, and emergency department records. General comorbidities were assessed using the Elixhauser Comorbidity, which includes 30 comorbidity categories and was developed by the Healthcare Cost and Utilization Project [ [20] , [21] , [22] , [23] ]. Our analysis also included 49 comorbidities associated with discrepant TT results, as identified in previous studies or by clinical experts [ Table S1 ] [ [24] , [25] , [26] ]. We used the dxpr package [ 27 ] to generate these comorbidities from diagnostic codes recorded in at least three encounters. 2.6. Sensitivity analysis We conducted univariable and multivariable sensitivity analyses to evaluate our method of comorbidity identification. To examine the effects of comorbidity definition, we redefined comorbidities as at least one or two recorded care encounters related to a given condition. 2.7. Statistical and machine learning methods We applied a range of modeling techniques with complementary strengths, including stepwise logistic regression, least absolute shrinkage and selection operator (LASSO) regression, eXtreme Gradient Boosting (XGBoost) [ 28 ], and LightGBM [ 29 ], to construct prediction models for discrepant TT results. Stepwise logistic regression is a traditional statistical approach widely applied in medical research. Interpretability is a key advantage in the medical domain [ 30 ]. LASSO regression addresses several limitations of stepwise methods by incorporating L1 regularization, which enables simultaneous feature selection and coefficient shrinkage [ 31 ]. Given the relatively large number of potential risk factors in our data set, the capacity of LASSO regression to shrink the coefficients of less important variables to zero mitigated multicollinearity concerns during model development. To achieve complex pattern recognition, we incorporated two gradient boosting algorithms: XGBoost and LightGBM. These ensemble methods can capture nonlinear relationships and interactions between variables, which regression-based approaches may overlook. XGBoost uses an ensemble of decision trees to improve predictive performance [ 28 ]. Its robust regularization techniques prevent overfitting, facilitating its widespread adoption in real-world applications [ [32] , [33] , [34] , [35] ]. LightGBM offers advantages in computational efficiency combined with high predictive performance through a leaf-wise growth strategy. It incorporates gradient-based one-side sampling, which prioritizes data points with large gradients for training, and exclusive feature bundling, which reduces dimensionality by combining mutually exclusive features to improve performance and efficiency. LightGBM is effective in managing categorical features, which comprise a major portion of our predictor variables [ 29 ]. 2.8. Model establishment and evaluation Demographic characteristics and comorbidities that significantly differed between both groups ( p < 0.1 for more than three cases [ 18 ]) were selected as independent variables for the development of models aimed at predicting the risk of TT discrepancy. To accurately evaluate model performance, we randomly divided the data into training and evaluation sets for 100 repetitions. In each training set, we fine-tuned the hyperparameters of the models through a grid search following a five-fold cross-validation approach [ Table S2 ]. Finally, we examined the performance of the prediction model by using the evaluation set [ [36] , [37] , [38] ]. We also assessed model performance by using the area under the receiver operating characteristic curve (AUC), and we estimated feature importance on the basis of information provided by the LightGBM model. 2.9. Positive predictive score of discrepant TT risk On the basis of the predictive probabilities generated by the machine learning models, we calculated a positive predictive score (PPS) to illustrate each individual's risk of discrepant TT results. We sorted all patients in descending order of their predictive probabilities and divided them into 20 equal groups (vigintiles) depending on their risk level [ 39 ]. For each vigintile, a PPS was calculated using the following formula: P P S = P r o p o r t i o n o f d i s c r e p a n t c a s e s i n t h e v i g i n t i l e B a s e l i n e p r o p o r t i o n o f d i s c r e p a n t c a s e s = n u m b e r o f d i s c r e p a n t T T s i n t h e v i g i n t i l e n u m b e r o f p a t i e n t s i n t h e v i g i n t i l e t o t a l n u m b e r o f d i s c r e p a n t T T s t o t a l n u m b e r o f p a t i e n t s In the first vigintile, patients in the top 5 % risk category for discrepant TT results were examined. To generate a PPS for the first vigintile, we divided the number of patients with discrepant TT results by the total number of patients in the first vigintile (numerator). We then normalized the resulting value by the baseline proportion of discrepant cases in the entire study population (denominator) to generate a PPS. The PPS provides clinical caregivers with an explicit illustration of the risk of discrepant TT results obtained for a random index case. 2.10. Review of syphilis treatments for patients with discrepant TT results We reviewed CGRD prescription records to determine whether patients with discrepant TT results received syphilis treatment. Treatment status was classified as “received treatment” for patients who received either recommended or suboptimal treatments or “no treatment” for patients who were not treated for syphilis [ 40 ]. We also recorded the time of treatment and analyzed the association between treatment status and discrepant TT results. 2.11. Statistical analysis A Kruskal–Wallis test was used to analyze differences in medians. Chi-square and Fisher's exact tests were used to conduct a univariate analysis of categorical variables. To identify factors associated with TT discrepancy, we performed multivariable logistic regression with Firth's correction to solve the problem of separation caused by rare events. Crude and adjusted odds ratios (aORs) were reported with corresponding 95 % confidence intervals (CIs) estimated using the profile penalized likelihood method [ 41 ]. All analyses were conducted using R software version 4.0.3 (R Foundation for Statistical Computing, Vienna, Austria) and SAS software version 9.4 (SAS Institute, Cary, NC, USA). All statistical tests were two-sided, with statistical significance set as p < 0.05. We followed the Strengthening the Reporting of Observational Studies in Epidemiology reporting guidelines [ 42 ]. 3. Results 3.1. Study population characteristics A total of 11,135 patients were available for inclusion after CGRD and NHIRD data were linked. Of these patients, 5780 met our inclusion criteria [ Figure S1 ]. Among patients who underwent at least two confirmatory tests, 133 (2.3 %) had discrepant TT results and 5647 (97.7 %) had nondiscrepant results. Patients with discrepant results (67.8, interquartile range [IQR] = 35.5) were significantly older than those with nondiscrepant results (44.8, IQR = 36.2, p < 0.001; [ Table S3 ]. No differences in sex distribution were observed between the two groups. [ Table S4 ] presents the TT result patterns for patients with discrepant TT results. The median interval between the initial positive and subsequent negative TT in the discrepancy group was 175 days (IQR = 36–434 days, [ Figure S2 ], which did not significantly differ from the interval between two positive TTs ( P = 0.40). The majority of patients with discrepant TT results had TPPA-positive followed by TPPA-negative results (94, 70.7 %). 3.2. Comorbidity analysis [ Table 1 ] presents the crude ORs and aORs for age and sex. The comorbidities associated with discrepant TT results were identified as HIV and AIDS (aOR = 2.6, 95 % CI = 1.4–4.7), other and unspecified osteoarthritis (aOR = 3.3, 95 % CI = 1.3–7.1), and pregnancy (aOR = 5.0, 95 % CI = 1.8–11.6). In the sensitivity analysis, HIV and AIDS and other and unspecified osteoarthritis were identified as risk factors associated with discrepant TT results on the basis of both one diagnosis [ Table S5 ] and two diagnoses comorbidities definitions [ Table S6 ]. Table 1. Association between comorbidities and discrepant TT results. Comorbidity OR Crude OR (95 % CI) P value aOR a (95 % CI) P value HIV and AIDS 1.68 (0.92–2.86) 0.070 2.63 (1.4–4.67) 0.004 ∗ Deficiency anemia 0.82 (0.2–2.19) 0.733 0.62 (0.17–1.6) 0.361 Rheumatoid arthritis collagen vascular diseases 1.91 (0.74–4.06) 0.128 1.87 (0.75–3.92) 0.163 Congestive heart failure 2.14 (0.83–4.55) 0.075 1.31 (0.52–2.77) 0.540 Chronic pulmonary disease 2.13 (1.24–3.44) 0.004 ∗ 1.49 (0.86–2.46) 0.146 Depression 1.87 (0.78–3.78) 0.115 1.44 (0.62–2.89) 0.366 Diabetes without chronic complications 1.84 (1.11–2.89) 0.012 ∗ 1.13 (0.67–1.81) 0.638 Diabetes with chronic complications 1.75 (0.88–3.14) 0.080 1.14 (0.58–2.05) 0.688 Hypertension, uncomplicated 1.67 (1.13–2.42) 0.009 ∗ 0.86 (0.56–1.32) 0.500 Hypertensive heart disease without heart failure 1.39 (0.49–3.12) 0.474 0.89 (0.33–1.96) 0.792 Liver disease 0.59 (0.23–1.24) 0.211 0.56\ (0.23–1.15) 0.124 Fluid and electrolyte disorders 4.46 (1.53–10.37) 0.002 ∗ 2.57 (0.91–5.96) 0.071 Other neurological disorders 1.98 (1–3.55) 0.034 ∗ 1.14 (0.57–2.09) 0.684 Peripheral vascular disease 3.82 (1.46–8.27) 0.002 ∗ 2.35 (0.92–5.08) 0.071 Solid tumor without metastasis 0.55 (0.14–1.47) 0.310 0.4 (0.11–1.03) 0.059 Chronic peptic ulcer disease 1.47 (0.52–3.3) 0.404 1.14 (0.42–2.51) 0.770 Valvular disease 2.28 (0.8–5.17) 0.077 1.62 (0.59–3.6) 0.317 Arthropathy 2.58 (0.62–7.15) 0.114 1.92 (0.52–5.07) 0.290 Inflammatory bowel disease 1.5 (0.46–3.65) 0.429 1.6 (0.52–3.75) 0.372 Inflammatory spondylopathies 2.24 (0.78–5.05) 0.085 1.69 (0.62–3.75) 0.280 Other and unspecified osteoarthritis 5.18 (1.96–11.39) <0.001∗ 3.27 (1.27–7.14) 0.017 ∗ Other and unspecified soft tissue disorders not elsewhere classified 2.07 (1.07–3.65) 0.019 ∗ 1.33 (0.69–2.38) 0.371 Pneumonia 2.52 (1.17–4.78) 0.009 ∗ 1.66 (0.78–3.19) 0.178 Polyosteoarthritis 3.41 (1.18–7.82) 0.010 ∗ 2.07 (0.75–4.72) 0.148 Pregnancy 2.79 (0.97–6.35) 0.029 ∗ 4.97 (1.76–11.6) 0.004 ∗ Rheumatoid arthritis 3.16 (0.76–8.82) 0.057 2.6 (0.7–6.97) 0.137 Urticaria 2.04 (0.79–4.33) 0.096 1.9 (0.76–3.97) 0.155 Vasomotor and allergic rhinitis 1.44 (0.67–2.71) 0.299 1.33 (0.63–2.48) 0.424 Open in a new tab a Adjusted for covariates: age and sex. ∗ p < 0.05. 3.3. Model prediction performance for discrepant TT results We used demographic characteristics and comorbidities that differed between the two groups ( p < 0.1) as variables to construct a prediction model for discrepant TT results. Among the four prediction algorithms tested, LightGBM significantly outperformed the others, with an AUC of 0.705 (95 % CI = 0.697–0.713, p < 0.001). Lower AUC values were observed for the other three prediction algorithms (Lasso regression, AUC = 0.657, 95 % CI = 0.648–0.666; stepwise logistic regression, AUC = 0.669, 95 % CI = 0.66–0.678; and gradient boosting algorithm, AUC = 0.676, 95 % CI = 0.666–0.687). The most important variables in the LightGBM model were HIV and AIDS, pregnancy, sex, and age [ Fig. 2 ]. Fig. 2. Open in a new tab Feature importance in the LightGBM model. Feature importance refers to information gain, which represents the average entropy difference before versus after a data set is split depending on each feature in trees. The middle line in each box represents the median. The dot points are single data points from 100 random training/evaluation splits for each feature. 3.4. PPS of discrepant TT risk The baseline proportion of discrepant cases within the study population was 0.02. We used the PPS formula to divide all patients into 20 categories depending on their risk of having discrepant TT results [ Fig. 3 ]. The majority of patients with discrepant TT results were classified into the first risk vigintile (with the top 5 % risk probability), in which the risk of having discrepant TT results was 10 times higher than the baseline frequency for the total study population. Using the PPS formula, we developed an Web-based application to estimate the risk of discrepant TT results [ Figure S3 ]. Fig. 3. Open in a new tab PPS of the LightGBM model. PPS refers to the proportion of discrepant test results within each vigintile divided by the baseline proportion of discrepant cases in the study population. For example, to generate a PPS for the first vigintile, we divided the number of patients with discrepant TT results by the total number of patients in the first vigintile (numerator). We then normalized the resulting value in accordance with the baseline proportion of discrepant TT results in the total study population (denominator) to generate a PPS. 3.5. Syphilis treatment for patients with discrepant TT results Given that the NTT titer is known to decrease in response to syphilis treatment [ 43 ], we examined the association between syphilis treatment and TT discrepancies. Of 133 patients with discrepant TT results, 71 (53.4 %) received antibiotic treatment for syphilis, whereas 62 (46.6 %) did not receive treatment. Among those who received treatment, 57 were treated before discrepant TT results were observed. A greater association was observed between syphilis treatment and negative TPPA results than between syphilis treatment and negative FTA-Abs results (OR = 14.7, 95 % CI = 1.9–115.4; [Table 2] ). Table 2. Association between TT type and syphilis treatment among patients with discrepant TT results. Negative TPPA was recorded after a positive TT; negative FTA-Abs was recorded after a positive TT. Without overlapping patients (w/) Negative TPPA Negative FTA-Abs No treatment before negative TT a 56 (61) 15 (20) Treatment before negative TT b 55 (56) 1 (2) Open in a new tab OR = 14.7 (1.9–115.4) c . a Four untreated patients and one patient treated after seroreversion had both negative TPPA and negative FTA-Abs results after a positive TT. b One patient received syphilis treatment before seroreversion had both negative TPPA and negative FTA-Abs results after a positive TT. b OR was calculated without overlapping patients ( n = 127). 4. Discussion Theoretically, practitioners need to administer only one type of TT in either traditional or reverse algorithms for syphilis diagnosis. However, in real-world settings, patients often undergo multiple TTs with different analytical methods. First, different TTs may be administered throughout the disease course. Second, different TTs may be administered across health-care institutions. Finally, different TTs may be ordered simultaneously by physicians who are less familiar with them. Given the rapid increase in new syphilis cases [ 7 , 44 , 45 ] and the limited research into discrepant results in confirmatory tests for syphilis, factors affecting TT results require identification. In this large-scale study, we analyzed demographic information and comorbidities to determine the features associated with discrepant TT results. Multivariable and discrepancy prediction model analysis revealed sex, age, HIV and AIDS, and pregnancy as the most relevant features. Syphilis treatment was more strongly associated with negative TPPA results than with negative FTA-Abs results after an initial positive TT. We also investigated the comorbidities associated with discrepant TT results and identified the TT type most likely to be affected by syphilis treatment. Given the crucial role of TTs in syphilis diagnosis, these results may contribute to diagnostic accuracy. TT discrepancies may be more common than previously assumed, particularly in settings with inexpensive testing (<US$10, fully covered by the national health insurance program of Taiwan) and particularly because of the widespread adoption of TTs in syphilis management. Although the requests for TTs in our study did not always follow clinical guidance, this practice allowed us to identify patients with multiple and discrepant TT results within the study population. Among 5780 patients with multiple TTs, discrepancies were observed in 2.3 %. This finding raises a concern that reliance on a single TT may increase the risk of false-negative results for patients with syphilis and false-positive results for patients without syphilis. Such errors can have serious consequences, including a missed diagnosis of syphilis in pregnant women or a delayed diagnosis in patients with HIV/AIDS. We also identified patterns in TT discrepancies, underscoring the complexity of TT result interpretation. To improve syphilis management, the frequency and criteria for screening and diagnosis with TTs may require reevaluation. A previous study revealed that HIV/AIDS and pregnancy may lead to false-positive TT results [ 13 ], which is consistent with our findings [ Table 1 ] and [ Fig. 2 ]. In our study, we identified unspecified osteoarthritis as a risk factor associated with TT discrepancies. The mechanism underlying the association between unspecified osteoarthritis and discrepant TT results has not been sufficiently explored. Immune cell dysregulation in osteoarthritis may contribute to discrepant TT results because of changes in the immune environment caused by the complex interplay of immune responses, which involve T cells, macrophages, and other immune cells and cytokines [ 46 ]. Immune system dysregulation can also affect the production of autoantibodies [ 47 ] and circulating immune complexes [ 48 ] in osteoarthritis. These immune products can interfere with various serological tests [ 49 ], including TTs. Overall, the large-scale nature of our study allowed us to comprehensively analyze TT discrepancies. Our results offer novel insights for the development of diagnostic tools for syphilis diagnosis or for the investigation of the mechanisms underlying TT discrepancies. We combined claims data (NHIRD) with EMRs (CGRD) to examine key laboratory test results that are unavailable in claims data. To the best of our knowledge, this is the first study to develop a machine learning model incorporating multiple comorbidities and demographic characteristics to identify factors associated with discrepant syphilis test results. Our findings contribute to the epidemiological understanding of discrepant TT results and establish a baseline to evaluate the effects of syphilis treatment on TT discrepancies. We also developed an online risk calculator based on our predictive model of TT discrepancies [ Figure S3 ]. Rather than risk probability, a PPS was applied to indicate the risk of discrepant TT results because it is more easily understandable [ Fig. 3 ]. The PPS describes an individual's risk compared to that of the overall population, and it provides valuable information for both physicians and patients. A tool is only useful in clinical practice if the information it conveys can be easily understood and communicated [ 50 ]. TTs are considered both a confirmatory test in syphilis diagnosis and a “syphilis scar,” so named because positive results typically last for a patient's lifetime, even after syphilis treatment. The TT discrepancies that we observed imply that TTs are not as confirmatory as has long been believed [ 51 ]. We also investigated the effect of syphilis treatment on TT discrepancies, observing that TPPA results were more likely than FTA-Abs results to become negative after treatment [ Table 2 ]. The reasons why TPPA is more vulnerable than FTA-Abs to this change are complex and have not yet been fully elucidated. A previous study revealed TPPA seroreversion after treatment [ 52 ], likely caused by the antigenicity of reagents. Although only part of the antigen is coated onto the particles in TPPA tests, the entire T. pallidum bacterium is used to capture antibodies in FTA-Abs tests [ 10 ]. Thus, the antigenicity of FTA-Abs likely remains more comprehensive and intact than that of TPPA because the entire surface antigens on T. pallidum are theoretically preserved in FTA-Abs. This comprehensive antigenicity may enable the detection of various antisyphilis antibodies in serum, even when some antibodies disappear after treatment. However, this explanation relies on the basic concept of testing and antigenicity, which may not apply to the numerous commercial TTs and epitopes used for testing [ 10 ]. Therefore, caution is recommended when extending our observations to other TTs from different providers. The LightGBM model demonstrated significantly higher predictive performance compared with all other models constructed in this study. This finding underscores the advantage of LightGBM and other ensemble tree methods in capturing complex interactions between demographic and comorbidity features that may contribute to TT discrepancies. Feature importance analysis within the LightGBM model revealed that HIV/AIDS status, pregnancy, sex, and age were the most influential predictors of discrepant TT results. This finding corroborates known clinical risk factors and highlights the utility of machine learning approaches in quantifying their relative contributions. This study has several limitations. First, out-of-pocket services were not included in the diagnosis data obtained from the NHIRD [ 53 ]. Second, we identified comorbidities through diagnosis codes. In cases where individual blood test results were unavailable, doctors may have provided a tentative diagnosis [ 54 ]. Therefore, we defined the presence of comorbidities as at least three recorded medical encounters related to a given condition. In addition, laboratory parameters and behavioral factors were not analyzed. These variables can be incorporated into future analyses to refine risk stratification, particularly for populations with complex clinical profiles. Third, disease stage was not available in the EMRs, hindering the analysis of the effect of disease stage on TT discrepancies. Including disease stages in medical records can enable researchers to better understand the relationship between these two factors. Fourth, although the LightGBM model demonstrated reasonable predictive performance, we must acknowledge several limitations that warrant further investigation. For instance, the model primarily incorporated demographic factors and comorbidities identified through claims data. Additional variables, such as intercurrent infections, socioeconomic status, vaccination history, and medication use, were not included because of data availability constraints. These factors may affect immune responses or test interference, and their inclusion may improve model accuracy if integrated. Finally, because we only included data from the Linkou branch of CGMH, the generalizability of our findings may be limited. However, as a major medical center in northern Taiwan with approximately 3800 beds that serves nearly 4 million outpatients every year, the Linkou branch of CGMH represents one of the largest patient cohorts in Taiwan [ 55 ]. Given these limitations, further research is required to better understand the association between comorbidities and discrepant TT results. This study reveals the associations of demographic and comorbidity data with discrepant TT results. We linked EMRs with claims data to identify risk factors that may affect TT results. Age, HIV and AIDS, pregnancy, and osteoarthritis are influential factors associated with discrepant TTs. These findings can serve as a reference for improving diagnostic decisions in preventive health care for syphilis and guiding related interventions. Further research is required to improve health care and clinical conditions related to syphilis, which remains a major public health concern. Ethical approval and consent to participate The Institutional Review Board of Chang Gung Medical Foundation approved this study (approval no. 201801524B0) and waived the requirement for patient consent. Consent for publication Not applicable. Availability of data and materials All data used in this study are restricted under the law regarding the protection of patient information in accordance with local and institutional requirements. The application process for access to these data can be made available upon reasonable request to the corresponding author. Funding This research was supported by the National Science and Technology Council of Taiwan (grant nos. NSTC 111-2320-B-182A-002-MY2 and NSTC 111-2628-E-A49-026-MY3), Chang Gung Memorial Hospital (grant nos. CORPG2P0072, CMRPG3M0851 and CMRPG3L1011), and the Higher Education Sprout Project of National Yang Ming Chiao Tung University and the Ministry of Education, Taiwan (grant no. CGMH-NYCU-114-CORPG2P0072). The funders had no role in the study design and procedures; data collection, management, analysis, and interpretation; manuscript preparation, review, and approval; or the decision to submit the manuscript for publication. Declaration of competing interest The authors have no conflicts of interest to declare. Acknowledgments The authors would like to thank the Health and Welfare Data Science Center, Ministry of Health and Welfare Chang Gung sub-centers and the data Health and Welfare Data Science Center, Ministry of Health and Welfare (HWDC, MOHW) for administrative support, promoting Chang Gung medical system to national health database application to conduct value-added research. They would also like to acknowledge the technical support provided by the Department of Information System Management, Chang Gung Memorial Hospital. Data for this study were collected from the CGRD of CGMH and the NHIRD of the Ministry of Health and Welfare. The interpretations and conclusions outlined herein do not represent the views of either institution. Footnotes Appendix A Supplementary data related to this article can be found at https://doi.org/10.1016/j.bj.2025.100890 . Appendix A. Supplementary data The following is the supplementary data related to this article: Multimedia component 1 mmc1.docx (584.2KB, docx) References 1. Çakmak SK, Tamer E, Karadağ AS, Waugh M. Syphilis: a great imitator. Clin Dermatol. 2019;37(3):182–191. doi: 10.1016/j.clindermatol.2019.01.007. [ DOI ] [ PubMed ] [ Google Scholar ] 2. Spiteri G, Unemo M, Mårdh O, Amato-Gauci AJ. The resurgence of syphilis in high-income countries in the 2000s: a focus on Europe. Epidemiol Infect. 2019;147 doi: 10.1017/S0950268819000281. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 3. Ghanem KG, Ram S, Rice PA. The modern epidemic of syphilis. N Engl J Med. 2020;382(9):845–854. doi: 10.1056/NEJMra1901593. [ DOI ] [ PubMed ] [ Google Scholar ] 4. de Souza RL, dos Santos Madeira LDP, Pereira MVS, da Silva RM, de Luna Sales JB, Azevedo VN, et al. Prevalence of syphilis in female sex workers in three countryside cities of the state of Pará, Brazilian Amazon. BMC Infect Dis. 2020;20(1):129. doi: 10.1186/s12879-020-4850-1. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 5. Harmon ED, Syphilis Robertson EW. A growing concern. Nurse Pract Am J Prim Health Care. 2019;44(8):21–28. doi: 10.1097/01.NPR.0000558159.61349.cb. [ DOI ] [ PubMed ] [ Google Scholar ] 6. World Health Organization Global health sector strategy on sexually transmitted infections 2016–2021 towards ending STIs. 2016. https://apps.who.int/iris/bitstream/handle/10665/246296/?sequence=1 [Accessed 5 August 2023] 7. Ghanem KG, Ram S, Rice PA. The modern epidemic of syphilis. N Engl J Med. 2020;382(9):845–854. doi: 10.1056/NEJMra1901593. [ DOI ] [ PubMed ] [ Google Scholar ] 8. Kitayama K, Segura ER, Lake JE, Perez-Brumer AG, Oldenburg CE, Myers BA, et al. Syphilis in the Americas: a protocol for a systematic review of syphilis prevalence and incidence in four high-risk groups, 1980-2016. Syst Rev. 2017;6(1):195. doi: 10.1186/s13643-017-0595-3. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 9. Lai SK, Chen CM, Kuo HW, Liu DP. Epidemiology of syphilis and gonorrhea in taiwan_ 2005–2016. Taiwan Epidemiol Bull. 2017;33:460–468. [ Google Scholar ] 10. Morshed MG, Singhb AE. Recent trends in the serologic diagnosis of syphilis. Clin Vaccine Immunol. 2015;22(2):137–147. doi: 10.1128/CVI.00681-14. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 11. Bristow CC, Klausner JD, Tran A. Clinical test performance of a rapid point-of-care syphilis treponemal antibody test: a systematic review and meta-analysis. Clin Infect Dis. 2020;71:S52–S57. doi: 10.1093/cid/ciaa350. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 12. Felman YM, Nikitas JA. Syphilis serology today. Arch Dermatol. 1980;116(1):84–89. [ PubMed ] [ Google Scholar ] 13. Forrestel AK, Kovarik C, Katz KA. Sexually acquired syphilis: laboratory diagnosis, management, and prevention. J Am Acad Dermatol. 2020;82(1):17–28. doi: 10.1016/j.jaad.2019.02.074. [ DOI ] [ PubMed ] [ Google Scholar ] 14. Hook EW. 3rd. Syphilis. Lancet. 2017;389(10078):1550–1557. doi: 10.1016/S0140-6736(16)32411-4. [ DOI ] [ PubMed ] [ Google Scholar ] 15. Shao SC, Chan YY, Kao Yang YH, Lin SJ, Hung MJ, Chien RN, et al. The Chang Gung Research Database—a multi-institutional electronic medical records database for real-world epidemiological studies in Taiwan. Pharmacoepidemiol Drug Saf. 2019;28(5):593–600. doi: 10.1002/pds.4713. [ DOI ] [ PubMed ] [ Google Scholar ] 16. Lin LY, Warren-Gash C, Smeeth L, Chen PC. Data resource profile: the national health insurance research database (NHIRD) Epidemiol Health. 2018;40 doi: 10.4178/epih.e2018062. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 17. Hsieh C-Y, Su C-C, Shao S-C, Sung S-F, Lin S-J, Yang Kao Y-H, et al. Taiwan’s national health insurance research database: past and future. Clin Epidemiol. 2019;11:349–358. doi: 10.2147/CLEP.S196293. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 18. Lin LY, Warren-Gash C, Smeeth L, Chen PC. Data resource profile: the national health insurance research database (NHIRD) Epidemiol Health. 2018;40 doi: 10.4178/epih.e2018062. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 19. Hsieh C-Y, Su C-C, Shao S-C, Sung S-F, Lin S-J, Yang Kao Y-H, et al. Taiwan’s national health insurance research database: past and future. Clin Epidemiol. 2019;11:349–358. doi: 10.2147/CLEP.S196293. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 20. Walraven C Van, Quan H, Forster AJ. A modification of the elixhauser comorbidity measures into a point system for hospital death using administrative data. Med Care. 2009;47(6):626–633. doi: 10.1097/MLR.0b013e31819432e5. [ DOI ] [ PubMed ] [ Google Scholar ] 21. Menendez ME, Neuhaus V, Van Dijk CN, Ring D. The Elixhauser comorbidity method outperforms the Charlson index in predicting inpatient death after orthopaedic surgery. Clin Orthop Relat Res. 2014;472(9):2878–2886. doi: 10.1007/s11999-014-3686-7. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 22. Elixhauser A, Steiner C, Harris D, Coffey R. Comorbidity measures for use with administrative data methods defining important comorbidities. Med Care. 1998;36(1):8–27. doi: 10.1097/00005650-199801000-00004. [ DOI ] [ PubMed ] [ Google Scholar ] 23. Quan H, Sundararajan V, Halfon P, Fong A, Burnand B, Luthi J-C, et al. Coding algorithms for defining comorbidities in ICD-9-CM and ICD-10 administrative data. Med Care. 2005;43(11):1130–1139. doi: 10.1097/01.mlr.0000182534.19832.83. [ DOI ] [ PubMed ] [ Google Scholar ] 24. van den Akker M., Buntinx F., Knottnerus J.A. Comorbidity or multimorbidity. Eur J Gen Pract. 1996;2(2):65–70. [ Google Scholar ] 25. Van den Akker M, Buntix F, Metsemakers JFM, Roos S, Knottnerus JA. Multimorbidity in general practice: prevalence, incidence, and determinants of co-occurring chronic and recurrent diseases. J Clin Epidemiol. 1998;51(5):367–375. doi: 10.1016/s0895-4356(97)00306-5. [ DOI ] [ PubMed ] [ Google Scholar ] 26. Navickas R, Petric V-K, Feigl AB, Seychell M. Multimorbidity: what do we know? What should we do? J Comorb. 2016;6(1):4–11. doi: 10.15256/joc.2016.6.72. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 27. Tseng Y-J, Chiu H-J, Chen CJ. Dxpr : an R package for generating analysis-ready data from electronic health records—diagnoses and procedures. PeerJ Comput Sci. 2021;7 doi: 10.7717/peerj-cs.520. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 28. Chen T., Guestrin C. Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining(KDD ’16), New York. 2016. XGBoost : A Scalable Tree Boosting System; pp. 785–794. [ Google Scholar ] 29. Ke G, Meng Q, Finley T, Wang T, Chen W, Ma W, et al. LightGBM: a highly efficient gradient boosting decision tree. Adv Neural Inf Process Syst. 2017:3147–3155. [ Google Scholar ] 30. Hosmer DW, Lemeshow S, Sturdivant RX. Applied Logistic Regression. 1st ed. Wiley; 2013. [ Google Scholar ] 31. Zou H., Hastie T. Regularization and variable selection via the elastic net. J R Stat Soc Series B Stat Methodol. 2005;67:301–320. [ Google Scholar ] 32. Feldman TC, Dienstag JL, Mandl KD, Tseng Y-J. Machine-learning-based predictions of direct-acting antiviral therapy duration for patients with hepatitis C. Int J Med Inform. 2021;154 doi: 10.1016/j.ijmedinf.2021.104562. [ DOI ] [ PubMed ] [ Google Scholar ] 33. Yu JR, Chen CH, Huang TW, Lu JJ, Chung CR. Lin TW,et al. Energy efficiency of inference algorithms for clinical laboratory data sets: green artificial intelligence study. J Med Internet Res. 2022;24(1) doi: 10.2196/28036. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 34. Tseng YJ, Wang YC, Hsueh PC, Wu CC. Development and validation of machine learning-based risk prediction models of oral squamous cell carcinoma using salivary autoantibody biomarkers. BMC Oral Health. 2022;22(1):534. doi: 10.1186/s12903-022-02607-2. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 35. Zhang J, Mucs D, Norinder U, Svensson F. LightGBM: an effective and scalable algorithm for prediction of chemical toxicity-application to the Tox21 and mutagenicity data sets. J Chem Inf Model. 2019;59(10):4150–4158. doi: 10.1021/acs.jcim.9b00633. [ DOI ] [ PubMed ] [ Google Scholar ] 36. Batten AJ, Thorpe J, Piegari RI, Rosland AM. A resampling based grid search method to improve reliability and robustness of mixture-item response theory models of multimorbid high-risk patients. IEEE J Biomed Health Inform. 2020;24(6):1780–1787. doi: 10.1109/JBHI.2019.2948734. [ DOI ] [ PubMed ] [ Google Scholar ] 37. Liu Y, Chen P-HC, Krause J, Peng L. How to read articles that use machine learning: users’ guides to the medical literature. JAMA. 2019;322(18):1806–1816. doi: 10.1001/jama.2019.16489. [ DOI ] [ PubMed ] [ Google Scholar ] 38. Uddin S, Khan A, Hossain ME, Moni MA. Comparing different supervised machine learning algorithms for disease prediction. BMC Med Inform Decis Mak. 2019;19(1):281. doi: 10.1186/s12911-019-1004-8. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 39. Wang H, Hung C, Chen C, Lee T, Huang K-Y, Ning H-C, et al. Increase Trichomonas vaginalis detection based on urine routine analysis through a machine learning approach. Sci Rep. 2019;9(1) doi: 10.1038/s41598-019-47361-8. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 40. World Health Organization. WHO guidelines for the treatment of treponema pallidum (Syphilis), https://www.who.int/publications/i/item/who-guidelines-for-the-treatment-of-treponema-pallidum-(syphilis) . [ PubMed ] 41. Heinze G, Ploner M, Beyea J. Confidence intervals after multiple imputation: combining profile likelihood information from logistic regressions. Stat Med. 2013;32(29):5062–5076. doi: 10.1002/sim.5899. [ DOI ] [ PubMed ] [ Google Scholar ] 42. Vandenbroucke JP. Strengthening the reporting of observational studies in Epidemiology (STROBE): explanation and elaboration. Ann Intern Med. 2007;147(8):W163–W194. doi: 10.7326/0003-4819-147-8-200710160-00010-w1. [ DOI ] [ PubMed ] [ Google Scholar ] 43. Pandey K, Fairley CK, Chen MY, Williamson DA, Bradshaw CS, Ong JJ, et al. Changes in the syphilis rapid plasma reagin titer between diagnosis and treatment. Clin Infect Dis. 2023;76(5):795–799. doi: 10.1093/cid/ciac843. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 44. Kerani RP, Handsfield HH, Stenger MS, Shafii T, Zick E, Brewer D, et al. Rising rates of syphilis in the era of syphilis elimination. Sex Transm Dis. 2007;34(3):154–161. doi: 10.1097/01.olq.0000233709.93891.e5. [ DOI ] [ PubMed ] [ Google Scholar ] 45. Huang S-Y, Hung J-H, Hu L-Y, Huang M-W, Lee S-C, Shen CC. Risk of sexually transmitted infections following depressive disorder. Medicine. 2018;97(43) doi: 10.1097/MD.0000000000012539. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 46. Lopes EBP, Filiberti A, Husain SA., Humphrey M.B. Immune contributions to osteoarthritis. Curr Osteoporos Rep. 2017;15(6):593–600. doi: 10.1007/s11914-017-0411-y. [ DOI ] [ PubMed ] [ Google Scholar ] 47. Camacho-Encina M, Balboa-Barreiro V, Rego-Perez I, Picchi F, Vanduin J, Qiu J, et al. Discovery of an autoantibody signature for the early diagnosis of knee osteoarthritis: data from the Osteoarthritis Initiative. Ann Rheum Dis. 2019;78(12):1699–1705. doi: 10.1136/annrheumdis-2019-215325. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 48. Savitskaya Y., Ramos L., Duarte C., Tellez R., Estrada E., Morales E., et al. Analysis of circulating immune complexes containing pain-associated molecules in patients with knee osteoarthritis: from the laboratory to the clinic. Osteoarthr Cartil. 2012;20:S257–S258. [ Google Scholar ] 49. Wauthier L, Plebani M, Favresse J. Interferences in immunoassays: review and practical algorithm. Clin Chem Lab Med. 2022;60(6):808–820. doi: 10.1515/cclm-2021-1288. [ DOI ] [ PubMed ] [ Google Scholar ] 50. Van de Velde S, Kunnamo I, Roshanov P, Kortteisto T, Aertgeerts B, Vandvik PO, et al. The GUIDES checklist: development of a tool to improve the successful use of guideline-based computerised clinical decision support. Implement Sci. 2018;13(1):86. doi: 10.1186/s13012-018-0772-3. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 51. Young H. Guidelines for serological testing for syphilis. Sex Transm Infect. 2000;76(5):403–405. doi: 10.1136/sti.76.5.403. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 52. Bosshard PP, Graf N, Knaute DF, Kündig T, Lautenschlager S, Weber R. Response of treponema pallidum particle agglutination test titers to treatment of syphilis. Clin Infect Dis. 2013;56(3):463–464. doi: 10.1093/cid/cis850. [ DOI ] [ PubMed ] [ Google Scholar ] 53. Chi C, Lee J, Tsai S, Chen W. Out‐of‐pocket payment for medical care under Taiwan’s National Health Insurance system. Health Econ. 2008;17(8):961–975. doi: 10.1002/hec.1312. [ DOI ] [ PubMed ] [ Google Scholar ] 54. Lin CC, Lai MS, Syu CY, Chang SC, Tseng FY. Accuracy of diabetes diagnosis in health insurance claims data in Taiwan. J Formos Med Assoc. 2005;104(3):157–163. [ PubMed ] [ Google Scholar ] 55. Lin JY, Kang EY, Yeh PH, Ling XC, Chen HC. Chen KJ,et al. Proposed measures to be taken by ophthalmologists during the coronavirus disease 2019 pandemic: experience from Chang Gung memorial hospital, Linkou, taiwan. Taiwan J Ophthalmol. 2021;12(1):115. doi: 10.4103/tjo.tjo_21_20. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Supplementary Materials Multimedia component 1 mmc1.docx (584.2KB, docx) Data Availability Statement All data used in this study are restricted under the law regarding the protection of patient information in accordance with local and institutional requirements. The application process for access to these data can be made available upon reasonable request to the corresponding author. Articles from Biomedical Journal are provided here courtesy of Chang Gung University ACTIONS View on publisher site PDF (2.5 MB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top