External Validation of a Web- and Artificial Intelligence-Based HIV/STI Risk Assessment Tool: Performance Evaluation Using Data from Sydney Sexual Health Centre

External Validation of a Web- and Artificial Intelligence-Based HIV/STI Risk Assessment Tool: Performance Evaluation Using Data from Sydney Sexual Health Centre | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article External Validation of a Web- and Artificial Intelligence-Based HIV/STI Risk Assessment Tool: Performance Evaluation Using Data from Sydney Sexual Health Centre Phyu M. Latt, Anik Ray, Heng Lu, Nyi N. Soe, Xianglong Xu, Yining Bao, and 5 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6757880/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 25 Nov, 2025 Read the published version in BMC Infectious Diseases → Version 1 posted 14 You are reading this latest preprint version Abstract Introduction HIV and sexually transmitted infections (STIs) continue to pose significant public health challenges globally. MySTIRisk , developed at Melbourne Sexual Health Centre (MSHC), is a machine learning-based tool that predicts individual risk for HIV, syphilis, gonorrhoea, and chlamydia using demographic and behavioural data. While initial validation showed promising results, external validation is crucial to assess its generalisability. This study externally validates MySTIRisk using data from the Sydney Sexual Health Centre (SSHC), Australia's second largest sexual health centre. Methods Following TRIPOD guidelines, we analysed consultations from patients aged 18 years and older attending SSHC between January 2013 and December 2023. Pre-trained MySTIRisk models were applied directly without modification. Performance was evaluated using the area under the receiver operating characteristic curve (AUC), sensitivity, and specificity at multiple thresholds, with subgroup analyses across demographic characteristics. Results We analysed 159,043 to 207,582 consultations at SSHC, with a median age of 30 years and 60.2–68.8% of the consultations involving men who have sex with men. The area under the receiver operating characteristic curve (AUC) values using data from SSHC were 0.67 (95% CI: 0.65–0.68) for HIV, 0.70 (95% CI: 0.69–0.71) for syphilis, 0.73 (95% CI: 0.73–0.74) for gonorrhoea, and 0.65 (95% CI: 0.65–0.66) for chlamydia, which were lower than the original MSHC validation metrics (0.74–0.87, all p < 0.001). Notably, model performance varied across demographic subgroups, with stronger HIV prediction among men who have sex with men with an AUC of 0.78 and better gonorrhoea prediction among younger attendees < 25 years with an AUC value of 0.79. At balanced sensitivity-specificity thresholds, the models identified 58.6–64.1% of infections while requiring testing of only 25.8–39.4% of the population. Conclusions Despite performance decrements in external validation using SSHC data, MySTIRisk maintained moderate to good predictive ability across all infections, demonstrating reasonable generalisability across different clinical populations. The demographic variations in performance highlight the importance of context-specific implementation and potential recalibration to optimise clinical utility. HIV sexually transmitted infections artificial intelligence risk assessment machine learning external validation predictive modelling sexual health digital health Figures Figure 1 Figure 2 Introduction Sexually transmitted infections (STIs) and HIV pose a significant public health challenge, with growing infection rates despite prevention efforts ( 1 – 3 ). Early detection is essential to reducing transmission and complications, yet many individuals delay testing due to limited awareness, barriers in healthcare access, and social stigma ( 4 – 6 ). Risk assessment tools can bridge this gap by encouraging timely testing and prevention. Existing risk prediction tools primarily focus on HIV and rely on conventional statistical methods, such as logistic regression ( 7 , 8 ). There remains an unmet need for tools that assess multiple STIs simultaneously while integrating more sophisticated analytical approaches. Machine learning offers potential benefits in this context, including the ability to handle complex, nonlinear, and multidimensional associations between risk factors and outcomes ( 9 – 11 ). To address this gap, the Melbourne Sexual Health Centre (MSHC) recently developed MySTIRisk , a machine learning-based tool for predicting individual risk of HIV, syphilis, gonorrhoea, and chlamydia ( 9 , 11 – 15 ). MySTIRisk uses demographic and behavioural data from the clinic attendees to generate personalised risk scores. Initial testing at MSHC showed promising results, with the area under the curve (AUC) values ranging from 0.70 to 0.84 across the four infections, indicating good discriminative ability ( 12 , 14 ). However, MySTIRisk was developed and validated using data from a single centre. To ensure its generalisability and wider applicability, external validation is necessary. Previous research has shown that machine learning models often exhibit performance variability when applied to new populations due to demographic and behavioural differences ( 16 – 21 ). Our recent systematic review of machine learning-based STI risk prediction tools revealed that only a small proportion of studies conducted external validation, with most limited to temporal validation within the same population ( 22 ). Without external validation, the effectiveness and reliability of MySTIRisk across diverse populations remain uncertain. This is particularly important for machine learning models, which can be sensitive to the specific characteristics of the training data ( 23 , 24 ). External validation helps assess whether the tool can be applied more broadly across Australian sexual health settings or requires adjustments for local implementation. This study aims to conduct the external validation of MySTIRisk using data from the Sydney Sexual Health Centre (SSHC). We selected SSHC for validation because it is the second largest sexual health centre in Australia, serving a diverse patient population with different demographics than MSHC, including higher overseas-born representation, and different sexual identity distributions. Together, MSHC and SSHC provide sexual health services to a substantial proportion of Australia's urban population attending a free publicly funded service. This study assessed MySTIRisk's predictive performance for all four STIs among the attendees at the SSHC. We compared AUCs, sensitivity, and specificity between MSHC and SSHC to evaluate the consistency of the tool's predictive accuracy across different settings. The findings from this study will inform decisions about implementing MySTIRisk in other Australian sexual health clinics and guide refinements for broader application. Validating this AI-based risk assessment tool across two major centres strengthens evidence for machine learning in public health and supports more effective STI prevention, resource allocation, and improved health outcomes. Methods Study Population and Design We conducted an external validation study of the MySTIRisk tool using data from SSHC following the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) statement guidelines ( 25 ). We used clinical consultation data from patients aged 18 years or older who attended SSHC between January 2013 and December 2023. Transgender individuals were excluded to maintain consistency with the original model development study, as the initial dataset did not include sufficient representation of this group. Data Collection and Cleaning At SSHC, we extracted data retrospectively from the electronic health record (EHR) of registered attendees, including patient demographics, sexual behaviours, recent STI contact and previous STI diagnoses, and HIV/STI test results. We performed systematic data cleaning to ensure compatibility with the original MySTIRisk model development approach. This included handling missing values, standardising categorical variables, and converting continuous variables to the appropriate format. Missing data were not imputed but treated as a separate category, consistent with the approach used in the original model development ( 9 , 12 ). We organised the data into four separate datasets corresponding to each infection endpoint (HIV, syphilis, gonorrhoea, and chlamydia). Each dataset included consultations where the respective infection was tested, accounting for differences in testing patterns across visits. Outcome Definitions We used the same outcome definitions as the original MySTIRisk development study ( 9 , 12 ). HIV infection was defined as a new diagnosis of HIV based on serology. Syphilis infection was defined as a new diagnosis of early syphilis (primary, secondary, or early latent using serological testing or nucleic acid amplification test (NAAT)). Gonorrhoea infection was defined as a new diagnosis using culture or NAAT at any anatomical site. Chlamydia infection was defined as a new diagnosis using NAAT at any anatomical site. All diagnoses were coded by clinicians following established clinical guidelines and standard laboratory procedures. Predictor Variables We used the same predictor variables identified in the original MySTIRisk tool development study ( 9 , 12 ). Key predictors included age, men who reported having sex with men (MSM), number of casual sexual partners (male and female) in the last 12 months, condom use with partners in the previous 12 months, sex overseas in the last 12 months, injecting drug use in the last 12 months, recent STI contact and previous STI diagnoses, and presence of STI symptoms. However, the "sex overseas" variable was not available in the SSHC dataset and was coded as missing for all records. For categorical variables, we used the same category definitions as the original study. For instance, condom use was categorised as "always," "never," "sometimes," or "not applicable", ensuring consistency with prior classifications. For numerical variables such as age and number of sexual partners, we maintained the same scaling and transformation approaches used in the original models. Risk Assessment Models The MySTIRisk comprises different machine learning models for each infection type. For HIV prediction, an ensemble approach combining elastic net regression, gradient boosting machine (GBM), and random forest models was used. For syphilis prediction, a GBM model was employed. Gonorrhoea risk was predicted using a random forest model, while chlamydia prediction utilised a GBM model ( 14 ). Full details of these model developments can be found in previously published studies ( 9 , 12 , 14 ). In brief, the development process used 5-fold cross-validation with data from the MSHC (2015–2018), with 80% used for training and 20% for validation. For our external validation, we applied these pre-trained models directly to the SSHC dataset without retraining or adjustments, following established external validation methodology ( 26 ). This approach ensures a robust assessment of model generalisability in a different clinical population without introducing modifications that might bias results. Statistical Analysis We calculated the area under the receiver operating characteristic curve (AUC) with bootstrapped 95% confidence intervals derived using 1,000 resamples to evaluate the model’s discrimination performance for each infection. This bootstrapping process provided robust estimates of variability in AUC values. We then compared these results with the performance metrics from the original MySTIRisk validation dataset (20% testing data from MSHC). To visually compare model performance, we overlaid receiver operating characteristic (ROC) curves from MSHC and SSHC validation datasets for each infection onto a single chart. We performed a Z-test to analyse differences between AUC values derived from the two independent datasets. This test relied on bootstrapped AUC distributions to calculate the Z-statistic and determine whether the differences were statistically significant. We considered a p-value of < 0.05 to indicate statistical significance. To evaluate the model's performance across clinically relevant risk thresholds, we calculated sensitivity and specificity at three key cutoffs for each infection: high sensitivity (targeting 90.0%), balanced sensitivity and specificity (determined using Youden’s index), and high specificity (targeting 90.0%). For each threshold, we also computed positive predictive value (PPV), negative predictive value (NPV), and the percentage of the population requiring testing. To calculate confidence intervals, we used the Wilson score method to estimate confidence intervals for proportion-based metrics in binary classification ( 27 ). Additionally, we conducted subgroup analyses to assess model performance consistency across key demographic characteristics. We calculated AUC values with 95% confidence intervals for population subgroups defined by population groups, age, and country of birth. For calibration assessment, we evaluated how well predicted probabilities aligned with observed outcomes. Following the original study's methodology, we divided predictions into 200 equally sized subgroups sorted by predicted probability and calculated the observed prevalence within each subgroup. We fitted logistic functions to these data points to create smooth calibration curves and visualise the relationship between predicted probabilities and observed prevalence. All statistical analyses were conducted using the Python programming language (version 3.9.12). Results Demographic Characteristics Between January 2013 and December 2023, we analysed 159,043 consultations for HIV, 168,443 for syphilis, and 207,582 for both gonorrhoea and chlamydia at the SSHC. Table 1 summarises the demographic and behavioural characteristics of these consultations. The median age was approximately 30 years across all datasets, with men who have sex with men (MSM) representing most of the consultations (60.2–68.8%). Most attendees (64.2–65.5%) were born overseas, and inconsistent condom use was reported in nearly half of all consultations. Table 1 Characteristics of Clinic Consultations in Individual Data Sets Predictors HIV (N = 159043 Consultations) Syphilis (N = 168443 Consultations) Gonorrhoea (N = 207582 Consultations) Chlamydia (N = 207582 Consultations) Age, median (IQR) 30 (25–37) 30 (25–37) 29 (25–36) 29 (25–36) Country of birth, n (%) Australia and New Zealand 56949 (35.8) 60336 (35.8) 71583 (34.5) 71583 (34.5) Overseas 102094 (64.2) 108107 (64.2) 135999 (65.5) 135999 (65.5) Missing STI Symptoms, n (%) Present 29953 (18.9) 32115 (19.1) 50575 (24.4) 50496 (24.3) Absent 129090 (81.2) 136328 (81.0) 157007 (75.6) 157086 (75.7) Population type, n (%) MSM 109362 (68.8) 115933 (68.8) 124963 (60.2) 124964 (60.2) Heterosexual male 23102 (14.5) 25516 (15.2) 37229 (18.0) 37228 (17.9) Female 26579 (16.7) 26994 (16.0) 45390 (21.9) 45390 (21.9) Condom use with male partners, n (%) Always 41447 (26.1) 43463 (25.8) 48936 (23.6) 48935 (23.6) Sometimes 76499 (48.1) 79712 (47.3) 100084 (48.2) 100095 (48.2) Never 27167 (17.1) 29221 (17.4) 38796 (18.7) 38795 (18.7) Not Applicable 10330 (6.5) 11510 (6.8) 13240 (6.4) 13235 (6.4) Unsure/Decline 46 (0.0) 50 (0.0) 74 (0.0) 74 (0.0) Missing 3554 (2.2) 4487 (2.7) 6452 (3.1) 6448 (3.1) Number of male sexual partners in last 12 months, median (IQR) 7 ( 3 – 15 ) 7 ( 3 – 15 ) 6 ( 3 – 15 ) 6 ( 3 – 15 ) Number of female sexual partners in last 12 months, median (IQR) 3 ( 1 – 7 ) 3 ( 1 – 6 ) 3 ( 1 – 7 ) 3 ( 1 – 7 ) Last time injected drugs not prescribed by doctor, n (%) Never 156433 (98.4) 165357 (98.2) 204079 (98.3) 204077 (98.3) Less than 3 months 1396 (0.9) 1720 (1.0) 1963 (1.0) 1962 (1.0) 3–12 months 577 (0.4) 709 (0.4) 789 (0.4) 789 (0.4) More than 12 months 637 (0.4) 657 (0.4) 751 (0.4) 754 (0.4) Past history of gonorrhoea, n (%) Yes 44512 (28.0) 49445 (29.4) 61432 (29.6) 61420 (29.6) No 114516 (72.0) 118982 (70.6) 146131 (70.4) 146143 (70.4) Missing 15 (0.0) 16 (0.0) 19 (0.0) 19 (0.0) Past history of nonspecific urethritis, n (%) Yes 16453 (10.4) 17972 (10.7) 24232 (11.7) 24211 (11.7) No 142575 (89.6) 150455 (89.3) 183331 (88.3) 183352 (88.3) Missing 15 (0.0) 16 (0.0) 19 (0.0) 19 (0.0) Past history of syphilis, n (%) Yes 19757 (12.4) 23550 (14.0) 28580 (13.8) 28557 (13.8) No 139271 (87.6) 144877 (86.0) 178983 (86.2) 179006 (86.2) Missing 15 (0.0) 16 (0.0) 19 (0.0) 19 (0.0) Sexual contact with someone diagnosed with gonorrhoea, n (%) Yes 3386 (2.1) 3635 (2.1) 5409 (2.6) 5381 (2.6) No 155657 (97.9) 164808 (97.9) 202173 (97.4) 202201 (97.4) Sexual contact with someone diagnosed with chlamydia, n (%) Yes 4657 (2.9) 4870 (2.9) 8578 (4.1) 8587 (4.1) No 154386 (97.1) 163573 (97.1) 199004 (95.9) 198995 (95.9) Sexual contact with someone diagnosed with syphilis, n (%) Yes 1426 (0.9) 1848 (1.1) 1573 (0.8) 1572 (0.7) No 157617 (99.1) 166595 (98.9) 206009 (99.2) 206010 (99.3) Infection positivity, n (%) Positive 1163 (0.7) 3410 (2.0) 15423 (7.4) 20801 (10.0) Negative 157880 (99.3) 165033 (98.0) 192159 (92.6) 186781 (90.0) Infection Positivity The HIV/STI positivity at SSHC were 0.7% for HIV, 2.0% for syphilis, 7.4% for gonorrhoea, and 10.0% for chlamydia (Table 1 ). Comparatively, MSHC reported lower positivity, with 0.3% for HIV, 1.7% for syphilis, 5.9% for gonorrhoea, and 8.1% for chlamydia ( 14 ), indicating potential variations in patient demographics, testing patterns, or underlying transmission dynamics between the two centres. External Validation Performance The external validation of MySTIRisk models at SSHC demonstrated lower discriminative ability than the original MSHC models across all four infections. The area under the ROC curve (AUC) values for SSHC were 0.67 (95% CI: 0.65–0.68) for HIV, 0.70 (95% CI: 0.69–0.71) for syphilis, 0.73 (95% CI: 0.73–0.74) for gonorrhoea, and 0.65 (95% CI: 0.65–0.66) for chlamydia. These results were significantly lower than the performance metrics at MSHC, which showed AUC values of 0.82 (95% CI: 0.80–0.85) for HIV, 0.87 (95% CI: 0.86–0.88) for syphilis, 0.84 (95% CI: 0.83–0.84) for gonorrhoea, and 0.74 (95% CI: 0.73–0.74) for chlamydia. Statistical comparison revealed significant differences in predictive performance across all four infections (all p-values < 0.001). Figure 1 displays the ROC curves for both centres across all four infections, illustrating consistent differences in model discrimination between these distinct clinical populations. Subgroup Analysis MySTIRisk demonstrated varying performance across demographic subgroups (Table 2 ). The HIV model showed better discrimination among MSM (AUC 0.78, 95% CI: 0.74–0.81) compared to heterosexual males (AUC 0.65, 95% CI: 0.63–0.68, p = 0.06) and females (AUC 0.61, 95% CI: 0.59–0.64, p = 0.002), with these differences reaching statistical significance. For gonorrhoea, performance was markedly lower in females (AUC 0.57, 95% CI: 0.55–0.59, p = 0.2) compared to MSM. Age-related variations were most notable for gonorrhoea, where younger attendees (< 25 years) exhibited significantly higher predictive performance (AUC 0.79, 95% CI: 0.78–0.80) than other age groups. The chlamydia model showed the most consistent performance across demographic categories (AUCs 0.64–0.67). Country of birth had minimal impact on model performance for all infections. Table 2 Subgroup Analyses of MySTIRisk Model Performance across Key Demographics Subgroup HIV Syphilis Gonorrhoea Chlamydia Overall 0.67 (0.65–0.68) 0.70 (0.69–0.71) 0.73 (0.73–0.74) 0.65 (0.65–0.66) Population type MSM (reference) 0.78 (0.74–0.81) 0.70 (0.68–0.71) 0.71 (0.70–0.71) 0.66 (0.65–0.66) Heterosexual male 0.65 (0.63–0.68) 0.63 (0.61–0.66)*** 0.72 (0.71–0.73) 0.66 (0.65–0.67) Female 0.61 (0.59–0.64)** 0.67 (0.62–0.71) 0.57 (0.55–0.59)*** 0.64 (0.63–0.65)** Age < 25 years 0.64 (0.62–0.66)*** 0.73 (0.70–0.75)*** 0.79 (0.78–0.80)*** 0.66 (0.65–0.67) 25–34 years (reference) 0.68 (0.64–0.72) 0.68 (0.66–0.69) 0.72 (0.72–0.73) 0.65 (0.65–0.66) ≥ 35 years 0.74 (0.68–0.80)* 0.70 (0.69–0.72)* 0.70 (0.69–0.70)*** 0.64 (0.63–0.65) Country of birth Australia and New Zealand (reference) 0.65 (0.62–0.69) 0.71 (0.69–0.72) 0.72 (0.71–0.73) 0.67 (0.66–0.67) Overseas 0.66 (0.65–0.68) 0.70 (0.69–0.71) 0.74 (0.73–0.74)*** 0.65 (0.64–0.65)*** *p < 0.05, **p < 0.01, ***p < 0.001 compared to reference group MSM: men who have sex with men; AUC: area under the receiver operating characteristic curve; CI: confidence interval Threshold Analysis Table 3 presents the performance of MySTIRisk across multiple risk thresholds for each infection. At thresholds calibrated for high sensitivity (90.0%), the proportion of the population requiring testing ranged from 70.3% (gonorrhoea) to 81.3% (chlamydia), with corresponding specificities between 19.7% and 31.3%. When optimising for balanced sensitivity and specificity, the models achieved more moderate but clinically useful performance. For HIV, a threshold of 0.62 resulted in 61.0% sensitivity (95% CI: 58.2–63.8%) and 64.1% specificity (95% CI: 63.8–64.3%), requiring testing of only 36.1% of the population. For syphilis, gonorrhoea, and chlamydia, the balanced thresholds achieved sensitivities of 58.6%, 64.1%, and 60.1%, respectively, with specificities ranging from 62.9–74.9%. At high specificity thresholds (90.0%), the models demonstrated lower sensitivities (23.5–37.0%) but substantially reduced the testing proportion to 10.1–12.0% of the population. Table 3 Performance of the MySTIRisk Models at Different Risk Thresholds Infections Scenario Threshold Sensitivity (95% CI) Specificity (95% CI) PPV (95% CI) NPV (95% CI) % Population Tested HIV Sensitivity at 90% 0.36 89.9% (88.1–91.5%) 27.2% (27.0-27.5%) 0.9% (0.8-1.0%) 99.7% (99.7–99.8%) 72.9% Balanced Sensitivity and Specificity* 0.62 61.0% (58.2–63.8%) 64.1% (63.8–64.3%) 1.2% (1.1–1.3%) 99.6% (99.5–99.6%) 36.1% Specificity at 90% 0.68 26.0% (23.5–28.6%) 90.0% (89.9–90.1%) 1.9% (1.7–2.1%) 99.4% (99.4–99.4%) 10.1% Syphilis Sensitivity at 90% 0.25 90.1% (89.0–91.0%) 25.6% (25.4–25.9%) 2.4% (2.4–2.5%) 99.2% (99.1–99.3%) 74.7% Balanced Sensitivity and Specificity* 0.46 58.6% (56.9–60.2%) 74.9% (74.7–75.1%) 4.6% (4.4–4.8%) 98.9% (98.8–98.9%) 25.8% Specificity at 90% 0.72 23.5% (22.1–25.0%) 90.1% (89.9–90.2%) 4.7% (4.4-5.0%) 98.3% (98.2–98.3%) 10.2% Gonorrhoea Sensitivity at 90% 0.33 90.0% (89.5–90.5%) 31.3% (31.1–31.5%) 9.5% (9.4–9.7%) 97.5% (97.4–97.6%) 70.3% Balanced Sensitivity and Specificity* 0.57 64.1% (63.3–64.9%) 70.3% (70.1–70.5%) 14.8% (14.5–15.0%) 96.1% (96.0-96.2%) 32.3% Specificity at 90% 0.66 37.0% (36.2–37.7%) 90.0% (89.9–90.1%) 22.9% (22.4–23.4%) 94.7% (94.6–94.8%) 12.0% Chlamydia Sensitivity at 90% 0.37 90.0% (89.6–90.4%) 19.7% (19.5–19.9%) 11.1% (10.9–11.2%) 94.6% (94.4–94.9%) 81.3% Balanced Sensitivity and Specificity* 0.52 60.1% (59.4–60.7%) 62.9% (62.7–63.1%) 15.3% (15.0-15.5%) 93.4% (93.3–93.5%) 39.4% Specificity at 90% 0.63 26.6% (26.0-27.2%) 90.0% (89.9–90.1%) 22.8% (22.3–23.4%) 91.7% (91.5–91.8%) 11.7% Sensitivity represents the percentage of true positive cases correctly identified by the model, while specificity measures the percentage of true negative cases correctly classified. The positive predictive value (PPV) represents the proportion of positive results that are truly positive cases, whereas the negative predictive value (NPV) reflects the proportion of negative results that are truly negative cases. The percentage of the population tested corresponds to the proportion of individuals who exceeded the given probability threshold. Calibration Assessment In our calibration assessment, we examined the association between predicted probabilities and observed prevalence for all four STIs (Fig. 2 ), as mentioned in the development study of MySTIRisk ( 12 ). The calibration plots demonstrated excellent agreement between predicted and observed risk across the probability spectrum for gonorrhoea and chlamydia, with data points closely following the logistic function fit. For HIV, the model showed good calibration at lower risk probabilities but slightly underestimated infection rates at higher probabilities (> 0.7), although this region contained fewer data points. The syphilis model demonstrated the most variation in calibration, with greater scatter around the fitted curve, particularly at mid-range probabilities (0.4–0.7). Discussion Our external validation of MySTIRisk demonstrated moderate to good predictive performance for all four STIs using the data from SSHC, although with significant decrements compared to the original MySTIRisk models developed at the MSHC. We found AUC values ranging from 0.65 for chlamydia and 0.73 for gonorrhoea, reflecting reasonable discriminative ability across infections, though lower than the original MSHC values (0.74–0.87). This performance decline aligns with established patterns in machine learning validation studies, where models typically show reduced effectiveness when applied to external populations due to differences in demographics, behavioural patterns, and testing practices ( 16 – 18 ). We observed notable population-specific variations, with the models performing better for HIV prediction among MSM (AUC 0.78) and for gonorrhoea among younger attendees (AUC 0.79). These findings underscore the importance of context-specific considerations in AI-based risk assessment tools. While MySTIRisk retains clinical utility across diverse settings, our results highlight the necessity of local validation and potential recalibration before widespread implementation, consistent with recent studies on clinical prediction models in infectious disease settings ( 10 , 16 , 18 , 20 , 28 ). Our study demonstrated that MySTIRisk models exhibited consistently lower discriminative performance when using the data from SSHC, with AUC values at 0.65–0.73 markedly reduced from those reported at MSHC (0.74–0.87). While such performance decrements are anticipated in external validation studies of machine learning models, several specific factors may explain these observed differences. First, the demographic and epidemiological composition differed substantially between sites—the SSHC dataset comprised a higher proportion of overseas-born attendees (approximately 65%) and MSM and lower proportions of heterosexual individuals than the more evenly distributed MSHC population. These population differences may influence how risk factors manifest and interact across settings. Second, differences in clinical documentation protocols may have affected model input quality; notably, while MSHC recorded sexual encounters overseas ("sex with someone outside Australia or New Zealand or had sex in Australia with someone from overseas"), SSHC did not capture this risk factor. Third, despite having fewer symptomatic presentations (18.8–24.4% vs. 28.3–35.7% at MSHC), SSHC paradoxically showed higher positivity across all infections, indicating population-specific testing and risk dynamics that models may struggle to capture. Studies have demonstrated that such disparities significantly impact machine learning model performance during external validation, particularly for models developed using comprehensive datasets ( 29 , 30 ). Fourth, the methods of triage, testing protocols, and patterns of symptomatic individuals may vary between the two populations. These differences may have led to differences in the patterns of risk factors for each infection between the MSHC and SSHC populations. Despite these differences, it is encouraging that MySTIRisk maintained moderate discriminative ability using the SSHC data, suggesting the core risk factors identified during development retain predictive value across Australian sexual health settings. Our subgroup analyses revealed important differences in MySTIRisk 's performance across demographic categories. For HIV prediction, the markedly superior performance among MSM (AUC 0.78) compared to heterosexual populations (AUC 0.61–0.65) suggested that the model captured MSM-specific risk patterns more effectively, potentially reflecting the higher prevalence and better-characterised transmission dynamics in this group. Age-stratified analysis demonstrated that younger attendees (< 25 years) showed significantly higher prediction accuracy for gonorrhoea (AUC 0.79), likely due to more consistent risk behaviours or testing patterns in this group. Interestingly, country of birth had minimal impact on model performance except for gonorrhoea and chlamydia. These demographic disparities in predictive performance align with findings from Franklin et al. ( 31 ), which demonstrated that demographic characteristics can introduce bias in machine learning models, affecting their generalisability across diverse populations. The observed variations in predictive performance across subgroups indicate that risk factors have different predictive weights in diverse populations, reflecting the complex interplay of behavioural, social, and biological determinants of STI transmissions. The threshold analysis findings provided valuable guidance for implementing MySTIRisk in diverse clinical settings. Setting risk thresholds to achieve high sensitivity (90.0%) would capture most infections but require testing 70.3–81.3% of attendees, which may place a significant strain on resources without substantially improving efficiency compared to universal testing. Conversely, high specificity thresholds (90.0%) would significantly reduce testing volume to only 10.1–12.0% of the population but would result in missing 63.0–76.5% of infections, potentially limiting early detection efforts. A balanced approach using Youden's index thresholds could represent an optimal middle ground by identifying 58.6–64.1% of infections while reducing the testing proportion to 25.8–39.4%. These results demonstrate how risk prediction tools can be calibrated according to local resource constraints, such as limited test kits, staffing shortages, or clinic capacity, while addressing public health priorities. In resource-limited settings, MySTIRisk could help prioritise testing for individuals at highest risk. Meanwhile, in well-resourced environments, it could complement universal testing by identifying candidates for more comprehensive screening. The flexibility to adjust thresholds based on specific clinical contexts represents a key advantage of machine learning-based risk assessment tools compared to traditional screening questionnaires, which rely on fixed criteria and may lack adaptability to evolving epidemiological trends. Future research should address several key areas to enhance the clinical utility of MySTIRisk and similar AI-based prediction tools. First, transfer learning approaches could be explored to adapt models to specific clinical settings while preserving their core predictive capabilities. This would involve fine-tuning the existing models with local data rather than complete retraining, potentially improving performance while maintaining generalisability ( 32 ). Second, future iterations should expand demographic representation, particularly including trans and gender diverse individuals who were excluded from current analyses, as there were no specific sexual behavioural questions for this population at the time when MySTIRisk was developed. Third, prospective implementation studies are needed to evaluate the real-world impact of integrating MySTIRisk into clinical workflows, including assessment of resource utilisation, infection detection rates, and user acceptance. Additional research could explore the temporal stability of these predictions, as sexual behaviour patterns and infection dynamics evolve over time. Finally, comparative effectiveness research should evaluate whether AI-based risk stratification outperforms traditional clinician judgement or simpler risk assessment tools in terms of both accuracy and cost-effectiveness. It would strengthen the evidence base for the broader adoption of machine learning approaches in sexual health services and inform best practices for their implementation across diverse healthcare settings. Our study has several notable strengths. This is the first external validation of an AI-based HIV/STI risk assessment tool across Australia's two largest publicly funded sexual health centres, representing a significant proportion of the country's urban sexual health services and providing robust statistical power through its extensive sample size. Our adherence to TRIPOD guidelines for prediction model validation ensures methodological quality and transparent reporting. The comprehensive assessment across four key STIs offers valuable comparative insights into prediction patterns across different infections. Additionally, our analysis of multiple risk thresholds provides practical implementation guidance that can be tailored to various clinical scenarios. By validating MySTIRisk in a demographically distinct setting, we have tested the model's resilience to population heterogeneity, a critical consideration for AI tools intended for widespread implementation across diverse healthcare environments. Our study has several important limitations that warrant consideration when interpreting the results. First, the retrospective nature of our analyses may have introduced selection bias though our large sample size spanning a decade helps mitigate some temporal variation effects. Second, we excluded transgender individuals to maintain consistency with the original model development, representing a significant gap in generalisability that future iterations must address; this exclusion was intentional because data systems did not capture epidemiological risk on these individuals until recently. With time adequate epidemiological risk data will accumulate and allow this analysis in the future. Third, differences in clinical documentation between sites meant that some risk factors used in the original model development were not identically captured at SSHC, potentially affecting model performance. Although we attempted to standardise variables where possible, some differences were unavoidable due to inherent variations in clinical workflows between centres. Fourth, while the study included a diverse patient population, both centres are urban sexual health clinics, potentially limiting generalisability to rural settings or primary care contexts where STI testing also occurs. However, these two centres serve the largest metropolitan areas in Australia, providing substantial population coverage. Fifth, we did not assess the models' performance for predicting multiple concurrent infections, which represents a clinically important scenario, as the original model development focused on individual infections. Sixth, the PPV were relatively low, particularly for HIV (0.9–1.9%), which should be interpreted in the context of the low prevalence of these infections in the study population. This reflects a common challenge in predictive modelling for low-prevalence conditions, where even highly accurate models may yield modest PPV values. Finally, while we did not directly evaluate patients' perspectives on AI-based risk assessment in this validation study, subsequent qualitative research conducted by our team has addressed these dimensions, revealing both the tool's potential value for overcoming barriers to sexual health information and areas for improving inclusivity and guidance ( 33 ). In conclusion, our external validation of MySTIRisk demonstrated moderate to good predictive performance across all four STIs, though with decreased discrimination compared to the original development site. Despite this performance drop, the tool maintained discriminative ability across different clinical populations, supporting its potential application in diverse sexual health settings. Population-specific variations highlight the need for contextualised implementation, with adjustable thresholds offering flexibility based on local resources and priorities. While improvements are needed, this first external validation across Australia's two largest sexual health centres provides evidence supporting the broader application of AI-based risk assessment in sexual health, though local validation remains essential before widespread implementation. Abbreviations AUC Area under the curve CI Confidence Interval EHR Electronic health record GBM Gradient boosting machine IQR Interquartile Range MSHC Melbourne Sexual Health Centre MSM Men who have sex with men NAAT Nucleic acid amplification test NHMRC National Health and Medical Research Council NPV Negative predictive value PPV Positive predictive value ROC Receiver operating characteristic SESLHD South Eastern Sydney Local Health District SSHC Sydney Sexual Health Centre STI Sexually transmitted infections TRIPOD Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis Declarations Ethics Approval and Consent to Participate This multi-site study received ethical approval from the Alfred Hospital Ethics Committee (Project 145/24) under the Human Research Ethics Application (HREA 105599). Site-specific approvals were obtained from both the South Eastern Sydney Local Health District (SESLHD) and Alfred Health (Victoria). The research was conducted in accordance with the ethical principles outlined in the Declaration of Helsinki. As this was a retrospective study using de-identified data, the Alfred Hospital Ethics Committee granted a consent waiver for the collection, use and disclosure of participants' health and personal information in accordance with the Victorian Office of the Health Services Commissioner's Statutory Guidelines on Research, the Health Records and Information Privacy Act 2002 (NSW): Statutory Guidelines on Research, and the NHMRC Guidelines approved under Sections 95 & 95A of the Privacy Act 1988, ensuring compliance with ethical standards. Acknowledgements The authors sincerely thank Monash University for providing a PhD scholarship for PL. We also extend our thanks to all contributors who played a role in this study. Author Contributions CKF, LZ and PL conceptualised and designed the study methodology. CKF provided overall supervision throughout all phases of the research. PL coordinated the project implementation, secured ethics approval, conducted the statistical analyses, and drafted the initial manuscript. AR, HL and RV were responsible for data extraction, validation, and cleaning processes. The development and technical implementation of the MySTIRisk web application was a collaborative effort involving PL, XX, YB, CKF, LZ, and EPFC. CKF, EPFC and JO provided statistical expertise and methodological guidance. All authors contributed to the manuscript's revision and read and approved the submitted version. Funding CKF is supported by a National Health and Medical Research Council (NHMRC) Leadership Investigator Grant (GNT1172900). JO is supported by the NHMRC Emerging Leadership Investigator Grant (GNT1193955). EPFC is supported by an NHMRC Leadership Investigator Grant (GNT2033299). Potential Conflicts of Interest PL, NS, XX, YB, LZ and CKF have licensed the use of MySTIRisk to Helfie (Level 5/171 La Trobe St, Melbourne VIC 3000). The remaining authors declare no conflicts of interest. Availability of Data and Materials The data supporting this study's findings are not openly available due to sensitivity and are available from the corresponding author, Dr. Phyu Mon Latt, at [email protected] , upon reasonable request. Data are in controlled access data storage at the Melbourne Sexual Health Centre. Consent for Publication Not applicable Clinical Trial Number Not applicable References Nattabi B, Matthews V, Bailie J, Rumbold A, Scrimgeour D, Schierhout G, et al. Wide variation in sexually transmitted infection testing and counselling at Aboriginal primary health care centres in Australia: analysis of longitudinal continuous quality improvement data. BMC Infect Dis. 2017;17(1):148. Rogers B, Tao J, Murphy M, Chan PA. The COVID-19 Pandemic and Sexually Transmitted Infections: Where Do We Go From Here? Sex Transm Dis. 2021;48(7):e94-e6. World Health Organization. Sexually transmitted infections (STIs) fact sheets Geneva20 July 2023 [Available from: https://www.who.int/news-room/fact-sheets/detail/sexually-transmitted-infections-(stis)/]. Okal J, Lango D, Matheka J, Obare F, Ngunu-Gituathi C, Mugambi M, et al. "It is always better for a man to know his HIV status" - A qualitative study exploring the context, barriers and facilitators of HIV testing among men in Nairobi, Kenya. PLoS One. 2020;15(4):e0231645. Khan AR, Altalbe A. Potential impacts of Russo-Ukraine conflict and its psychological consequences among Ukrainian adults: the post-COVID-19 era. Front Public Health. 2023;11:1280423. Farquharson RM, Fairley CK, Abraham E, Bradshaw CS, Plummer EL, Ong JJ, et al. Time to healthcare seeking following the onset of symptoms among men and women attending a sexual health clinic in Melbourne, Australia. Front Med (Lausanne). 2022;9:915399. Scott H, Vittinghoff E, Irvin R, Liu A, Nelson L, Del Rio C, et al. Development and Validation of the Personalized Sexual Health Promotion (SexPro) HIV Risk Prediction Model for Men Who Have Sex with Men in the United States. AIDS Behav. 2020;24(1):274-83. Hoenigl M, Weibel N, Mehta SR, Anderson CM, Jenks J, Green N, et al. Development and validation of the San Diego Early Test Score to predict acute and early HIV infection risk in men who have sex with men. Clin Infect Dis. 2015;61(3):468-75. Bao Y, Medland NA, Fairley CK, Wu J, Shang X, Chow EPF, et al. Predicting the diagnosis of HIV and sexually transmitted infections among men who have sex with men using machine learning approaches. J Infect. 2021;82(1):48-59. Balzer LB, Havlir DV, Kamya MR, Chamie G, Charlebois ED, Clark TD, et al. Machine Learning to Identify Persons at High-Risk of Human Immunodeficiency Virus Acquisition in Rural Kenya and Uganda. Clinical Infectious Diseases. 2020;71(9):2326-33. Phyu Mon L, Nyi Nyi S, Xianglong X, Rashidur R, Eric PFC, Jason JO, et al. Assessing disparity in the distribution of HIV and sexually transmitted infections in Australia: a retrospective cross-sectional study using Gini coefficients. BMJ Public Health. 2023;1(1):e000012. Xu X, Yu Z, Ge Z, Chow EPF, Bao Y, Ong JJ, et al. Web-Based Risk Prediction Tool for an Individual's Risk of HIV and Sexually Transmitted Infections Using Machine Learning Algorithms: Development and External Validation Study. J Med Internet Res. 2022;24(8):e37850. Xu X, Ge Z, Chow EPF, Yu Z, Lee D, Wu J, et al. A Machine-Learning-Based Risk-Prediction Tool for HIV and Sexually Transmitted Infections Acquisition over the Next 12 Months. J Clin Med. 2022;11(7). Latt PM, Soe NN, Xu X, Ong JJ, Chow EPF, Fairley CK, et al. Identifying individuals at high risk for HIV and sexually transmitted infections with an artificial intelligence-based risk assessment tool. Open Forum Infectious Diseases. 2024. Melbourne Sexual Health Centre. MySTIRisk [Available from: https://mystirisk.mshc.org.au/]. Gruber S, Krakower D, Menchaca JT, Hsu K, Hawrusik R, Maro JC, et al. Using electronic health records to identify candidates for human immunodeficiency virus pre-exposure prophylaxis: An application of super learning to risk prediction when the outcome is rare. Statistics in Medicine. 2020;39(23):3059-73. He J, Li J, Jiang S, Cheng W, Jiang J, Xu Y, et al. Application of machine learning algorithms in predicting HIV infection among men who have sex with men: Model development and validation. Frontiers in Public Health. 2022;10. Krakower DS, Gruber S, Hsu K, Menchaca JT, Maro JC, Kruskal BA, et al. Development and validation of an automated HIV prediction algorithm to identify candidates for pre-exposure prophylaxis: a modelling study. The Lancet HIV. 2019;6(10):e696-e704. Majam M, Segal B, Fieggen J, Smith E, Hermans L, Singh L, et al. Utility of a machine-guided tool for assessing risk behaviour associated with contracting HIV in three sites in South Africa. Informatics in Medicine Unlocked. 2023;37. Marcus JL, Hurley LB, Krakower DS, Alexeeff S, Silverberg MJ, Volk JE. Use of electronic health record data and machine learning to identify candidates for HIV pre-exposure prophylaxis: a modelling study. The Lancet HIV. 2019;6(10):e688-e95. Xu X, Yu Z, Ge Z, Chow EPF, Bao Y, Ong JJ, et al. Web-Based Risk Prediction Tool for an Individual's Risk of HIV and Sexually Transmitted Infections Using Machine Learning Algorithms: Development and External Validation Study. Journal of Medical Internet Research. 2022;24(8). Latt PM, Soe NN, Fairley CK, Chow EPF, Johnson CC, Shah P, et al. Machine Learning for Personalised Risk Assessment of HIV, Syphilis, Gonorrhoea, and Chlamydia: A Systematic Review and Meta-analysis. International Journal of Infectious Diseases. 2025:107922. Gu J, Epland M, Ma X, Park J, Sanchez RJ, Li Y. A machine-learning algorithm using claims data to identify patients with homozygous familial hypercholesterolemia. Scientific Reports. 2024;14(1):8890. Alelyani S. Detection and Evaluation of Machine Learning Bias. Applied Sciences. 2021;11(14):6271. Collins GS, Moons KGM, Dhiman P, Riley RD, Beam AL, Van Calster B, et al. TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. Bmj. 2024;385:e078378. Riley RD, Archer L, Snell KIE, Ensor J, Dhiman P, Martin GP, et al. Evaluation of clinical prediction models (part 2): how to undertake an external validation study. BMJ. 2024;384:e074820. Brown LD, Cai TT, DasGupta A. Interval estimation for a binomial proportion. Statistical science. 2001;16(2):101-33. Latt PM, Soe NN, Xu X, Ong JJ, Chow EPF, Fairley CK, et al. Identifying Individuals at High Risk for HIV and Sexually Transmitted Infections With an Artificial Intelligence-Based Risk Assessment Tool. Open Forum Infectious Diseases. 2024;11(3). Chen X, Hu L, Yu R. Development and external validation of machine learning-based models to predict patients with cellulitis developing sepsis during hospitalisation. BMJ Open. 2024;14(7):e084183. Rios R, Miller RJH, Manral N, Sharir T, Einstein AJ, Fish MB, et al. Handling missing values in machine learning to predict patient-specific risk of adverse cardiac events: Insights from REFINE SPECT registry. Comput Biol Med. 2022;145:105449. Franklin G, Stephens R, Piracha M, Tiosano S, Lehouillier F, Koppel R, et al. The Sociodemographic Biases in Machine Learning Algorithms: A Biomedical Informatics Perspective. Life (Basel). 2024;14(6). Ebbehoj A, Thunbo M, Andersen OE, Glindtvad MV, Hulman A. Transfer learning for non-image data in clinical research: A scoping review. PLOS Digit Health. 2022;1(2):e0000014. King AJ, Latt PM, Soe NN, Temple-Smith M, Fairley CK, Chow EP, et al. User experiences of an AI application for predicting risk of sexually transmitted infections. Digit Health. 2024;10:20552076241289646. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 25 Nov, 2025 Read the published version in BMC Infectious Diseases → Version 1 posted Editorial decision: Revision requested 15 Sep, 2025 Reviews received at journal 30 Jul, 2025 Reviews received at journal 25 Jul, 2025 Reviews received at journal 22 Jul, 2025 Reviewers agreed at journal 16 Jul, 2025 Reviewers agreed at journal 15 Jul, 2025 Reviewers agreed at journal 15 Jul, 2025 Reviews received at journal 23 Jun, 2025 Reviewers agreed at journal 10 Jun, 2025 Reviewers agreed at journal 05 Jun, 2025 Reviewers invited by journal 01 Jun, 2025 Editor assigned by journal 30 May, 2025 Submission checks completed at journal 30 May, 2025 First submitted to journal 30 May, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6757880","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":465206866,"identity":"9d4c18fb-6d12-4136-bbbc-18273ed0be53","order_by":0,"name":"Phyu M. Latt","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA20lEQVRIiWNgGAWjYDACCRBRAMTsDWB+AhAbEKHFgEGCh+cAyVokEojUYi7dYybBYGBXZy/5OvFxQU1dHgN78zYJfFos55wBaUmW4JHO3Ww849jhYgaeY2V4tRjcyAFpYQZp2SbNw3YgsUECJEJYS70Ej+RZoJZ/dYkN8m+I0nIY6H3ebdK8bcxAW3jwa7GckVZskWBwXLLnDNAvvH2HE9t4gCL4tJhLJG+88aGimp+9/ezGxzzf6hL72Q9vvIHXYQwcBuC4gAM2fMohWtgfEFIzCkbBKBgFIx0AALFXP5O7bGffAAAAAElFTkSuQmCC","orcid":"","institution":"Alfred Health","correspondingAuthor":true,"prefix":"","firstName":"Phyu","middleName":"M.","lastName":"Latt","suffix":""},{"id":465206867,"identity":"72636da1-9044-4747-8d8c-54d1fb890790","order_by":1,"name":"Anik Ray","email":"","orcid":"","institution":"South Eastern Sydney Local Health District","correspondingAuthor":false,"prefix":"","firstName":"Anik","middleName":"","lastName":"Ray","suffix":""},{"id":465206868,"identity":"5bc63300-ddff-45c7-a302-22cd94e16aff","order_by":2,"name":"Heng Lu","email":"","orcid":"","institution":"South Eastern Sydney Local Health District","correspondingAuthor":false,"prefix":"","firstName":"Heng","middleName":"","lastName":"Lu","suffix":""},{"id":465206869,"identity":"fd3a35bd-b886-4211-ae2c-bfc3517bbfb7","order_by":3,"name":"Nyi N. Soe","email":"","orcid":"","institution":"Alfred Health","correspondingAuthor":false,"prefix":"","firstName":"Nyi","middleName":"N.","lastName":"Soe","suffix":""},{"id":465206870,"identity":"752b7943-956f-4464-b2b2-30919dc5459a","order_by":4,"name":"Xianglong Xu","email":"","orcid":"","institution":"Alfred Health","correspondingAuthor":false,"prefix":"","firstName":"Xianglong","middleName":"","lastName":"Xu","suffix":""},{"id":465206871,"identity":"3dce5838-2a3a-43d6-a597-c8c9d521bfa0","order_by":5,"name":"Yining Bao","email":"","orcid":"","institution":"Alfred Health","correspondingAuthor":false,"prefix":"","firstName":"Yining","middleName":"","lastName":"Bao","suffix":""},{"id":465206872,"identity":"a37aa340-5b0b-44bd-a458-cb7f01b0e2c8","order_by":6,"name":"Jason J Ong","email":"","orcid":"","institution":"Monash University","correspondingAuthor":false,"prefix":"","firstName":"Jason","middleName":"J","lastName":"Ong","suffix":""},{"id":465206873,"identity":"48cc3831-e329-43ef-82ff-22cbb99f6ef3","order_by":7,"name":"Eric P F Chow","email":"","orcid":"","institution":"Monash University","correspondingAuthor":false,"prefix":"","firstName":"Eric","middleName":"P F","lastName":"Chow","suffix":""},{"id":465206874,"identity":"bfc37f99-164e-49e7-ac86-1cb8f3becfd1","order_by":8,"name":"Rick Varma","email":"","orcid":"","institution":"South Eastern Sydney Local Health District","correspondingAuthor":false,"prefix":"","firstName":"Rick","middleName":"","lastName":"Varma","suffix":""},{"id":465206875,"identity":"70a8ba45-f0e5-42dd-8e6b-73e963305600","order_by":9,"name":"Lei Zhang","email":"","orcid":"","institution":"Alfred Health","correspondingAuthor":false,"prefix":"","firstName":"Lei","middleName":"","lastName":"Zhang","suffix":""},{"id":465206876,"identity":"a686a365-ad69-42ee-afbf-1c938a53e576","order_by":10,"name":"Christopher K. Fairley","email":"","orcid":"","institution":"Monash University","correspondingAuthor":false,"prefix":"","firstName":"Christopher","middleName":"K.","lastName":"Fairley","suffix":""}],"badges":[],"createdAt":"2025-05-27 09:38:20","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6757880/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6757880/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1186/s12879-025-12087-8","type":"published","date":"2025-11-25T15:57:22+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":84185686,"identity":"7500ba66-e99e-452b-994c-e6d5163ce5e8","added_by":"auto","created_at":"2025-06-09 05:28:54","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":192139,"visible":true,"origin":"","legend":"\u003cp\u003eExternal Validation of MySTIRisk: Model Performance across Two Sexual Health Centres\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-6757880/v1/760e75ff01561fb4279c0741.png"},{"id":84185687,"identity":"599904fa-8368-43da-9364-4544aba9dc7f","added_by":"auto","created_at":"2025-06-09 05:28:54","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":150083,"visible":true,"origin":"","legend":"\u003cp\u003eCalibration Plots of Model-Predicted Probabilities Versus Observed Prevalence for MySTIRisk models on SSHC data\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-6757880/v1/26fe02a9c11622ddcc52daa8.png"},{"id":97178601,"identity":"0faecf22-c00e-425c-937f-cef60c65caa9","added_by":"auto","created_at":"2025-12-01 16:11:24","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1614618,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6757880/v1/839b6fec-6b3f-40dd-9ce2-0971416aa579.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"External Validation of a Web- and Artificial Intelligence-Based HIV/STI Risk Assessment Tool: Performance Evaluation Using Data from Sydney Sexual Health Centre","fulltext":[{"header":"Introduction","content":"\u003cp\u003eSexually transmitted infections (STIs) and HIV pose a significant public health challenge, with growing infection rates despite prevention efforts (\u003cspan additionalcitationids=\"CR2\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e). Early detection is essential to reducing transmission and complications, yet many individuals delay testing due to limited awareness, barriers in healthcare access, and social stigma (\u003cspan additionalcitationids=\"CR5\" citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e). Risk assessment tools can bridge this gap by encouraging timely testing and prevention.\u003c/p\u003e \u003cp\u003eExisting risk prediction tools primarily focus on HIV and rely on conventional statistical methods, such as logistic regression (\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e). There remains an unmet need for tools that assess multiple STIs simultaneously while integrating more sophisticated analytical approaches. Machine learning offers potential benefits in this context, including the ability to handle complex, nonlinear, and multidimensional associations between risk factors and outcomes (\u003cspan additionalcitationids=\"CR10\" citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eTo address this gap, the Melbourne Sexual Health Centre (MSHC) recently developed \u003cem\u003eMySTIRisk\u003c/em\u003e, a machine learning-based tool for predicting individual risk of HIV, syphilis, gonorrhoea, and chlamydia (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan additionalcitationids=\"CR12 CR13 CR14\" citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e). \u003cem\u003eMySTIRisk\u003c/em\u003e uses demographic and behavioural data from the clinic attendees to generate personalised risk scores. Initial testing at MSHC showed promising results, with the area under the curve (AUC) values ranging from 0.70 to 0.84 across the four infections, indicating good discriminative ability (\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eHowever, \u003cem\u003eMySTIRisk\u003c/em\u003e was developed and validated using data from a single centre. To ensure its generalisability and wider applicability, external validation is necessary. Previous research has shown that machine learning models often exhibit performance variability when applied to new populations due to demographic and behavioural differences (\u003cspan additionalcitationids=\"CR17 CR18 CR19 CR20\" citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e). Our recent systematic review of machine learning-based STI risk prediction tools revealed that only a small proportion of studies conducted external validation, with most limited to temporal validation within the same population (\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e). Without external validation, the effectiveness and reliability of \u003cem\u003eMySTIRisk\u003c/em\u003e across diverse populations remain uncertain. This is particularly important for machine learning models, which can be sensitive to the specific characteristics of the training data (\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e). External validation helps assess whether the tool can be applied more broadly across Australian sexual health settings or requires adjustments for local implementation.\u003c/p\u003e \u003cp\u003eThis study aims to conduct the external validation of \u003cem\u003eMySTIRisk\u003c/em\u003e using data from the Sydney Sexual Health Centre (SSHC). We selected SSHC for validation because it is the second largest sexual health centre in Australia, serving a diverse patient population with different demographics than MSHC, including higher overseas-born representation, and different sexual identity distributions. Together, MSHC and SSHC provide sexual health services to a substantial proportion of Australia's urban population attending a free publicly funded service. This study assessed \u003cem\u003eMySTIRisk's\u003c/em\u003e predictive performance for all four STIs among the attendees at the SSHC. We compared AUCs, sensitivity, and specificity between MSHC and SSHC to evaluate the consistency of the tool's predictive accuracy across different settings.\u003c/p\u003e \u003cp\u003eThe findings from this study will inform decisions about implementing \u003cem\u003eMySTIRisk\u003c/em\u003e in other Australian sexual health clinics and guide refinements for broader application. Validating this AI-based risk assessment tool across two major centres strengthens evidence for machine learning in public health and supports more effective STI prevention, resource allocation, and improved health outcomes.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003eStudy Population and Design\u003c/p\u003e \u003cp\u003eWe conducted an external validation study of the \u003cem\u003eMySTIRisk\u003c/em\u003e tool using data from SSHC following the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) statement guidelines (\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e). We used clinical consultation data from patients aged 18 years or older who attended SSHC between January 2013 and December 2023. Transgender individuals were excluded to maintain consistency with the original model development study, as the initial dataset did not include sufficient representation of this group.\u003c/p\u003e \u003cp\u003eData Collection and Cleaning\u003c/p\u003e \u003cp\u003eAt SSHC, we extracted data retrospectively from the electronic health record (EHR) of registered attendees, including patient demographics, sexual behaviours, recent STI contact and previous STI diagnoses, and HIV/STI test results.\u003c/p\u003e \u003cp\u003eWe performed systematic data cleaning to ensure compatibility with the original \u003cem\u003eMySTIRisk\u003c/em\u003e model development approach. This included handling missing values, standardising categorical variables, and converting continuous variables to the appropriate format. Missing data were not imputed but treated as a separate category, consistent with the approach used in the original model development (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e). We organised the data into four separate datasets corresponding to each infection endpoint (HIV, syphilis, gonorrhoea, and chlamydia). Each dataset included consultations where the respective infection was tested, accounting for differences in testing patterns across visits.\u003c/p\u003e \u003cp\u003eOutcome Definitions\u003c/p\u003e \u003cp\u003eWe used the same outcome definitions as the original \u003cem\u003eMySTIRisk\u003c/em\u003e development study (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e). HIV infection was defined as a new diagnosis of HIV based on serology. Syphilis infection was defined as a new diagnosis of early syphilis (primary, secondary, or early latent using serological testing or nucleic acid amplification test (NAAT)). Gonorrhoea infection was defined as a new diagnosis using culture or NAAT at any anatomical site. Chlamydia infection was defined as a new diagnosis using NAAT at any anatomical site. All diagnoses were coded by clinicians following established clinical guidelines and standard laboratory procedures.\u003c/p\u003e \u003cp\u003ePredictor Variables\u003c/p\u003e \u003cp\u003eWe used the same predictor variables identified in the original \u003cem\u003eMySTIRisk\u003c/em\u003e tool development study (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e). Key predictors included age, men who reported having sex with men (MSM), number of casual sexual partners (male and female) in the last 12 months, condom use with partners in the previous 12 months, sex overseas in the last 12 months, injecting drug use in the last 12 months, recent STI contact and previous STI diagnoses, and presence of STI symptoms. However, the \"sex overseas\" variable was not available in the SSHC dataset and was coded as missing for all records.\u003c/p\u003e \u003cp\u003eFor categorical variables, we used the same category definitions as the original study. For instance, condom use was categorised as \"always,\" \"never,\" \"sometimes,\" or \"not applicable\", ensuring consistency with prior classifications. For numerical variables such as age and number of sexual partners, we maintained the same scaling and transformation approaches used in the original models.\u003c/p\u003e \u003cp\u003eRisk Assessment Models\u003c/p\u003e \u003cp\u003eThe \u003cem\u003eMySTIRisk\u003c/em\u003e comprises different machine learning models for each infection type. For HIV prediction, an ensemble approach combining elastic net regression, gradient boosting machine (GBM), and random forest models was used. For syphilis prediction, a GBM model was employed. Gonorrhoea risk was predicted using a random forest model, while chlamydia prediction utilised a GBM model (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e). Full details of these model developments can be found in previously published studies (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e). In brief, the development process used 5-fold cross-validation with data from the MSHC (2015\u0026ndash;2018), with 80% used for training and 20% for validation.\u003c/p\u003e \u003cp\u003eFor our external validation, we applied these pre-trained models directly to the SSHC dataset without retraining or adjustments, following established external validation methodology (\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e). This approach ensures a robust assessment of model generalisability in a different clinical population without introducing modifications that might bias results.\u003c/p\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eStatistical Analysis\u003c/h2\u003e \u003cp\u003eWe calculated the area under the receiver operating characteristic curve (AUC) with bootstrapped 95% confidence intervals derived using 1,000 resamples to evaluate the model\u0026rsquo;s discrimination performance for each infection. This bootstrapping process provided robust estimates of variability in AUC values. We then compared these results with the performance metrics from the original \u003cem\u003eMySTIRisk\u003c/em\u003e validation dataset (20% testing data from MSHC).\u003c/p\u003e \u003cp\u003eTo visually compare model performance, we overlaid receiver operating characteristic (ROC) curves from MSHC and SSHC validation datasets for each infection onto a single chart. We performed a Z-test to analyse differences between AUC values derived from the two independent datasets. This test relied on bootstrapped AUC distributions to calculate the Z-statistic and determine whether the differences were statistically significant. We considered a p-value of \u0026lt;\u0026thinsp;0.05 to indicate statistical significance.\u003c/p\u003e \u003cp\u003eTo evaluate the model's performance across clinically relevant risk thresholds, we calculated sensitivity and specificity at three key cutoffs for each infection: high sensitivity (targeting 90.0%), balanced sensitivity and specificity (determined using Youden\u0026rsquo;s index), and high specificity (targeting 90.0%). For each threshold, we also computed positive predictive value (PPV), negative predictive value (NPV), and the percentage of the population requiring testing. To calculate confidence intervals, we used the Wilson score method to estimate confidence intervals for proportion-based metrics in binary classification (\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eAdditionally, we conducted subgroup analyses to assess model performance consistency across key demographic characteristics. We calculated AUC values with 95% confidence intervals for population subgroups defined by population groups, age, and country of birth.\u003c/p\u003e \u003cp\u003eFor calibration assessment, we evaluated how well predicted probabilities aligned with observed outcomes. Following the original study's methodology, we divided predictions into 200 equally sized subgroups sorted by predicted probability and calculated the observed prevalence within each subgroup. We fitted logistic functions to these data points to create smooth calibration curves and visualise the relationship between predicted probabilities and observed prevalence. All statistical analyses were conducted using the Python programming language (version 3.9.12).\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cp\u003eDemographic Characteristics\u003c/p\u003e \u003cp\u003eBetween January 2013 and December 2023, we analysed 159,043 consultations for HIV, 168,443 for syphilis, and 207,582 for both gonorrhoea and chlamydia at the SSHC. Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e summarises the demographic and behavioural characteristics of these consultations. The median age was approximately 30 years across all datasets, with men who have sex with men (MSM) representing most of the consultations (60.2\u0026ndash;68.8%). Most attendees (64.2\u0026ndash;65.5%) were born overseas, and inconsistent condom use was reported in nearly half of all consultations.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eCharacteristics of Clinic Consultations in Individual Data Sets\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePredictors\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHIV\u003c/p\u003e \u003cp\u003e(N\u0026thinsp;=\u0026thinsp;159043\u003c/p\u003e \u003cp\u003eConsultations)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSyphilis\u003c/p\u003e \u003cp\u003e(N\u0026thinsp;=\u0026thinsp;168443\u003c/p\u003e \u003cp\u003eConsultations)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eGonorrhoea\u003c/p\u003e \u003cp\u003e(N\u0026thinsp;=\u0026thinsp;207582\u003c/p\u003e \u003cp\u003eConsultations)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eChlamydia\u003c/p\u003e \u003cp\u003e(N\u0026thinsp;=\u0026thinsp;207582\u003c/p\u003e \u003cp\u003eConsultations)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAge, median (IQR)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e30 (25\u0026ndash;37)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e30 (25\u0026ndash;37)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e29 (25\u0026ndash;36)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e29 (25\u0026ndash;36)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eCountry of birth, n (%)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAustralia and New Zealand\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e56949 (35.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e60336 (35.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e71583 (34.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e71583 (34.5)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOverseas\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e102094 (64.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e108107 (64.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e135999 (65.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e135999 (65.5)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMissing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSTI Symptoms, n (%)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePresent\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e29953 (18.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e32115 (19.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e50575 (24.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e50496 (24.3)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAbsent\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e129090 (81.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e136328 (81.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e157007 (75.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e157086 (75.7)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003ePopulation type, n (%)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMSM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e109362 (68.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e115933 (68.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e124963 (60.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e124964 (60.2)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHeterosexual male\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e23102 (14.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e25516 (15.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e37229 (18.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e37228 (17.9)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFemale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e26579 (16.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e26994 (16.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e45390 (21.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e45390 (21.9)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eCondom use with male partners, n (%)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAlways\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e41447 (26.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e43463 (25.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e48936 (23.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e48935 (23.6)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSometimes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e76499 (48.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e79712 (47.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e100084 (48.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e100095 (48.2)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNever\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e27167 (17.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e29221 (17.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e38796 (18.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e38795 (18.7)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNot Applicable\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e10330 (6.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e11510 (6.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e13240 (6.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e13235 (6.4)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUnsure/Decline\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e46 (0.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e50 (0.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e74 (0.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e74 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMissing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3554 (2.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4487 (2.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e6452 (3.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e6448 (3.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNumber of male sexual partners in last 12 months, median (IQR)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e7 (\u003cspan additionalcitationids=\"CR4 CR5 CR6 CR7 CR8 CR9 CR10 CR11 CR12 CR13 CR14\" citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7 (\u003cspan additionalcitationids=\"CR4 CR5 CR6 CR7 CR8 CR9 CR10 CR11 CR12 CR13 CR14\" citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e6 (\u003cspan additionalcitationids=\"CR4 CR5 CR6 CR7 CR8 CR9 CR10 CR11 CR12 CR13 CR14\" citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e6 (\u003cspan additionalcitationids=\"CR4 CR5 CR6 CR7 CR8 CR9 CR10 CR11 CR12 CR13 CR14\" citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNumber of female sexual partners in last 12 months, median (IQR)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3 (\u003cspan additionalcitationids=\"CR2 CR3 CR4 CR5 CR6\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3 (\u003cspan additionalcitationids=\"CR2 CR3 CR4 CR5\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3 (\u003cspan additionalcitationids=\"CR2 CR3 CR4 CR5 CR6\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e3 (\u003cspan additionalcitationids=\"CR2 CR3 CR4 CR5 CR6\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eLast time injected drugs not prescribed by doctor, n (%)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNever\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e156433 (98.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e165357 (98.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e204079 (98.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e204077 (98.3)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLess than 3 months\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1396 (0.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1720 (1.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1963 (1.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1962 (1.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e3\u0026ndash;12 months\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e577 (0.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e709 (0.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e789 (0.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e789 (0.4)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMore than 12 months\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e637 (0.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e657 (0.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e751 (0.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e754 (0.4)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003ePast history of gonorrhoea, n (%)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e44512 (28.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e49445 (29.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e61432 (29.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e61420 (29.6)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e114516 (72.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e118982 (70.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e146131 (70.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e146143 (70.4)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMissing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e15 (0.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e16 (0.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e19 (0.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e19 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003ePast history of nonspecific urethritis, n (%)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e16453 (10.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e17972 (10.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e24232 (11.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e24211 (11.7)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e142575 (89.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e150455 (89.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e183331 (88.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e183352 (88.3)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMissing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e15 (0.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e16 (0.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e19 (0.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e19 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003ePast history of syphilis, n (%)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e19757 (12.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e23550 (14.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e28580 (13.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e28557 (13.8)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e139271 (87.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e144877 (86.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e178983 (86.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e179006 (86.2)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMissing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e15 (0.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e16 (0.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e19 (0.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e19 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSexual contact with someone diagnosed with gonorrhoea, n (%)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3386 (2.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3635 (2.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e5409 (2.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e5381 (2.6)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e155657 (97.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e164808 (97.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e202173 (97.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e202201 (97.4)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSexual contact with someone diagnosed with chlamydia, n (%)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e4657 (2.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4870 (2.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e8578 (4.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e8587 (4.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e154386 (97.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e163573 (97.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e199004 (95.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e198995 (95.9)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSexual contact with someone diagnosed with syphilis, n (%)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1426 (0.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1848 (1.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1573 (0.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1572 (0.7)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e157617 (99.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e166595 (98.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e206009 (99.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e206010 (99.3)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eInfection positivity, n (%)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePositive\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1163 (0.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3410 (2.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e15423 (7.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e20801 (10.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNegative\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e157880 (99.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e165033 (98.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e192159 (92.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e186781 (90.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eInfection Positivity\u003c/p\u003e \u003cp\u003eThe HIV/STI positivity at SSHC were 0.7% for HIV, 2.0% for syphilis, 7.4% for gonorrhoea, and 10.0% for chlamydia (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Comparatively, MSHC reported lower positivity, with 0.3% for HIV, 1.7% for syphilis, 5.9% for gonorrhoea, and 8.1% for chlamydia (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e), indicating potential variations in patient demographics, testing patterns, or underlying transmission dynamics between the two centres.\u003c/p\u003e \u003cp\u003eExternal Validation Performance\u003c/p\u003e \u003cp\u003eThe external validation of \u003cem\u003eMySTIRisk\u003c/em\u003e models at SSHC demonstrated lower discriminative ability than the original MSHC models across all four infections. The area under the ROC curve (AUC) values for SSHC were 0.67 (95% CI: 0.65\u0026ndash;0.68) for HIV, 0.70 (95% CI: 0.69\u0026ndash;0.71) for syphilis, 0.73 (95% CI: 0.73\u0026ndash;0.74) for gonorrhoea, and 0.65 (95% CI: 0.65\u0026ndash;0.66) for chlamydia. These results were significantly lower than the performance metrics at MSHC, which showed AUC values of 0.82 (95% CI: 0.80\u0026ndash;0.85) for HIV, 0.87 (95% CI: 0.86\u0026ndash;0.88) for syphilis, 0.84 (95% CI: 0.83\u0026ndash;0.84) for gonorrhoea, and 0.74 (95% CI: 0.73\u0026ndash;0.74) for chlamydia. Statistical comparison revealed significant differences in predictive performance across all four infections (all p-values\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Figure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e displays the ROC curves for both centres across all four infections, illustrating consistent differences in model discrimination between these distinct clinical populations.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eSubgroup Analysis\u003c/p\u003e \u003cp\u003e \u003cem\u003eMySTIRisk\u003c/em\u003e demonstrated varying performance across demographic subgroups (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). The HIV model showed better discrimination among MSM (AUC 0.78, 95% CI: 0.74\u0026ndash;0.81) compared to heterosexual males (AUC 0.65, 95% CI: 0.63\u0026ndash;0.68, p\u0026thinsp;=\u0026thinsp;0.06) and females (AUC 0.61, 95% CI: 0.59\u0026ndash;0.64, p\u0026thinsp;=\u0026thinsp;0.002), with these differences reaching statistical significance. For gonorrhoea, performance was markedly lower in females (AUC 0.57, 95% CI: 0.55\u0026ndash;0.59, p\u0026thinsp;=\u0026thinsp;0.2) compared to MSM. Age-related variations were most notable for gonorrhoea, where younger attendees (\u0026lt;\u0026thinsp;25 years) exhibited significantly higher predictive performance (AUC 0.79, 95% CI: 0.78\u0026ndash;0.80) than other age groups. The chlamydia model showed the most consistent performance across demographic categories (AUCs 0.64\u0026ndash;0.67). Country of birth had minimal impact on model performance for all infections.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSubgroup Analyses of MySTIRisk Model Performance across Key Demographics\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSubgroup\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHIV\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSyphilis\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eGonorrhoea\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eChlamydia\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOverall\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.67 (0.65\u0026ndash;0.68)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.70 (0.69\u0026ndash;0.71)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.73 (0.73\u0026ndash;0.74)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.65 (0.65\u0026ndash;0.66)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e \u003cp\u003ePopulation type\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMSM (reference)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.78 (0.74\u0026ndash;0.81)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.70 (0.68\u0026ndash;0.71)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.71 (0.70\u0026ndash;0.71)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.66 (0.65\u0026ndash;0.66)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHeterosexual male\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.65 (0.63\u0026ndash;0.68)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.63 (0.61\u0026ndash;0.66)***\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.72 (0.71\u0026ndash;0.73)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.66 (0.65\u0026ndash;0.67)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFemale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.61 (0.59\u0026ndash;0.64)**\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.67 (0.62\u0026ndash;0.71)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.57 (0.55\u0026ndash;0.59)***\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.64 (0.63\u0026ndash;0.65)**\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e \u003cp\u003eAge\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;25 years\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.64 (0.62\u0026ndash;0.66)***\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.73 (0.70\u0026ndash;0.75)***\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.79 (0.78\u0026ndash;0.80)***\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.66 (0.65\u0026ndash;0.67)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e25\u0026ndash;34 years (reference)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.68 (0.64\u0026ndash;0.72)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.68 (0.66\u0026ndash;0.69)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.72 (0.72\u0026ndash;0.73)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.65 (0.65\u0026ndash;0.66)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026ge;\u0026thinsp;35 years\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.74 (0.68\u0026ndash;0.80)*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.70 (0.69\u0026ndash;0.72)*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.70 (0.69\u0026ndash;0.70)***\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.64 (0.63\u0026ndash;0.65)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e \u003cp\u003eCountry of birth\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAustralia and New Zealand (reference)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.65 (0.62\u0026ndash;0.69)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.71 (0.69\u0026ndash;0.72)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.72 (0.71\u0026ndash;0.73)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.67 (0.66\u0026ndash;0.67)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOverseas\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.66 (0.65\u0026ndash;0.68)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.70 (0.69\u0026ndash;0.71)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.74 (0.73\u0026ndash;0.74)***\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.65 (0.64\u0026ndash;0.65)***\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"5\"\u003e*p\u0026thinsp;\u0026lt;\u0026thinsp;0.05, **p\u0026thinsp;\u0026lt;\u0026thinsp;0.01, ***p\u0026thinsp;\u0026lt;\u0026thinsp;0.001 compared to reference group\u003c/td\u003e\u003c/tr\u003e \u003ctr\u003e\u003ctd colspan=\"5\"\u003eMSM: men who have sex with men; AUC: area under the receiver operating characteristic curve; CI: confidence interval\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eThreshold Analysis\u003c/p\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e presents the performance of \u003cem\u003eMySTIRisk\u003c/em\u003e across multiple risk thresholds for each infection. At thresholds calibrated for high sensitivity (90.0%), the proportion of the population requiring testing ranged from 70.3% (gonorrhoea) to 81.3% (chlamydia), with corresponding specificities between 19.7% and 31.3%. When optimising for balanced sensitivity and specificity, the models achieved more moderate but clinically useful performance. For HIV, a threshold of 0.62 resulted in 61.0% sensitivity (95% CI: 58.2\u0026ndash;63.8%) and 64.1% specificity (95% CI: 63.8\u0026ndash;64.3%), requiring testing of only 36.1% of the population. For syphilis, gonorrhoea, and chlamydia, the balanced thresholds achieved sensitivities of 58.6%, 64.1%, and 60.1%, respectively, with specificities ranging from 62.9\u0026ndash;74.9%. At high specificity thresholds (90.0%), the models demonstrated lower sensitivities (23.5\u0026ndash;37.0%) but substantially reduced the testing proportion to 10.1\u0026ndash;12.0% of the population.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePerformance of the MySTIRisk Models at Different Risk Thresholds\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"8\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInfections\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eScenario\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eThreshold\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSensitivity \u003c/p\u003e \u003cp\u003e(95% CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSpecificity \u003c/p\u003e \u003cp\u003e(95% CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003ePPV \u003c/p\u003e \u003cp\u003e(95% CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eNPV \u003c/p\u003e \u003cp\u003e(95% CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003e% Population \u003c/p\u003e \u003cp\u003eTested\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eHIV\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSensitivity at 90%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.36\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e89.9% \u003c/p\u003e \u003cp\u003e(88.1\u0026ndash;91.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e27.2% \u003c/p\u003e \u003cp\u003e(27.0-27.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.9% \u003c/p\u003e \u003cp\u003e(0.8-1.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e99.7% \u003c/p\u003e \u003cp\u003e(99.7\u0026ndash;99.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e72.9%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eBalanced Sensitivity and Specificity*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.62\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e61.0% \u003c/p\u003e \u003cp\u003e(58.2\u0026ndash;63.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e64.1% \u003c/p\u003e \u003cp\u003e(63.8\u0026ndash;64.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.2% \u003c/p\u003e \u003cp\u003e(1.1\u0026ndash;1.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e99.6% \u003c/p\u003e \u003cp\u003e(99.5\u0026ndash;99.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e36.1%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSpecificity at 90%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e26.0% \u003c/p\u003e \u003cp\u003e(23.5\u0026ndash;28.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e90.0% \u003c/p\u003e \u003cp\u003e(89.9\u0026ndash;90.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.9% \u003c/p\u003e \u003cp\u003e(1.7\u0026ndash;2.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e99.4% \u003c/p\u003e \u003cp\u003e(99.4\u0026ndash;99.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e10.1%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eSyphilis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSensitivity at 90%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.25\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e90.1% \u003c/p\u003e \u003cp\u003e(89.0\u0026ndash;91.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e25.6% \u003c/p\u003e \u003cp\u003e(25.4\u0026ndash;25.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2.4% \u003c/p\u003e \u003cp\u003e(2.4\u0026ndash;2.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e99.2% \u003c/p\u003e \u003cp\u003e(99.1\u0026ndash;99.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e74.7%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eBalanced Sensitivity and Specificity*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.46\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e58.6% \u003c/p\u003e \u003cp\u003e(56.9\u0026ndash;60.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e74.9% \u003c/p\u003e \u003cp\u003e(74.7\u0026ndash;75.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e4.6% \u003c/p\u003e \u003cp\u003e(4.4\u0026ndash;4.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e98.9% \u003c/p\u003e \u003cp\u003e(98.8\u0026ndash;98.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e25.8%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSpecificity at 90%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.72\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e23.5% \u003c/p\u003e \u003cp\u003e(22.1\u0026ndash;25.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e90.1% \u003c/p\u003e \u003cp\u003e(89.9\u0026ndash;90.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e4.7% \u003c/p\u003e \u003cp\u003e(4.4-5.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e98.3% \u003c/p\u003e \u003cp\u003e(98.2\u0026ndash;98.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e10.2%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eGonorrhoea\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSensitivity at 90%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.33\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e90.0% \u003c/p\u003e \u003cp\u003e(89.5\u0026ndash;90.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e31.3% \u003c/p\u003e \u003cp\u003e(31.1\u0026ndash;31.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e9.5% \u003c/p\u003e \u003cp\u003e(9.4\u0026ndash;9.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e97.5% \u003c/p\u003e \u003cp\u003e(97.4\u0026ndash;97.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e70.3%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eBalanced Sensitivity and Specificity*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.57\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e64.1% \u003c/p\u003e \u003cp\u003e(63.3\u0026ndash;64.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e70.3% \u003c/p\u003e \u003cp\u003e(70.1\u0026ndash;70.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e14.8% \u003c/p\u003e \u003cp\u003e(14.5\u0026ndash;15.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e96.1% \u003c/p\u003e \u003cp\u003e(96.0-96.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e32.3%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSpecificity at 90%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.66\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e37.0% \u003c/p\u003e \u003cp\u003e(36.2\u0026ndash;37.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e90.0% \u003c/p\u003e \u003cp\u003e(89.9\u0026ndash;90.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e22.9% \u003c/p\u003e \u003cp\u003e(22.4\u0026ndash;23.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e94.7% \u003c/p\u003e \u003cp\u003e(94.6\u0026ndash;94.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e12.0%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eChlamydia\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSensitivity at 90%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.37\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e90.0% \u003c/p\u003e \u003cp\u003e(89.6\u0026ndash;90.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e19.7% \u003c/p\u003e \u003cp\u003e(19.5\u0026ndash;19.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e11.1% \u003c/p\u003e \u003cp\u003e(10.9\u0026ndash;11.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e94.6% \u003c/p\u003e \u003cp\u003e(94.4\u0026ndash;94.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e81.3%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eBalanced Sensitivity and Specificity*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.52\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e60.1% \u003c/p\u003e \u003cp\u003e(59.4\u0026ndash;60.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e62.9% \u003c/p\u003e \u003cp\u003e(62.7\u0026ndash;63.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e15.3% \u003c/p\u003e \u003cp\u003e(15.0-15.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e93.4% \u003c/p\u003e \u003cp\u003e(93.3\u0026ndash;93.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e39.4%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSpecificity at 90%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.63\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e26.6% \u003c/p\u003e \u003cp\u003e(26.0-27.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e90.0% \u003c/p\u003e \u003cp\u003e(89.9\u0026ndash;90.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e22.8% \u003c/p\u003e \u003cp\u003e(22.3\u0026ndash;23.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e91.7% \u003c/p\u003e \u003cp\u003e(91.5\u0026ndash;91.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e11.7%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"8\"\u003eSensitivity represents the percentage of true positive cases correctly identified by the model, while specificity measures the percentage of true negative cases correctly classified. The positive predictive value (PPV) represents the proportion of positive results that are truly positive cases, whereas the negative predictive value (NPV) reflects the proportion of negative results that are truly negative cases. The percentage of the population tested corresponds to the proportion of individuals who exceeded the given probability threshold.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eCalibration Assessment\u003c/p\u003e \u003cp\u003eIn our calibration assessment, we examined the association between predicted probabilities and observed prevalence for all four STIs (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e), as mentioned in the development study of \u003cem\u003eMySTIRisk\u003c/em\u003e (\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e). The calibration plots demonstrated excellent agreement between predicted and observed risk across the probability spectrum for gonorrhoea and chlamydia, with data points closely following the logistic function fit. For HIV, the model showed good calibration at lower risk probabilities but slightly underestimated infection rates at higher probabilities (\u0026gt;\u0026thinsp;0.7), although this region contained fewer data points. The syphilis model demonstrated the most variation in calibration, with greater scatter around the fitted curve, particularly at mid-range probabilities (0.4\u0026ndash;0.7).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eOur external validation of \u003cem\u003eMySTIRisk\u003c/em\u003e demonstrated moderate to good predictive performance for all four STIs using the data from SSHC, although with significant decrements compared to the original \u003cem\u003eMySTIRisk\u003c/em\u003e models developed at the MSHC. We found AUC values ranging from 0.65 for chlamydia and 0.73 for gonorrhoea, reflecting reasonable discriminative ability across infections, though lower than the original MSHC values (0.74\u0026ndash;0.87). This performance decline aligns with established patterns in machine learning validation studies, where models typically show reduced effectiveness when applied to external populations due to differences in demographics, behavioural patterns, and testing practices (\u003cspan additionalcitationids=\"CR17\" citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e). We observed notable population-specific variations, with the models performing better for HIV prediction among MSM (AUC 0.78) and for gonorrhoea among younger attendees (AUC 0.79). These findings underscore the importance of context-specific considerations in AI-based risk assessment tools. While \u003cem\u003eMySTIRisk\u003c/em\u003e retains clinical utility across diverse settings, our results highlight the necessity of local validation and potential recalibration before widespread implementation, consistent with recent studies on clinical prediction models in infectious disease settings (\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eOur study demonstrated that \u003cem\u003eMySTIRisk\u003c/em\u003e models exhibited consistently lower discriminative performance when using the data from SSHC, with AUC values at 0.65\u0026ndash;0.73 markedly reduced from those reported at MSHC (0.74\u0026ndash;0.87). While such performance decrements are anticipated in external validation studies of machine learning models, several specific factors may explain these observed differences. First, the demographic and epidemiological composition differed substantially between sites\u0026mdash;the SSHC dataset comprised a higher proportion of overseas-born attendees (approximately 65%) and MSM and lower proportions of heterosexual individuals than the more evenly distributed MSHC population. These population differences may influence how risk factors manifest and interact across settings. Second, differences in clinical documentation protocols may have affected model input quality; notably, while MSHC recorded sexual encounters overseas (\"sex with someone outside Australia or New Zealand or had sex in Australia with someone from overseas\"), SSHC did not capture this risk factor. Third, despite having fewer symptomatic presentations (18.8\u0026ndash;24.4% vs. 28.3\u0026ndash;35.7% at MSHC), SSHC paradoxically showed higher positivity across all infections, indicating population-specific testing and risk dynamics that models may struggle to capture. Studies have demonstrated that such disparities significantly impact machine learning model performance during external validation, particularly for models developed using comprehensive datasets (\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e). Fourth, the methods of triage, testing protocols, and patterns of symptomatic individuals may vary between the two populations. These differences may have led to differences in the patterns of risk factors for each infection between the MSHC and SSHC populations. Despite these differences, it is encouraging that \u003cem\u003eMySTIRisk\u003c/em\u003e maintained moderate discriminative ability using the SSHC data, suggesting the core risk factors identified during development retain predictive value across Australian sexual health settings.\u003c/p\u003e \u003cp\u003eOur subgroup analyses revealed important differences in \u003cem\u003eMySTIRisk\u003c/em\u003e's performance across demographic categories. For HIV prediction, the markedly superior performance among MSM (AUC 0.78) compared to heterosexual populations (AUC 0.61\u0026ndash;0.65) suggested that the model captured MSM-specific risk patterns more effectively, potentially reflecting the higher prevalence and better-characterised transmission dynamics in this group. Age-stratified analysis demonstrated that younger attendees (\u0026lt;\u0026thinsp;25 years) showed significantly higher prediction accuracy for gonorrhoea (AUC 0.79), likely due to more consistent risk behaviours or testing patterns in this group. Interestingly, country of birth had minimal impact on model performance except for gonorrhoea and chlamydia. These demographic disparities in predictive performance align with findings from Franklin et al. (\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e), which demonstrated that demographic characteristics can introduce bias in machine learning models, affecting their generalisability across diverse populations. The observed variations in predictive performance across subgroups indicate that risk factors have different predictive weights in diverse populations, reflecting the complex interplay of behavioural, social, and biological determinants of STI transmissions.\u003c/p\u003e \u003cp\u003eThe threshold analysis findings provided valuable guidance for implementing \u003cem\u003eMySTIRisk\u003c/em\u003e in diverse clinical settings. Setting risk thresholds to achieve high sensitivity (90.0%) would capture most infections but require testing 70.3\u0026ndash;81.3% of attendees, which may place a significant strain on resources without substantially improving efficiency compared to universal testing. Conversely, high specificity thresholds (90.0%) would significantly reduce testing volume to only 10.1\u0026ndash;12.0% of the population but would result in missing 63.0\u0026ndash;76.5% of infections, potentially limiting early detection efforts. A balanced approach using Youden's index thresholds could represent an optimal middle ground by identifying 58.6\u0026ndash;64.1% of infections while reducing the testing proportion to 25.8\u0026ndash;39.4%. These results demonstrate how risk prediction tools can be calibrated according to local resource constraints, such as limited test kits, staffing shortages, or clinic capacity, while addressing public health priorities. In resource-limited settings, \u003cem\u003eMySTIRisk\u003c/em\u003e could help prioritise testing for individuals at highest risk. Meanwhile, in well-resourced environments, it could complement universal testing by identifying candidates for more comprehensive screening. The flexibility to adjust thresholds based on specific clinical contexts represents a key advantage of machine learning-based risk assessment tools compared to traditional screening questionnaires, which rely on fixed criteria and may lack adaptability to evolving epidemiological trends.\u003c/p\u003e \u003cp\u003eFuture research should address several key areas to enhance the clinical utility of \u003cem\u003eMySTIRisk\u003c/em\u003e and similar AI-based prediction tools. First, transfer learning approaches could be explored to adapt models to specific clinical settings while preserving their core predictive capabilities. This would involve fine-tuning the existing models with local data rather than complete retraining, potentially improving performance while maintaining generalisability (\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e). Second, future iterations should expand demographic representation, particularly including trans and gender diverse individuals who were excluded from current analyses, as there were no specific sexual behavioural questions for this population at the time when \u003cem\u003eMySTIRisk\u003c/em\u003e was developed. Third, prospective implementation studies are needed to evaluate the real-world impact of integrating \u003cem\u003eMySTIRisk\u003c/em\u003e into clinical workflows, including assessment of resource utilisation, infection detection rates, and user acceptance. Additional research could explore the temporal stability of these predictions, as sexual behaviour patterns and infection dynamics evolve over time. Finally, comparative effectiveness research should evaluate whether AI-based risk stratification outperforms traditional clinician judgement or simpler risk assessment tools in terms of both accuracy and cost-effectiveness. It would strengthen the evidence base for the broader adoption of machine learning approaches in sexual health services and inform best practices for their implementation across diverse healthcare settings.\u003c/p\u003e \u003cp\u003eOur study has several notable strengths. This is the first external validation of an AI-based HIV/STI risk assessment tool across Australia's two largest publicly funded sexual health centres, representing a significant proportion of the country's urban sexual health services and providing robust statistical power through its extensive sample size. Our adherence to TRIPOD guidelines for prediction model validation ensures methodological quality and transparent reporting. The comprehensive assessment across four key STIs offers valuable comparative insights into prediction patterns across different infections. Additionally, our analysis of multiple risk thresholds provides practical implementation guidance that can be tailored to various clinical scenarios. By validating \u003cem\u003eMySTIRisk\u003c/em\u003e in a demographically distinct setting, we have tested the model's resilience to population heterogeneity, a critical consideration for AI tools intended for widespread implementation across diverse healthcare environments.\u003c/p\u003e \u003cp\u003eOur study has several important limitations that warrant consideration when interpreting the results. First, the retrospective nature of our analyses may have introduced selection bias though our large sample size spanning a decade helps mitigate some temporal variation effects. Second, we excluded transgender individuals to maintain consistency with the original model development, representing a significant gap in generalisability that future iterations must address; this exclusion was intentional because data systems did not capture epidemiological risk on these individuals until recently. With time adequate epidemiological risk data will accumulate and allow this analysis in the future. Third, differences in clinical documentation between sites meant that some risk factors used in the original model development were not identically captured at SSHC, potentially affecting model performance. Although we attempted to standardise variables where possible, some differences were unavoidable due to inherent variations in clinical workflows between centres. Fourth, while the study included a diverse patient population, both centres are urban sexual health clinics, potentially limiting generalisability to rural settings or primary care contexts where STI testing also occurs. However, these two centres serve the largest metropolitan areas in Australia, providing substantial population coverage. Fifth, we did not assess the models' performance for predicting multiple concurrent infections, which represents a clinically important scenario, as the original model development focused on individual infections. Sixth, the PPV were relatively low, particularly for HIV (0.9\u0026ndash;1.9%), which should be interpreted in the context of the low prevalence of these infections in the study population. This reflects a common challenge in predictive modelling for low-prevalence conditions, where even highly accurate models may yield modest PPV values. Finally, while we did not directly evaluate patients' perspectives on AI-based risk assessment in this validation study, subsequent qualitative research conducted by our team has addressed these dimensions, revealing both the tool's potential value for overcoming barriers to sexual health information and areas for improving inclusivity and guidance (\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eIn conclusion, our external validation of \u003cem\u003eMySTIRisk\u003c/em\u003e demonstrated moderate to good predictive performance across all four STIs, though with decreased discrimination compared to the original development site. Despite this performance drop, the tool maintained discriminative ability across different clinical populations, supporting its potential application in diverse sexual health settings. Population-specific variations highlight the need for contextualised implementation, with adjustable thresholds offering flexibility based on local resources and priorities. While improvements are needed, this first external validation across Australia's two largest sexual health centres provides evidence supporting the broader application of AI-based risk assessment in sexual health, though local validation remains essential before widespread implementation.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eAUC Area under the curve\u003c/p\u003e\n\u003cp\u003eCI Confidence Interval\u003c/p\u003e\n\u003cp\u003eEHR Electronic health record\u003c/p\u003e\n\u003cp\u003eGBM Gradient boosting machine\u003c/p\u003e\n\u003cp\u003eIQR Interquartile Range \u003c/p\u003e\n\u003cp\u003eMSHC Melbourne Sexual Health Centre\u003c/p\u003e\n\u003cp\u003eMSM Men who have sex with men\u003c/p\u003e\n\u003cp\u003eNAAT Nucleic acid amplification test\u003c/p\u003e\n\u003cp\u003eNHMRC National Health and Medical Research Council\u003c/p\u003e\n\u003cp\u003eNPV Negative predictive value\u003c/p\u003e\n\u003cp\u003ePPV Positive predictive value\u003c/p\u003e\n\u003cp\u003eROC Receiver operating characteristic\u003c/p\u003e\n\u003cp\u003eSESLHD South Eastern Sydney Local Health District\u003c/p\u003e\n\u003cp\u003eSSHC Sydney Sexual Health Centre\u003c/p\u003e\n\u003cp\u003eSTI Sexually transmitted infections\u003c/p\u003e\n\u003cp\u003eTRIPOD Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eEthics Approval and Consent to Participate\u003c/h2\u003e\n\u003cp\u003eThis multi-site study received ethical approval from the Alfred Hospital Ethics Committee (Project 145/24) under the Human Research Ethics Application (HREA 105599). Site-specific approvals were obtained from both the South Eastern Sydney Local Health District (SESLHD) and Alfred Health (Victoria). The research was conducted in accordance with the ethical principles outlined in the Declaration of Helsinki. As this was a retrospective study using de-identified data, the Alfred Hospital Ethics Committee granted a consent waiver for the collection, use and disclosure of participants\u0026apos; health and personal information in accordance with the Victorian Office of the Health Services Commissioner\u0026apos;s Statutory Guidelines on Research, the Health Records and Information Privacy Act 2002 (NSW): Statutory Guidelines on Research, and the NHMRC Guidelines approved under Sections 95 \u0026amp; 95A of the Privacy Act 1988, ensuring compliance with ethical standards.\u003c/p\u003e\n\u003ch2\u003eAcknowledgements\u003c/h2\u003e\n\u003cp\u003eThe authors sincerely thank Monash University for providing a PhD scholarship for PL. We also extend our thanks to all contributors who played a role in this study.\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003eAuthor Contributions\u003c/h2\u003e\n\u003cp\u003eCKF, LZ and PL conceptualised and designed the study methodology. CKF provided overall supervision throughout all phases of the research. PL coordinated the project implementation, secured ethics approval, conducted the statistical analyses, and drafted the initial manuscript. AR, HL and RV were responsible for data extraction, validation, and cleaning processes. The development and technical implementation of the \u003cem\u003eMySTIRisk\u003c/em\u003e web application was a collaborative effort involving PL, XX, YB, CKF, LZ, and EPFC. CKF, EPFC and JO provided statistical expertise and methodological guidance. All authors contributed to the manuscript\u0026apos;s revision and read and approved the submitted version.\u003c/p\u003e\n\u003ch2\u003eFunding\u003c/h2\u003e\n\u003cp\u003eCKF is supported by a National Health and Medical Research Council (NHMRC) Leadership Investigator Grant (GNT1172900). JO is supported by the NHMRC Emerging Leadership Investigator Grant (GNT1193955). EPFC is supported by an NHMRC Leadership Investigator Grant (GNT2033299).\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003ePotential Conflicts of Interest\u003c/h2\u003e\n\u003cp\u003ePL, NS, XX, YB, LZ and CKF have licensed the use of \u003cem\u003eMySTIRisk\u003c/em\u003e to Helfie (Level 5/171 La Trobe St, Melbourne VIC 3000). The remaining authors declare no conflicts of interest.\u003c/p\u003e\n\u003ch2\u003eAvailability of Data and Materials\u003c/h2\u003e\n\u003cp\u003eThe data supporting this study\u0026apos;s findings are not openly available due to sensitivity and are available from the corresponding author, Dr. Phyu Mon Latt, at [email protected], upon reasonable request. Data are in controlled access data storage at the Melbourne Sexual Health Centre.\u003c/p\u003e\n\u003ch2\u003eConsent for Publication\u003c/h2\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e\n\u003ch2\u003eClinical Trial Number\u003c/h2\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eNattabi B, Matthews V, Bailie J, Rumbold A, Scrimgeour D, Schierhout G, et al. Wide variation in sexually transmitted infection testing and counselling at Aboriginal primary health care centres in Australia: analysis of longitudinal continuous quality improvement data. BMC Infect Dis. 2017;17(1):148.\u003c/li\u003e\n\u003cli\u003eRogers B, Tao J, Murphy M, Chan PA. The COVID-19 Pandemic and Sexually Transmitted Infections: Where Do We Go From Here? Sex Transm Dis. 2021;48(7):e94-e6.\u003c/li\u003e\n\u003cli\u003eWorld Health Organization. Sexually transmitted infections (STIs) fact sheets Geneva20 July 2023 [Available from: https://www.who.int/news-room/fact-sheets/detail/sexually-transmitted-infections-(stis)/].\u003c/li\u003e\n\u003cli\u003eOkal J, Lango D, Matheka J, Obare F, Ngunu-Gituathi C, Mugambi M, et al. \u0026quot;It is always better for a man to know his HIV status\u0026quot; - A qualitative study exploring the context, barriers and facilitators of HIV testing among men in Nairobi, Kenya. PLoS One. 2020;15(4):e0231645.\u003c/li\u003e\n\u003cli\u003eKhan AR, Altalbe A. Potential impacts of Russo-Ukraine conflict and its psychological consequences among Ukrainian adults: the post-COVID-19 era. Front Public Health. 2023;11:1280423.\u003c/li\u003e\n\u003cli\u003eFarquharson RM, Fairley CK, Abraham E, Bradshaw CS, Plummer EL, Ong JJ, et al. Time to healthcare seeking following the onset of symptoms among men and women attending a sexual health clinic in Melbourne, Australia. Front Med (Lausanne). 2022;9:915399.\u003c/li\u003e\n\u003cli\u003eScott H, Vittinghoff E, Irvin R, Liu A, Nelson L, Del Rio C, et al. Development and Validation of the Personalized Sexual Health Promotion (SexPro) HIV Risk Prediction Model for Men Who Have Sex with Men in the United States. AIDS Behav. 2020;24(1):274-83.\u003c/li\u003e\n\u003cli\u003eHoenigl M, Weibel N, Mehta SR, Anderson CM, Jenks J, Green N, et al. Development and validation of the San Diego Early Test Score to predict acute and early HIV infection risk in men who have sex with men. Clin Infect Dis. 2015;61(3):468-75.\u003c/li\u003e\n\u003cli\u003eBao Y, Medland NA, Fairley CK, Wu J, Shang X, Chow EPF, et al. Predicting the diagnosis of HIV and sexually transmitted infections among men who have sex with men using machine learning approaches. J Infect. 2021;82(1):48-59.\u003c/li\u003e\n\u003cli\u003eBalzer LB, Havlir DV, Kamya MR, Chamie G, Charlebois ED, Clark TD, et al. Machine Learning to Identify Persons at High-Risk of Human Immunodeficiency Virus Acquisition in Rural Kenya and Uganda. Clinical Infectious Diseases. 2020;71(9):2326-33.\u003c/li\u003e\n\u003cli\u003ePhyu Mon L, Nyi Nyi S, Xianglong X, Rashidur R, Eric PFC, Jason JO, et al. Assessing disparity in the distribution of HIV and sexually transmitted infections in Australia: a retrospective cross-sectional study using Gini coefficients. BMJ Public Health. 2023;1(1):e000012.\u003c/li\u003e\n\u003cli\u003eXu X, Yu Z, Ge Z, Chow EPF, Bao Y, Ong JJ, et al. Web-Based Risk Prediction Tool for an Individual\u0026apos;s Risk of HIV and Sexually Transmitted Infections Using Machine Learning Algorithms: Development and External Validation Study. J Med Internet Res. 2022;24(8):e37850.\u003c/li\u003e\n\u003cli\u003eXu X, Ge Z, Chow EPF, Yu Z, Lee D, Wu J, et al. A Machine-Learning-Based Risk-Prediction Tool for HIV and Sexually Transmitted Infections Acquisition over the Next 12 Months. J Clin Med. 2022;11(7).\u003c/li\u003e\n\u003cli\u003eLatt PM, Soe NN, Xu X, Ong JJ, Chow EPF, Fairley CK, et al. Identifying individuals at high risk for HIV and sexually transmitted infections with an artificial intelligence-based risk assessment tool. Open Forum Infectious Diseases. 2024.\u003c/li\u003e\n\u003cli\u003eMelbourne Sexual Health Centre. MySTIRisk [Available from: https://mystirisk.mshc.org.au/].\u003c/li\u003e\n\u003cli\u003eGruber S, Krakower D, Menchaca JT, Hsu K, Hawrusik R, Maro JC, et al. Using electronic health records to identify candidates for human immunodeficiency virus pre-exposure prophylaxis: An application of super learning to risk prediction when the outcome is rare. Statistics in Medicine. 2020;39(23):3059-73.\u003c/li\u003e\n\u003cli\u003eHe J, Li J, Jiang S, Cheng W, Jiang J, Xu Y, et al. Application of machine learning algorithms in predicting HIV infection among men who have sex with men: Model development and validation. Frontiers in Public Health. 2022;10.\u003c/li\u003e\n\u003cli\u003eKrakower DS, Gruber S, Hsu K, Menchaca JT, Maro JC, Kruskal BA, et al. Development and validation of an automated HIV prediction algorithm to identify candidates for pre-exposure prophylaxis: a modelling study. The Lancet HIV. 2019;6(10):e696-e704.\u003c/li\u003e\n\u003cli\u003eMajam M, Segal B, Fieggen J, Smith E, Hermans L, Singh L, et al. Utility of a machine-guided tool for assessing risk behaviour associated with contracting HIV in three sites in South Africa. Informatics in Medicine Unlocked. 2023;37.\u003c/li\u003e\n\u003cli\u003eMarcus JL, Hurley LB, Krakower DS, Alexeeff S, Silverberg MJ, Volk JE. Use of electronic health record data and machine learning to identify candidates for HIV pre-exposure prophylaxis: a modelling study. The Lancet HIV. 2019;6(10):e688-e95.\u003c/li\u003e\n\u003cli\u003eXu X, Yu Z, Ge Z, Chow EPF, Bao Y, Ong JJ, et al. Web-Based Risk Prediction Tool for an Individual\u0026apos;s Risk of HIV and Sexually Transmitted Infections Using Machine Learning Algorithms: Development and External Validation Study. Journal of Medical Internet Research. 2022;24(8).\u003c/li\u003e\n\u003cli\u003eLatt PM, Soe NN, Fairley CK, Chow EPF, Johnson CC, Shah P, et al. Machine Learning for Personalised Risk Assessment of HIV, Syphilis, Gonorrhoea, and Chlamydia: A Systematic Review and Meta-analysis. International Journal of Infectious Diseases. 2025:107922.\u003c/li\u003e\n\u003cli\u003eGu J, Epland M, Ma X, Park J, Sanchez RJ, Li Y. A machine-learning algorithm using claims data to identify patients with homozygous familial hypercholesterolemia. Scientific Reports. 2024;14(1):8890.\u003c/li\u003e\n\u003cli\u003eAlelyani S. Detection and Evaluation of Machine Learning Bias. Applied Sciences. 2021;11(14):6271.\u003c/li\u003e\n\u003cli\u003eCollins GS, Moons KGM, Dhiman P, Riley RD, Beam AL, Van Calster B, et al. TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. Bmj. 2024;385:e078378.\u003c/li\u003e\n\u003cli\u003eRiley RD, Archer L, Snell KIE, Ensor J, Dhiman P, Martin GP, et al. Evaluation of clinical prediction models (part 2): how to undertake an external validation study. BMJ. 2024;384:e074820.\u003c/li\u003e\n\u003cli\u003eBrown LD, Cai TT, DasGupta A. Interval estimation for a binomial proportion. Statistical science. 2001;16(2):101-33.\u003c/li\u003e\n\u003cli\u003eLatt PM, Soe NN, Xu X, Ong JJ, Chow EPF, Fairley CK, et al. Identifying Individuals at High Risk for HIV and Sexually Transmitted Infections With an Artificial Intelligence-Based Risk Assessment Tool. Open Forum Infectious Diseases. 2024;11(3).\u003c/li\u003e\n\u003cli\u003eChen X, Hu L, Yu R. Development and external validation of machine learning-based models to predict patients with cellulitis developing sepsis during hospitalisation. BMJ Open. 2024;14(7):e084183.\u003c/li\u003e\n\u003cli\u003eRios R, Miller RJH, Manral N, Sharir T, Einstein AJ, Fish MB, et al. Handling missing values in machine learning to predict patient-specific risk of adverse cardiac events: Insights from REFINE SPECT registry. Comput Biol Med. 2022;145:105449.\u003c/li\u003e\n\u003cli\u003eFranklin G, Stephens R, Piracha M, Tiosano S, Lehouillier F, Koppel R, et al. The Sociodemographic Biases in Machine Learning Algorithms: A Biomedical Informatics Perspective. Life (Basel). 2024;14(6).\u003c/li\u003e\n\u003cli\u003eEbbehoj A, Thunbo M, Andersen OE, Glindtvad MV, Hulman A. Transfer learning for non-image data in clinical research: A scoping review. PLOS Digit Health. 2022;1(2):e0000014.\u003c/li\u003e\n\u003cli\u003eKing AJ, Latt PM, Soe NN, Temple-Smith M, Fairley CK, Chow EP, et al. User experiences of an AI application for predicting risk of sexually transmitted infections. Digit Health. 2024;10:20552076241289646.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-infectious-diseases","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"infd","sideBox":"Learn more about [BMC Infectious Diseases](http://bmcinfectdis.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/infd","title":"BMC Infectious Diseases","twitterHandle":"#bmcinfectdis","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"HIV, sexually transmitted infections, artificial intelligence, risk assessment, machine learning, external validation, predictive modelling, sexual health, digital health","lastPublishedDoi":"10.21203/rs.3.rs-6757880/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6757880/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eIntroduction\u003c/h2\u003e \u003cp\u003eHIV and sexually transmitted infections (STIs) continue to pose significant public health challenges globally. \u003cem\u003eMySTIRisk\u003c/em\u003e, developed at Melbourne Sexual Health Centre (MSHC), is a machine learning-based tool that predicts individual risk for HIV, syphilis, gonorrhoea, and chlamydia using demographic and behavioural data. While initial validation showed promising results, external validation is crucial to assess its generalisability. This study externally validates \u003cem\u003eMySTIRisk\u003c/em\u003e using data from the Sydney Sexual Health Centre (SSHC), Australia's second largest sexual health centre.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003e Following TRIPOD guidelines, we analysed consultations from patients aged 18 years and older attending SSHC between January 2013 and December 2023. Pre-trained \u003cem\u003eMySTIRisk\u003c/em\u003e models were applied directly without modification. Performance was evaluated using the area under the receiver operating characteristic curve (AUC), sensitivity, and specificity at multiple thresholds, with subgroup analyses across demographic characteristics.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eWe analysed 159,043 to 207,582 consultations at SSHC, with a median age of 30 years and 60.2\u0026ndash;68.8% of the consultations involving men who have sex with men. The area under the receiver operating characteristic curve (AUC) values using data from SSHC were 0.67 (95% CI: 0.65\u0026ndash;0.68) for HIV, 0.70 (95% CI: 0.69\u0026ndash;0.71) for syphilis, 0.73 (95% CI: 0.73\u0026ndash;0.74) for gonorrhoea, and 0.65 (95% CI: 0.65\u0026ndash;0.66) for chlamydia, which were lower than the original MSHC validation metrics (0.74\u0026ndash;0.87, all p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Notably, model performance varied across demographic subgroups, with stronger HIV prediction among men who have sex with men with an AUC of 0.78 and better gonorrhoea prediction among younger attendees\u0026thinsp;\u0026lt;\u0026thinsp;25 years with an AUC value of 0.79. At balanced sensitivity-specificity thresholds, the models identified 58.6\u0026ndash;64.1% of infections while requiring testing of only 25.8\u0026ndash;39.4% of the population.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eDespite performance decrements in external validation using SSHC data, \u003cem\u003eMySTIRisk\u003c/em\u003e maintained moderate to good predictive ability across all infections, demonstrating reasonable generalisability across different clinical populations. The demographic variations in performance highlight the importance of context-specific implementation and potential recalibration to optimise clinical utility.\u003c/p\u003e","manuscriptTitle":"External Validation of a Web- and Artificial Intelligence-Based HIV/STI Risk Assessment Tool: Performance Evaluation Using Data from Sydney Sexual Health Centre","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-06-09 05:28:49","doi":"10.21203/rs.3.rs-6757880/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-09-15T11:33:08+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-07-30T16:14:33+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-07-25T15:23:50+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-07-22T16:58:05+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"28124695377141778114253312585119045805","date":"2025-07-16T08:19:03+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"213513309273931969614672885687341150483","date":"2025-07-16T03:01:49+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"87196558802191032306395439714111930557","date":"2025-07-16T01:21:46+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-06-23T08:14:05+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"248741461112296908560785468597917141102","date":"2025-06-10T07:29:35+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"64228294325720171023803884890488648302","date":"2025-06-05T13:41:51+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-06-02T03:28:26+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-05-30T11:30:19+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-05-30T10:44:22+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Infectious Diseases","date":"2025-05-30T10:41:17+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-infectious-diseases","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"infd","sideBox":"Learn more about [BMC Infectious Diseases](http://bmcinfectdis.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/infd","title":"BMC Infectious Diseases","twitterHandle":"#bmcinfectdis","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"0cab567d-584d-40d7-b3cd-f4b9a868c4c7","owner":[],"postedDate":"June 9th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2025-12-01T16:04:43+00:00","versionOfRecord":{"articleIdentity":"rs-6757880","link":"https://doi.org/10.1186/s12879-025-12087-8","journal":{"identity":"bmc-infectious-diseases","isVorOnly":false,"title":"BMC Infectious Diseases"},"publishedOn":"2025-11-25 15:57:22","publishedOnDateReadable":"November 25th, 2025"},"versionCreatedAt":"2025-06-09 05:28:49","video":"","vorDoi":"10.1186/s12879-025-12087-8","vorDoiUrl":"https://doi.org/10.1186/s12879-025-12087-8","workflowStages":[]},"version":"v1","identity":"rs-6757880","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6757880","identity":"rs-6757880","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}