External Validation of Predictive Models for High-Grade Cervical Lesions in Thai Women Undergoing Colposcopy

External Validation of Predictive Models for High-Grade Cervical Lesions in Thai Women Undergoing Colposcopy | 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 Predictive Models for High-Grade Cervical Lesions in Thai Women Undergoing Colposcopy Atikom Eidbur, Thiti Atjimakul This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8615288/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 15 You are reading this latest preprint version Abstract Background Predictive models for high-grade cervical lesions (CIN2+/HSIL) have been developed in various populations; however, their generalizability across different healthcare settings is uncertain. External validation in specific populations is required before clinical implementation. Methods This retrospective external validation study included Thai women who underwent colposcopy between 2022 and 2024 at a tertiary referral center. The primary outcome was histologically confirmed CIN2+/HSIL. Two previously published predictive models were applied without modifications. Model performance was evaluated in terms of discrimination using the area under the receiver operating characteristic curve (AUROC), calibration using calibration plots, calibration metrics, Brier scores, and clinical utility using a decision curve analysis. Results A total of 536 women were included, of whom 28.7% were diagnosed with CIN2+/HSIL. Despite differences in demographic characteristics, HPV genotype distribution, cytological findings, and colposcopic features compared to the original development cohorts, both models demonstrated robust performance in the Thai population. The Xue model showed excellent discrimination, with an AUROC of 0.85 (95% CI: 0.81–0.89), whereas the Sheng model demonstrated good discrimination, with an AUROC of 0.80 (95% CI: 0.77–0.85). At the optimal thresholds, the Xue model achieved a sensitivity of 77% and specificity of 82%, whereas the Sheng model achieved a sensitivity of 63% and specificity of 84%. Both models showed good calibration, with calibration slopes of 1.00, calibration-in-the-large of 0.00, and Brier scores of 0.127 and 0.147. Decision curve analysis demonstrated a positive net clinical benefit across the clinically relevant threshold probabilities. Conclusion Both predictive models showed good external validity and clinical utility in Thai colposcopy patients. These findings support individualized risk stratification in women with abnormal screening results for guiding clinical decisions. Cervical Intraepithelial Neoplasia High-Grade Squamous Intraepithelial Lesion Colposcopy Risk Prediction Model External Validation Human Papillomavirus Figures Figure 1 Figure 2 Figure 3 Figure 4 Background Cervical cancer remains a major public health problem and one of the leading causes of cancer-related morbidity and mortality in women worldwide 1 . Persistent infection with high-risk human papillomavirus (HPV) is the principal etiological factor for the development of cervical precancerous lesions and invasive cervical cancer 2 . The implementation of cervical cancer screening programs, particularly HPV-based screening, has substantially reduced cervical cancer incidence and mortality rates of cervical cancer in many regions 3 , 4 . Nevertheless, effective triage of women with abnormal screening results remains a significant clinical challenge. In Thailand, cervical cancer is the second most common cancer among women 5 , and national screening programs increasingly rely on human papillomavirus (HPV) testing, including both clinician-collected and self-collected samples. Although HPV-based screening has improved the sensitivity for detecting at-risk women, it has also resulted in a growing number of referrals for colposcopy. Importantly, only a minority of women referred for colposcopy are ultimately diagnosed with high-grade cervical lesions (CIN2+/HSIL) on histopathological examination. Recent Thai data indicate that approximately one-quarter of women undergoing colposcopy following a positive HPV test have CIN2+/HSIL, highlighting the substantial burden of unnecessary diagnostic procedures and potential overtreatment 5 . The loop electrosurgical excision procedure (LEEP) is one of the most commonly performed diagnostic and therapeutic interventions for high-grade cervical lesions 6 , 7 . Although effective, LEEP is associated with short- and long-term complications, including bleeding, infection, cervical stenosis, and adverse reproductive outcomes, such as preterm birth and cervical insufficiency 8 . These risks are particularly relevant in women of reproductive age and underscore the importance of accurately identifying those who would benefit most from excisional treatment while avoiding unnecessary procedures in low-risk individuals 9 – 11 . To address this challenge, several multivariable predictive models have been developed to estimate the individual risk of CIN2+/HSIL using combinations of demographic characteristics, cytological findings, HPV genotype information, and colposcopic features. These models aim to support risk-based decision-making and align with contemporary management strategies advocated by international guidelines 4 . However, the majority of existing models have been developed in Western or East Asian populations, and their performance may be influenced by differences in HPV genotype prevalence, cytological interpretation, referral patterns, and healthcare systems 12 , 13 . Before predictive models can be implemented in clinical practice, rigorous external validation in independent populations is essential to assess their generalizability and transportability 14 . External validation is particularly important when applying models to populations with distinct epidemiological and clinical characteristics, such as those encountered in Southeast Asia. Despite this need, data on the external validation of cervical precancer prediction models in the Thai population remain limited. Two predictive models recently developed by Xue et al. and Sheng et al. 12,13 , derived from large Asian cohorts, demonstrated promising discriminative performance in identifying CIN2+/HSIL. These models incorporate routinely available clinical, cytological, HPV, and colposcopic variables, making them potentially applicable to real-world clinical settings. However, their external validity in Thai women who have undergone colposcopy has not yet been formally evaluated. Therefore, this study aimed to externally validate these two predictive models in a cohort of Thai women undergoing colposcopy at a tertiary referral center. Specifically, we assessed the model’s discrimination, calibration, and clinical utility using decision curve analysis. By evaluating model performance in a real-world Thai population with different demographic and clinical characteristics from the original development cohorts, this study sought to determine the transportability and clinical feasibility of these models for risk-based triage and management of women with abnormal cervical cancer screening results. Methods Study Design and Setting The cohort data were routinely collected in clinical practice between January 2022 and December 2024 in accordance with the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) statement 15 . The study was performed at Songklanagarind Hospital, a tertiary referral center in Southern Thailand. The study protocol was approved by the Human Research Ethics Committee of the Faculty of Medicine, Prince of Songkla University (REC.68-007-12-3). The requirement for informed consent to participate was waived by the Ethics Committee due to the retrospective nature of the study and the use of de-identified routinely collected clinical data. The study was conducted in accordance with the Declaration of Helsinki. Study Population Eligible participants were Thai women aged 25–65 years who underwent colposcopy following abnormal cervical cytology or a positive HPV screening result and subsequently underwent histological evaluation by LEEP or cervical conization. Women were excluded if they were pregnant or within six weeks postpartum, had a history of hysterectomy involving removal of the cervix, had previously received treatment for cervical lesions (including conization, cryotherapy, or laser ablation), or had a history of gynecological malignancy. Patients who underwent only endocervical curettage or punch biopsy without excisional treatment were excluded from the study. Outcome Definition The primary outcome was histologically confirmed high-grade cervical lesions, defined as cervical intraepithelial neoplasia grade ≥2 (CIN2+/HSIL), including CIN2, CIN3, carcinoma in situ, adenocarcinoma in situ, and invasive cervical cancer. Histopathological diagnoses were obtained from excisional specimens and classified according to standard pathological criteria. Participants were categorized as having either CIN2+/HSIL or non–CIN2 disease (<CIN2) 16 . Predictor Variables Predictor variables were selected based on the original Xue and Sheng model 12,13 specifications and were obtained from electronic medical records. These included demographic factors (age, parity, gravidity, and menopausal status), cervical cytology results, HPV status and genotype (HPV16, HPV18/45, other high-risk HPV types, and combined infections), and colposcopic findings. The colposcopic variables included the transformation zone type, visibility of the squamocolumnar junction, lesion size, acetowhite epithelium, mosaic or punctation patterns, atypical vessels, and overall colposcopic impression. To ensure consistency, predictors were defined and categorized according to the criteria used in the original model-development studies. Models for External Two previously published multivariable logistic regression–based predictive models for CIN2+/HSIL were externally validated. The models developed by Xue et al. and Sheng et al 12,13 . were applied to the Thai validation cohort without modification or recalibration of the regression coefficients. The predicted probability of CIN2+/HSIL were calculated for each participant using original model equations Statistical Analysis Descriptive statistics were used to summarize the baseline characteristics of the validation cohort and to compare them with those of the original model development cohorts. Categorical variables are expressed as frequencies and percentages, while continuous variables are summarized as means with standard deviations or medians with interquartile ranges, as appropriate. Group comparisons were performed using the chi-square test or Fisher’s exact test for categorical variables and the independent t-test or Mann–Whitney U test for continuous variables. Model discrimination was assessed using the area under the receiver operating characteristic curve (AUROC) with 95% confidence intervals estimated using the DeLong method. Diagnostic indices, including sensitivity, specificity, and Youden’s J index, were calculated at optimal cut-off values. The model calibration was evaluated using calibration plots, calibration-in-the-large, calibration slopes, and Brier scores. Clinical utility was assessed using decision curve analysis by comparing the net benefits of the predictive models with the default strategies of treating all patients or not across a range of threshold probabilities. All statistical analyses were performed using R software (version 4.5.1; R Foundation for Statistical Computing, Vienna, Austria) and Stata software (version 19; StataCorp, College Station, TX, USA). Statistical significance was defined as a two-sided p-value of <0.05. • PROBAST justification In accordance with the Prediction Model Risk of Bias Assessment Tool (PROBAST) 17 , this study was designed to minimize potential sources of bias in external validation. The study population consisted of women who underwent colposcopy with histologically confirmed outcomes, to ensure a well-defined target population for model validation. The predictors were defined and coded to match the original model specifications and were routinely available at the time of clinical decision-making. The outcome (CIN2+/HSIL) was based on the histopathological diagnosis of the excisional specimens, providing a robust and objective reference standard. The published prediction models were applied without modification, and the model performance was evaluated using appropriate measures of discrimination, calibration, and clinical utility, consistent with the PROBAST recommendations. Although the retrospective design and tertiary referral setting may introduce selection-related case-mix differences, these characteristics reflect real-world clinical practice and were explicitly considered in the interpretation of the model performance and applicability. Results Study Population and Baseline Characteristics A total of 536 Thai women who underwent colposcopy of histologically confirmed results were included in the external validation cohort. The baseline demographic and clinical characteristics of the validation cohort, compared with those of the original Xue and Sheng model development cohorts, are summarized in Tables 1A and 1B. This comparison provides an important context for assessing the external validity and transportability of predictive models across populations. Compared with the Xue cohort, women in the validation cohort were more frequently represented in the 30–39 and 40–49-year age groups (31% vs. 37%, p = 0.007, 30.4% vs. 26.6%, p=0.031), whereas the proportion of menopausal women was similar between the cohorts (14.0% vs. 21.8%, p = 0.206). Differences in reproductive characteristics were observed, including a lower proportion of women with high gravidity in the training cohort (4.1% vs. 8.0%, p = 0.005) and a slightly higher proportion of women with parity greater than two in the validation cohort (10.4% vs. 8.0%, p = 0.024). The distribution of HPV status and cytologic categories was largely comparable; however, a higher proportion of women with negative intraepithelial lesion or malignancy (NILM) cytology was observed in the Thai cohort (46.5% vs. 41.8%, p = 0.046), reflecting potential differences in screening strategies and referral criteria. Compared to the Sheng cohort, the validation cohort demonstrated notable differences in HPV genotype distribution, cytological findings, and colposcopic features. Combined high-risk HPV infections (12.6% vs. 9.1%, p <0.001) and ASC-US/AGC cytology (36.1% vs. 20.5%, p = 0.002) were less frequent in the validation cohort, whereas mosaic and punctation findings were more commonly observed (32.5% vs. 17.7%, p <0.001). In contrast, atypical vascular patterns were less prevalent (4.6% vs. 2.4%, P = 0.001). These variations highlight differences in HPV epidemiology, colposcopic interpretation, and case mix across healthcare settings. The prevalence of histologically confirmed CIN2+/HSIL in the validation cohort was 28.7%, compared with 25.1% in the Xue cohort and 11.4% in the Sheng cohort. Despite population-level differences, the validation cohort encompassed a broad spectrum of demographic, cytological, and colposcopic characteristics, providing a robust dataset for external model validation. Multivariable Logistic Regression Analyses Associations between clinical predictors and CIN2+/HSIL were evaluated using univariable and multivariable logistic regression analyses (Tables 2A and 2B). These analyses were conducted to quantify the contribution of individual predictors within the external validation cohort and assess their consistency with the original model derivation study. Multivariable analyses (Xue model) Multivariable logistic regression analyses for the predictors of high-grade cervical lesions (CIN2+/HSIL) based on the Xue model are presented in Table 2A. In the multivariable analysis, ASC-H (adjusted OR 3.78, 95% CI 1.39–10.31) and HSIL cytology (adjusted OR 8.79, 95% CI 3.35–23.02) remained independent predictors of CIN2 diagnosis. Lesion size >1/3–2/3 of the cervical area retained a strong association (adjusted OR 4.05, 95% CI 2.30–7.12), while lesions >2/3 showed borderline significance (adjusted OR 9.28, 95% CI 0.99–87.06). High-grade colposcopic impressions remained a robust predictor (adjusted OR 11.37, 95% CI 2.70–47.92). The model demonstrated a good discriminative ability, with an AUROC of 0.85. HPV genotype categories, including HPV16/18 and other high-risk HPV types, were significantly associated with CIN2+/HSIL in the univariable analysis but did not retain independent significance after multivariable adjustment. Overall, these findings indicate that the core predictors of the Xue model preserved their predictive relevance in an external population, despite differences in baseline characteristics. Multivariable analyses (Sheng model) In the multivariable analysis (Table 2B), HPV16 infection remained an independent predictor (adjusted OR 2.31, 95% CI 1.26–4.24). Cytological abnormalities showed the strongest effects, with ASC-H (adjusted OR 8.89, 95% CI 3.44–23.00) and HSIL/SCC (adjusted OR 20.18, 95% CI 7.67–53.05) demonstrating markedly increased odds of CIN2+/HSIL. Among the colposcopic findings, acetowhite epithelium (adjusted OR 3.32, 95% CI 1.05–10.54), mosaic or punctation (adjusted OR 3.31, 95% CI 2.06–5.32), and atypical vessels (adjusted OR 4.77, 95% CI 1.22–18.67) remained independently associated. The Sheng model showed good discrimination, with an AUROC of 0.80. These results are consistent with the structure of the original Sheng model and support the transportability of its key predictors when applied to a Thai colposcopy population with different epidemiological and clinical characteristics. Overall, both externally validated models demonstrated consistent associations between cytologic severity, HPV genotype, and colposcopic features and the risk of CIN2+/HSIL, supporting their applicability for risk stratification in Thai women undergoing colposcopy. • Diagnostic Performance of Predictive Models The diagnostic performance of the predictive models in the validation cohort is summarized in Tables 3 and 4 and Figure 2. Receiver operating characteristic (ROC) curve analysis demonstrated good discriminative ability for both models (Figure 2). The Xue model achieved an area under the ROC curve (AUROC) of 0.85 (95% CI: 0.81–0.89), while the Sheng model achieved an AUROC of 0.80 (95% CI: 0.77–0.85). These findings indicate that both models effectively distinguished between women with and without CIN2+/HSIL in an external population, with the Xue model demonstrating superior overall discrimination. The diagnostic performances of the predictive models in the external validation cohort is summarized in Tables 3 and 4. At model-specific optimal cutoff thresholds determined by the Youden index (Table 3), the Xue model demonstrated a sensitivity of 74.7% and specificity of 84.0%, yielding a Youden’s J index of 0.59. In comparison, the Sheng model showed a sensitivity of 72.7% and specificity of 75.9%, with a Youden’s J index of 0.49. These findings indicate a more favorable balance between sensitivity and specificity for the Xue model at data-driven optimal thresholds. A marked trade-off between sensitivity and specificity was observed for both models when predefined model-specific probability thresholds were applied (Table 4). The Xue model, which used a higher probability threshold, demonstrated reduced sensitivity (56.5%) but substantially increased specificity (93.7%), whereas the Sheng model, which used a lower threshold, achieved high sensitivity (93.5%) at the expense of low specificity (34.3%). These results highlight how threshold selection substantially influences diagnostic performance and underscore the importance of context-specific threshold choices when applying predictive models to external populations. • Calibration and Overall Model Accuracy The calibration performance is shown in Figure 3. For both models, the predicted probabilities closely aligned with the observed event frequencies across the range of risk estimates. Quantitative calibration metrics demonstrated excellent agreement, with an observed-to-expected ratio of 1.00, a calibration-in-the-large of 0.00, and a calibration slope of 1.00 for both models, indicating no evidence of systematic overestimation or underestimation of risk. The overall predictive accuracy was further supported by Brier scores (Figure 4). The Brier score was 0.127 for the Xue model and 0.147 for the Sheng model, reflecting good overall accuracy of the predicted probabilities. Together, these findings indicate that both models provide reliable absolute risk estimates for CIN2+/HSIL in the external validation cohort. • Clinical Utility Assessed by Decision Curve Analysis The clinical utility of the predictive models was evaluated using a decision curve analysis (Figure 5). Both models demonstrated a positive net clinical benefit across a wide range of threshold probabilities compared to the default strategies for treating all patients. The net benefit was most pronounced across clinically relevant threshold probabilities, approximately between 0.05 and 0.35, corresponding to the decision thresholds commonly used in colposcopy management and excisional treatment. These findings suggest that the application of predictive models in clinical practice could reduce unnecessary procedures while maintaining appropriate detection of high-grade cervical lesions. Overall, the decision curve analysis supports the potential role of these models in individualized risk-based decision-making in Thai women undergoing colposcopy. Discussion This study validated two predictive models for high-grade cervical lesions (CIN2+/HSIL) in a Thai colposcopy population. Despite the demographic and clinical differences from the original cohorts, both models demonstrated good discrimination and clinical benefits, supporting their transportability to populations with distinct characteristics. The findings of this external validation study align with those of previous reports showing good discriminative performance of the prediction models for high-grade cervical lesions. In the development studies by Xue et al. and Sheng et al. 12,13 , both models achieved AUROC values of 0.80–0.90, indicating strong ability to distinguish CIN2+/HSIL from lower grade lesions. The Xue model showed higher sensitivity, whereas the Sheng model demonstrated higher specificity, reflecting variations in the model structure and intended use. In the Thai validation cohort, both models exhibited good discrimination, with AUROC values of 0.85 and 0.80. This consistency across distinct populations suggests that the core predictors - cytological severity, HPV genotype, and colposcopic features - capture the fundamental determinants of high-grade cervical disease. Calibration is a critical and often underreported aspect of model validation. In this study, both models showed good agreement between predicted probabilities and observed outcomes, as shown by the calibration plots, calibration-in-the-large, calibration slope, and Brier score 18 . The absence of systematic overestimation or underestimation indicates that the models produced reliable risk estimates in the validation cohort. This finding is particularly relevant for clinical decision-making, as poorly calibrated models may misinform patient counseling, even when discrimination appears acceptable. A key observation of this study is the strong dependency of diagnostic performance on the probability threshold selection. When model-specific optimal cutoffs were applied, a balanced trade-off between sensitivity and specificity was achieved, whereas predefined thresholds resulted in shifts that favored either sensitivity or specificity 19 . These findings indicate that prediction models should be interpreted as continuous risk estimators rather than binary diagnostic tests. Therefore, threshold selection should be context-specific, considering disease prevalence, healthcare resources, and the consequences of false-positive and false-negative decisions. Rigid adoption of fixed thresholds across populations may undermine the benefits of risk-based approaches. Decision curve analysis further supported both models' clinical utility, demonstrating positive net benefits across relevant threshold probabilities compared with the treat-all and treat-none strategies 20 . This suggests that these models may improve decision-making by aligning interventions with individual risk profiles. This approach is particularly relevant in cervical cancer prevention, where balancing early detection while avoiding unnecessary procedures is central. These findings have implications for HPV-based cervical cancer screening, which is increasingly adopted in Thailand and other low- and middle-income countries 21 . While HPV testing improves sensitivity, it also increases referral rates to colposcopy, straining healthcare systems and exposing women to potentially unnecessary procedures. By integrating clinical, cytological, HPV, and colposcopic variables into individualized risk estimates, validated prediction models may support refined colposcopy triage strategies. This aligns with risk-based management paradigms and may help optimize resource utilization while ensuring patient safety. Colposcopic assessment remains subjective in cervical cancer prevention, with variability related to operator experience and practice patterns. Notably, colposcopic features retained predictive relevance in both models despite inter-observer variability, suggesting that these features capture meaningful disease information in heterogeneous real-world settings 22 . Future studies incorporating standardized colposcopy training, image-based assessments, or AI-assisted interpretation may further enhance colposcopic predictor reliability and improve the model performance. Population characteristics vary across regions and healthcare systems and may influence model performance. Despite these variations, both models demonstrated good calibration and maintained clinical utility in the Thai population, thus underscoring their robustness. Similar observations have been reported in other external validation studies of cervical precancer prediction models 9 , where acceptable performance was preserved despite the differences in screening strategies. These findings reinforce the importance of external validation and suggest that prediction models based on routine clinical variables can be applied across diverse populations. This study had several strengths. It provides rigorous external validation of two predictive models in an independent cohort distinct from the development populations which is- essential for assessing model transportability. All outcomes used histologically confirmed diagnoses from the excisional specimens to ensure reliable reference standards. The evaluation incorporated discrimination, calibration, and clinical utility through decision curve analysis. The models' use of routine clinical, cytological, HPV, and colposcopic variables enhanced their real-world applicability. This study has limitations due to its retrospective design and potential selection bias from the tertiary referral center population with high CIN2+/HSIL prevalence compared with general screening. Some predictors were not uniformly available or had inter-observer variability, particularly in colposcopic assessment, affecting model performance. The study excluded emerging triage biomarkers such as dual-stained p16/Ki-67 cytology. While the overall calibration was acceptable, local recalibration may be required before implementation, and prospective validation is required to evaluate the model-guided decision-making impact. The findings of this external validation study have important clinical implications for the management of women with abnormal cervical cancer screening results. The validated predictive models support a shift from uniform strategies toward risk-based triage in colposcopy by providing individualized CIN2+/HSIL risk estimates. Women with low predicted risk may undergo surveillance, whereas those with a higher risk may need excisional treatment. This approach can reduce unnecessary procedures while maintaining the detection of significant disease. These implications are especially relevant in settings with high HPV-positive referrals and limited colposcopy resources, such as Thailand and other low-income countries. Risk-based decisions using validated prediction models may optimize resources and reduce patient anxiety and procedure-related morbidities. The models used routine clinical, cytological, HPV, and colposcopic variables for easy workflow integration. However, prospective studies are required to confirm improved outcomes in practice. In conclusion, this external validation study demonstrated that the two predictive models for CIN2+/HSIL retain robust performance, good calibration, and meaningful clinical utility when applied to a real-world Thai colposcopy population. These findings support the transportability of risk-based prediction models across populations and highlight their potential role in advancing individualized, context-aware cervical cancer prevention strategies. Declarations Data availability The data supporting the findings of this study are available from the corresponding author (T. Atjimakul), upon reasonable request. Acknowledgements We would like to sincerely thank the Department of Epidemiology, Songklanagarind Hospital, for assisting us with sample size calculation, data collection, and data analysis. Funding This research was supported by (REC.68-007-12-3) the Faculty of Medicine, Prince of Songkla University. Declaration of interest The authors declare no conflicts of interest. Consent for publication Not applicable Author Contribution Statement All authors met the authorship criteria established by the International Committee of Medical Journal Editors (ICMJE). A. Eidbur : Conceptualization, Methodology, Formal analysis, Investigation, Resources, Data curation, Validation, Writing-original draft, Writing-review & Editing T. Atjimakul : Conceptualization, Methodology, Formal analysis, Investigation, Resources, Data curation, Supervision, Software, Validation, Visualization, Writing-original draft, Writing-review & Editing, Writing-original draft, Project administration, Funding acquisition References Bray F, Laversanne M, Sung H, Ferlay J, Siegel RL, Soerjomataram I, et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2024;74(3):229–263. doi:10.3322/caac.21834 Schiffman M, Castle PE, Jeronimo J, Rodriguez AC, Wacholder S. Human papillomavirus and cervical cancer. Lancet. 2007;370(9590):890–907. doi:10.1016/S0140-6736(07)61416-0 World Health Organization. WHO guideline for screening and treatment of cervical pre-cancer lesions for cervical cancer prevention . 2nd ed. Geneva: World Health Organization; 2021. PMID:34314129 Schiffman M, Wentzensen N, Perkins RB, Guido RS. An introduction to the 2019 ASCCP risk-based management consensus guidelines. J Low Genit Tract Dis. 2020;24(2):87–89. doi:10.1097/LGT.0000000000000531 Thai National Cancer Institute. Hospital-based cancer registry 2022 . Medical Record and Database Cancer Unit; 2022 [cited 2025 Oct 14]. Available from: https://www.nci.go.th/e_book/hosbased_2565/index.html Ozturk E, Ince O, Karaca İ, Demirayak G, Comba C, Alay I, et al. The impact of concurrent HPV infections on the presentation of high-grade cervical intraepithelial lesions. Ginekol Pol. 2020;91(6):324–330. doi:10.5603/GP.a2020.0055 Carrillo-García A, Ponce-de-León-Rosales S, Cantú-de-León D, Fragoso-Ontiveros V, Martínez-Ramírez I, Orozco-Colín A, et al. Impact of human papillomavirus coinfections on the risk of high-grade squamous intraepithelial lesion and cervical cancer. Gynecol Oncol. 2014;134(3):534–539. doi:10.1016/j.ygyno.2014.05.033 Kyrgiou M, Athanasiou A, Kalliala IEJ, Paraskevaidi M, Mitra A, Martin-Hirsch PP, et al. Obstetric outcomes after conservative treatment for cervical intraepithelial lesions and early invasive disease. Cochrane Database Syst Rev. 2017;11(11):CD012847. doi:10.1002/14651858.CD012847 Wittenborn J, Kupec T, Iborra S, Stickeler E, Najjari L, Kennes LN. HPV high-risk multiple infection is a key predictor of cervical dysplasia in diagnostic LEEPs: a retrospective cohort analysis. Geburtshilfe Frauenheilkd. 2022;82(12):1387–1396. doi:10.1055/a-1749-6420 Arbyn M, Kyrgiou M, Simoens C, Raifu AO, Koliopoulos G, Martin-Hirsch P, et al. Perinatal mortality and other severe adverse pregnancy outcomes associated with treatment of cervical intraepithelial neoplasia: meta-analysis. BMJ. 2008;337:a1284. doi:10.1136/bmj.a1284 Kyrgiou M, Athanasiou A, Paraskevaidi M, Mitra A, Kalliala I, Martin-Hirsch P, et al. Adverse obstetric outcomes after local treatment for cervical preinvasive and early invasive disease according to cone depth: systematic review and meta-analysis. BMJ. 2016;354:i3633. doi:10.1136/bmj.i3633 Sheng B, Yao D, Du X, Chen D, Zhou L. Establishment and validation of a risk prediction model for high-grade cervical lesions. Eur J Obstet Gynecol Reprod Biol. 2023;281:1–6. doi:10.1016/j.ejogrb.2023.01.003 Xue P, Seery S, Wang S, Jiang Y, Qiao Y. Developing a predictive nomogram for colposcopists: a retrospective, multicenter study of cervical precancer identification in China. BMC Cancer. 2023;23(1):163. doi:10.1186/s12885-023-09919-3 Ramspek CL, Jager KJ, Dekker FW, Zoccali C, van Diepen M. External validation of prognostic models: what, why, how, when, and where? Clin Kidney J. 2020;14(1):49–58. doi:10.1093/ckj/sfz123 Collins GS, Reitsma JB, Altman DG, Moons KGM. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. BMJ. 2015;350:g7594. doi:10.1136/bmj.g7594 Prendiville W, Sankaranarayanan R. Colposcopy and treatment of cervical precancer. In: Colposcopic terminology: the 2011 IFCPC nomenclature . Lyon (FR): International Agency for Research on Cancer; 2017. p. 91–115. (IARC Technical Report No. 45). Wolff RF, Moons KGM, Riley RD, Whiting PF, Westwood M, Collins GS, et al; PROBAST Group. PROBAST: a tool to assess the risk of bias and applicability of prediction model studies. Ann Intern Med. 2019;170(1):51–58. doi:10.7326/M18-1376 Van Calster B, Steyerberg EW, Van Smeden M, McLernon DJ, Wynants L. Calibration: the Achilles heel of predictive analytics. BMC Med. 2019;17(1):230. doi:10.1186/s12916-019-1466-7 Akarachantachote N, Chadcham S, Saithanu K. Cutoff threshold of variable importance in projection for variable selection. Int J Pure Appl Math. 2014;94(3):307–322. doi:10.12732/ijpam.v94i3.2 Zhang Z, Rousson V, Qian X, Chen M, Lee WC, Guo Y, et al. Decision curve analysis: a technical note. Ann Transl Med. 2018;6(15):308. doi:10.21037/atm.2018.07.02 Sahasrabuddhe VV, Parham GP, Vermund SH, Mwanahamuntu MH. Cervical cancer prevention in low- and middle-income countries: feasible, affordable, essential. Cancer Prev Res (Phila). 2012;5(1):11–17. doi:10.1158/1940-6207.CAPR-11-0540 Khan MJ, Einstein MH, Huh WK, Conageski C, Moscicki AB, Waxman AG, et al. ASCCP colposcopy standards: role of colposcopy, benefits, potential harms, and terminology for colposcopic practice. J Low Genit Tract Dis. 2017;21(4):223–229. doi:10.1097/LGT.0000000000000338 Tables Tables 1 to 4 are available in the Supplementary Files section. Additional Declarations No competing interests reported. Supplementary Files Table1A.docx Table1B.docx Table2AFindingsofmultivariablelogisticregression.docx Table2BFindingsofmultivariablelogisticregressionintheShengmodelcohortandValidationcohort.docx Table3.docx Table4.docx Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 28 Apr, 2026 Reviews received at journal 22 Feb, 2026 Reviews received at journal 21 Feb, 2026 Reviews received at journal 21 Feb, 2026 Reviews received at journal 20 Feb, 2026 Reviewers agreed at journal 16 Feb, 2026 Reviewers agreed at journal 15 Feb, 2026 Reviewers agreed at journal 15 Feb, 2026 Reviewers agreed at journal 15 Feb, 2026 Reviewers agreed at journal 10 Feb, 2026 Reviewers invited by journal 09 Feb, 2026 Editor assigned by journal 06 Feb, 2026 Editor invited by journal 19 Jan, 2026 Submission checks completed at journal 18 Jan, 2026 First submitted to journal 18 Jan, 2026 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-8615288","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":589412334,"identity":"45e36ac3-8b5b-4ea2-99d0-af3105e4010b","order_by":0,"name":"Atikom Eidbur","email":"","orcid":"","institution":"Prince of Songkla University","correspondingAuthor":false,"prefix":"","firstName":"Atikom","middleName":"","lastName":"Eidbur","suffix":""},{"id":589412335,"identity":"82eda1c8-00bd-4f78-b894-3daea34109e8","order_by":1,"name":"Thiti Atjimakul","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA90lEQVRIiWNgGAWjYDACCcYGhgQGBmY2IGZgKLABihDQwSPB2NiQwGAA1WKQRowWBqA1DAYgNkjLYcJa7KWb2x88qPnDzid2+LExj8H5xP7ZzQcfMNTYROO0ReYg0GHHgA6TTjNO5jG4nTjjzrFkA4ZjabkNOB2WCNTCBtKSYHwYpKXhRo4ZMEwOE9DyD6Ql/TNQy7nE+URpSWwDackBOexA4gaCWm4kNs5I7DMGaSk2nGOQbLzxRlqyQQIev7DPSH/w8cc3uWT52embJd5U2MnOu5F88MGHGhucWmAgGcZwBKtMIKAcBOxgDHsiFI+CUTAKRsEIAwD42VdgtwcKBAAAAABJRU5ErkJggg==","orcid":"","institution":"Prince of Songkla University","correspondingAuthor":true,"prefix":"","firstName":"Thiti","middleName":"","lastName":"Atjimakul","suffix":""}],"badges":[],"createdAt":"2026-01-16 04:38:30","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8615288/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8615288/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":102746859,"identity":"bc9d6158-4d3a-4537-9911-138828379a22","added_by":"auto","created_at":"2026-02-16 09:02:12","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":176250,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFlowchart for the validation of predictive model\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-8615288/v1/29f44d9595288401c5432edf.png"},{"id":102516075,"identity":"d511db8b-d32e-48e6-bd88-5a4da28909cc","added_by":"auto","created_at":"2026-02-12 13:47:25","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":553040,"visible":true,"origin":"","legend":"\u003cp\u003eReceiver Operating Characteristic (ROC) curve for the predictive model which validated with Xue model (2A) and Sheng model (2B)\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-8615288/v1/8d86ec870424414ae45462d1.png"},{"id":102516073,"identity":"28cb4534-24d7-477a-b56e-1e787ee6ab90","added_by":"auto","created_at":"2026-02-12 13:47:25","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":300068,"visible":true,"origin":"","legend":"\u003cp\u003eCalibration plot for the predictive model to develop observed frequency and predicted probability for the predictive model in validation set with Xue (3A) and Sheng model (3B)\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-8615288/v1/69f5ddf3e3e12a944d6e5e7a.png"},{"id":102516078,"identity":"a3397f32-0a63-4547-bb11-331905201c00","added_by":"auto","created_at":"2026-02-12 13:47:25","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":142091,"visible":true,"origin":"","legend":"\u003cp\u003eBrier score and calibration performance of the predictive model for high-grade cervical lesion (CIN2 ) in validation set with Xue (4A) and Sheng model (4B)\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-8615288/v1/e2854b19d0a4fa8bbf3fd117.png"},{"id":102750712,"identity":"fecb91ee-2655-4c27-abc7-bc27b483d615","added_by":"auto","created_at":"2026-02-16 09:21:40","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1529903,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8615288/v1/11ae25d7-1aff-497f-b74b-c5a9f1b87a6d.pdf"},{"id":102747082,"identity":"8142cb83-85e3-4a8f-8fa5-0a4dc91aa56c","added_by":"auto","created_at":"2026-02-16 09:03:46","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":24526,"visible":true,"origin":"","legend":"","description":"","filename":"Table1A.docx","url":"https://assets-eu.researchsquare.com/files/rs-8615288/v1/d3db3fdbe04161f1431283ff.docx"},{"id":102746250,"identity":"10e2b25e-12b0-4330-88fd-d60eec926e26","added_by":"auto","created_at":"2026-02-16 08:56:17","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":22671,"visible":true,"origin":"","legend":"","description":"","filename":"Table1B.docx","url":"https://assets-eu.researchsquare.com/files/rs-8615288/v1/4b6daedc54669e22b0dd80c3.docx"},{"id":102746385,"identity":"bfbf9e3d-f518-45a4-8689-9e5af51810c7","added_by":"auto","created_at":"2026-02-16 08:57:14","extension":"docx","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":19921,"visible":true,"origin":"","legend":"","description":"","filename":"Table2AFindingsofmultivariablelogisticregression.docx","url":"https://assets-eu.researchsquare.com/files/rs-8615288/v1/78c08d7a9c5848f1b2b721a5.docx"},{"id":102516079,"identity":"372bcd8b-86cc-4613-9b0b-8b5890fa7709","added_by":"auto","created_at":"2026-02-12 13:47:26","extension":"docx","order_by":4,"title":"","display":"","copyAsset":false,"role":"supplement","size":18245,"visible":true,"origin":"","legend":"","description":"","filename":"Table2BFindingsofmultivariablelogisticregressionintheShengmodelcohortandValidationcohort.docx","url":"https://assets-eu.researchsquare.com/files/rs-8615288/v1/37f49a4ac0e4b4e3571a3668.docx"},{"id":102746323,"identity":"102797d8-9c07-4284-8507-4f2ce250e06d","added_by":"auto","created_at":"2026-02-16 08:56:43","extension":"docx","order_by":5,"title":"","display":"","copyAsset":false,"role":"supplement","size":15763,"visible":true,"origin":"","legend":"","description":"","filename":"Table3.docx","url":"https://assets-eu.researchsquare.com/files/rs-8615288/v1/59092a6e80a3955cc04e4d4a.docx"},{"id":102516077,"identity":"3a112de5-e87d-435a-ab85-ef43538de329","added_by":"auto","created_at":"2026-02-12 13:47:25","extension":"docx","order_by":6,"title":"","display":"","copyAsset":false,"role":"supplement","size":15771,"visible":true,"origin":"","legend":"","description":"","filename":"Table4.docx","url":"https://assets-eu.researchsquare.com/files/rs-8615288/v1/9f5fd31454c5633a659c43e7.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"\u003cp\u003eExternal Validation of Predictive Models for High-Grade Cervical Lesions in Thai Women Undergoing Colposcopy\u003c/p\u003e","fulltext":[{"header":"Background","content":"\u003cp\u003eCervical cancer remains a major public health problem and one of the leading causes of cancer-related morbidity and mortality in women worldwide\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. Persistent infection with high-risk human papillomavirus (HPV) is the principal etiological factor for the development of cervical precancerous lesions and invasive cervical cancer\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e. The implementation of cervical cancer screening programs, particularly HPV-based screening, has substantially reduced cervical cancer incidence and mortality rates of cervical cancer in many regions\u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e,\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e. Nevertheless, effective triage of women with abnormal screening results remains a significant clinical challenge.\u003c/p\u003e \u003cp\u003eIn Thailand, cervical cancer is the second most common cancer among women\u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e, and national screening programs increasingly rely on human papillomavirus (HPV) testing, including both clinician-collected and self-collected samples. Although HPV-based screening has improved the sensitivity for detecting at-risk women, it has also resulted in a growing number of referrals for colposcopy. Importantly, only a minority of women referred for colposcopy are ultimately diagnosed with high-grade cervical lesions (CIN2+/HSIL) on histopathological examination. Recent Thai data indicate that approximately one-quarter of women undergoing colposcopy following a positive HPV test have CIN2+/HSIL, highlighting the substantial burden of unnecessary diagnostic procedures and potential overtreatment\u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eThe loop electrosurgical excision procedure (LEEP) is one of the most commonly performed diagnostic and therapeutic interventions for high-grade cervical lesions\u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e,\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e. Although effective, LEEP is associated with short- and long-term complications, including bleeding, infection, cervical stenosis, and adverse reproductive outcomes, such as preterm birth and cervical insufficiency\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e. These risks are particularly relevant in women of reproductive age and underscore the importance of accurately identifying those who would benefit most from excisional treatment while avoiding unnecessary procedures in low-risk individuals\u003csup\u003e\u003cspan additionalcitationids=\"CR10\" citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eTo address this challenge, several multivariable predictive models have been developed to estimate the individual risk of CIN2+/HSIL using combinations of demographic characteristics, cytological findings, HPV genotype information, and colposcopic features. These models aim to support risk-based decision-making and align with contemporary management strategies advocated by international guidelines\u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e. However, the majority of existing models have been developed in Western or East Asian populations, and their performance may be influenced by differences in HPV genotype prevalence, cytological interpretation, referral patterns, and healthcare systems\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e,\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eBefore predictive models can be implemented in clinical practice, rigorous external validation in independent populations is essential to assess their generalizability and transportability\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e. External validation is particularly important when applying models to populations with distinct epidemiological and clinical characteristics, such as those encountered in Southeast Asia. Despite this need, data on the external validation of cervical precancer prediction models in the Thai population remain limited.\u003c/p\u003e \u003cp\u003eTwo predictive models recently developed by Xue et al. and Sheng et al.\u003csup\u003e12,13\u003c/sup\u003e, derived from large Asian cohorts, demonstrated promising discriminative performance in identifying CIN2+/HSIL. These models incorporate routinely available clinical, cytological, HPV, and colposcopic variables, making them potentially applicable to real-world clinical settings. However, their external validity in Thai women who have undergone colposcopy has not yet been formally evaluated.\u003c/p\u003e \u003cp\u003eTherefore, this study aimed to externally validate these two predictive models in a cohort of Thai women undergoing colposcopy at a tertiary referral center. Specifically, we assessed the model\u0026rsquo;s discrimination, calibration, and clinical utility using decision curve analysis. By evaluating model performance in a real-world Thai population with different demographic and clinical characteristics from the original development cohorts, this study sought to determine the transportability and clinical feasibility of these models for risk-based triage and management of women with abnormal cervical cancer screening results.\u003c/p\u003e"},{"header":"Methods","content":"\u003cul\u003e\n \u003cli\u003e\u003cstrong\u003eStudy Design and Setting\u003c/strong\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eThe cohort data were routinely collected in clinical\u0026nbsp;practice between January 2022 and December 2024 in accordance with the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) statement\u003csup\u003e15\u003c/sup\u003e. The study was performed at Songklanagarind Hospital, a tertiary referral center in Southern Thailand. The study protocol was approved by the Human Research Ethics Committee of the Faculty of Medicine, Prince of Songkla University (REC.68-007-12-3). The requirement for informed consent to participate was waived by the Ethics Committee due to the retrospective nature of the study and the use of de-identified routinely collected clinical data. The study was conducted in accordance with the Declaration of Helsinki.\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003e\u003cstrong\u003eStudy Population\u003c/strong\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eEligible participants were Thai women aged 25\u0026ndash;65 years who underwent colposcopy\u0026nbsp;\u003c/p\u003e\n\u003cp\u003efollowing abnormal cervical cytology or a positive HPV screening result and subsequently underwent histological evaluation by LEEP or cervical conization. Women were excluded if they were pregnant or within six weeks postpartum, had a history of hysterectomy involving removal of the cervix, had previously received treatment for cervical lesions (including conization, cryotherapy, or laser ablation), or had a history of gynecological malignancy. Patients who underwent only endocervical curettage or punch biopsy without excisional treatment were excluded from the study.\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003e\u003cstrong\u003eOutcome Definition\u003c/strong\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eThe primary outcome was histologically confirmed high-grade cervical lesions, defined as cervical intraepithelial neoplasia grade\u0026nbsp;\u0026ge;2 (CIN2+/HSIL), including CIN2, CIN3, carcinoma in situ, adenocarcinoma in situ, and invasive cervical cancer. Histopathological diagnoses were obtained from excisional specimens and classified according to standard pathological criteria. Participants were categorized as having either CIN2+/HSIL or non\u0026ndash;CIN2 disease (\u0026lt;CIN2)\u003csup\u003e16\u003c/sup\u003e.\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003e\u003cstrong\u003ePredictor Variables\u003c/strong\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003ePredictor variables were selected based on the original Xue and Sheng model\u003csup\u003e12,13\u003c/sup\u003e specifications and were obtained from electronic medical records. These included demographic factors (age, parity, gravidity, and menopausal status), cervical cytology results, HPV status and genotype (HPV16, HPV18/45, other high-risk HPV types, and combined infections), and colposcopic findings. The colposcopic variables included the transformation zone type, visibility of the squamocolumnar junction, lesion size, acetowhite epithelium, mosaic or punctation patterns, atypical vessels, and overall colposcopic impression. To ensure consistency, predictors were defined and categorized according to the criteria used in the original model-development studies.\u0026nbsp;\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003e\u003cstrong\u003eModels for External\u003c/strong\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eTwo previously published multivariable logistic regression\u0026ndash;based predictive models for CIN2+/HSIL were externally validated. The models developed by Xue et al. and Sheng et al\u003csup\u003e12,13\u003c/sup\u003e. were applied to the Thai validation cohort without modification or recalibration of the regression coefficients. The predicted probability of CIN2+/HSIL were calculated for each participant using original model equations\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003e\u003cstrong\u003eStatistical Analysis\u003c/strong\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eDescriptive statistics were used to summarize the baseline characteristics of the validation cohort and to compare them with those of the original model development cohorts. Categorical variables are expressed as frequencies and percentages, while continuous variables are summarized as means with standard deviations or medians with interquartile ranges, as appropriate. Group comparisons were performed using the chi-square test or Fisher\u0026rsquo;s exact test for categorical variables and the independent t-test or Mann\u0026ndash;Whitney U test for continuous variables.\u003c/p\u003e\n\u003cp\u003eModel discrimination was assessed using the area under the receiver operating characteristic curve (AUROC) with 95% confidence intervals estimated using the DeLong method. Diagnostic indices, including sensitivity, specificity, and Youden\u0026rsquo;s J index, were calculated at optimal cut-off values. The model calibration was evaluated using calibration plots, calibration-in-the-large, calibration slopes, and Brier scores. Clinical utility was assessed using decision curve analysis by comparing the net benefits of the predictive models with the default strategies of treating all patients or not across a range of threshold probabilities.\u003c/p\u003e\n\u003cp\u003eAll statistical analyses were performed using R software (version 4.5.1; R Foundation for Statistical Computing, Vienna, Austria) and Stata software (version 19; StataCorp, College Station, TX, USA). Statistical significance was defined as a two-sided p-value of \u0026lt;0.05.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026bull;\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;PROBAST justification\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn accordance with the Prediction Model Risk of Bias Assessment Tool (PROBAST)\u003csup\u003e17\u003c/sup\u003e, this study was designed to minimize potential sources of bias in external validation. The study population consisted of women who underwent colposcopy with histologically confirmed outcomes, to ensure a well-defined target population for model validation. The predictors were defined and coded to match the original model specifications and were routinely available at the time of clinical decision-making. The outcome (CIN2+/HSIL) was based on the histopathological diagnosis of the excisional specimens, providing a robust and objective reference standard. The published prediction models were applied without modification, and the model performance was evaluated using appropriate measures of discrimination, calibration, and clinical utility, consistent with the PROBAST recommendations. Although the retrospective design and tertiary referral setting may introduce selection-related case-mix differences, these characteristics reflect real-world clinical practice and were explicitly considered in the interpretation of the model performance and applicability.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cstrong\u003eStudy Population and Baseline Characteristics\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eA total of 536 Thai women who underwent colposcopy of histologically confirmed results were included in the external validation cohort. The baseline demographic and clinical characteristics of the validation cohort, compared with those of the original Xue and Sheng model development cohorts, are summarized in Tables 1A and 1B. This comparison provides an important context for assessing the external validity and transportability of predictive models across populations.\u003c/p\u003e\n\u003cp\u003eCompared with the Xue cohort, women in the validation cohort were more frequently represented in the 30\u0026ndash;39 and 40\u0026ndash;49-year age groups (31% vs. 37%, p = 0.007, 30.4% vs. 26.6%, p=0.031), whereas the proportion of menopausal women was similar between the cohorts (14.0% vs. 21.8%, p = 0.206). Differences in reproductive characteristics were observed, including a lower proportion of women with high gravidity in the training cohort (4.1% vs. 8.0%, p = 0.005) and a slightly higher proportion of women with parity greater than two in the validation cohort (10.4% vs. 8.0%, p = 0.024). The distribution of HPV status and cytologic categories was largely comparable; however, a higher proportion of women with negative intraepithelial lesion or malignancy (NILM) cytology was observed in the Thai cohort (46.5% vs. 41.8%, p = 0.046), reflecting potential differences in screening strategies and referral criteria.\u003c/p\u003e\n\u003cp\u003eCompared to the Sheng cohort, the validation cohort demonstrated notable differences in HPV genotype distribution, cytological findings, and colposcopic features. Combined high-risk HPV infections (12.6% vs. 9.1%, p \u0026lt;0.001) and ASC-US/AGC cytology (36.1% vs. 20.5%, p = 0.002) were less frequent in the validation cohort, whereas mosaic and punctation findings were more commonly observed (32.5% vs. 17.7%, p \u0026lt;0.001). In contrast, atypical vascular patterns were less prevalent (4.6% vs. 2.4%, P = 0.001). These variations highlight differences in HPV epidemiology, colposcopic interpretation, and case mix across healthcare settings.\u003c/p\u003e\n\u003cp\u003eThe prevalence of histologically confirmed CIN2+/HSIL in the validation cohort was 28.7%, compared with 25.1% in the Xue cohort and 11.4% in the Sheng cohort. Despite population-level differences, the validation cohort encompassed a broad spectrum of demographic, cytological, and colposcopic characteristics, providing a robust dataset for external model validation.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMultivariable Logistic Regression Analyses\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAssociations between clinical predictors and CIN2+/HSIL were evaluated using univariable and multivariable logistic regression analyses (Tables 2A and 2B). These analyses were conducted to quantify the contribution of individual predictors within the external validation cohort and assess their consistency with the original model derivation study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMultivariable analyses (Xue model)\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eMultivariable logistic regression analyses for the predictors of high-grade cervical lesions (CIN2+/HSIL) based on the Xue model are presented in Table 2A.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn the multivariable analysis, ASC-H (adjusted OR 3.78, 95% CI 1.39\u0026ndash;10.31) and HSIL cytology (adjusted OR 8.79, 95% CI 3.35\u0026ndash;23.02) remained independent predictors of CIN2 diagnosis. Lesion size \u0026gt;1/3\u0026ndash;2/3 of the cervical area retained a strong association (adjusted OR 4.05, 95% CI 2.30\u0026ndash;7.12), while lesions \u0026gt;2/3 showed borderline significance (adjusted OR 9.28, 95% CI 0.99\u0026ndash;87.06). High-grade colposcopic impressions remained a robust predictor (adjusted OR 11.37, 95% CI 2.70\u0026ndash;47.92). The model demonstrated a good discriminative ability, with an AUROC of 0.85. HPV genotype categories, including HPV16/18 and other high-risk HPV types, were significantly associated with CIN2+/HSIL in the univariable analysis but did not retain independent significance after multivariable adjustment. Overall, these findings indicate that the core predictors of the Xue model preserved their predictive relevance in an external population, despite differences in baseline characteristics.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMultivariable analyses (Sheng model)\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn the multivariable analysis (Table 2B), HPV16 infection remained an independent predictor (adjusted OR 2.31, 95% CI 1.26\u0026ndash;4.24). Cytological abnormalities showed the strongest effects, with ASC-H (adjusted OR 8.89, 95% CI 3.44\u0026ndash;23.00) and HSIL/SCC (adjusted OR 20.18, 95% CI 7.67\u0026ndash;53.05) demonstrating markedly increased odds of CIN2+/HSIL. Among the colposcopic findings, acetowhite epithelium (adjusted OR 3.32, 95% CI 1.05\u0026ndash;10.54), mosaic or punctation (adjusted OR 3.31, 95% CI 2.06\u0026ndash;5.32), and atypical vessels (adjusted OR 4.77, 95% CI 1.22\u0026ndash;18.67) remained independently associated. The Sheng model showed good discrimination, with an AUROC of 0.80. These results are consistent with the structure of the original Sheng model and support the transportability of its key predictors when applied to a Thai colposcopy population with different epidemiological and clinical characteristics.\u003c/p\u003e\n\u003cp\u003eOverall, both externally validated models demonstrated consistent associations between cytologic severity, HPV genotype, and colposcopic features and the risk of CIN2+/HSIL, supporting their applicability for risk stratification in Thai women undergoing colposcopy.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026bull; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;Diagnostic Performance of Predictive Models\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe diagnostic performance of the predictive models in the validation cohort is summarized in Tables 3 and 4 and Figure 2.\u003c/p\u003e\n\u003cp\u003eReceiver operating characteristic (ROC) curve analysis demonstrated good discriminative ability for both models (Figure 2). The Xue model achieved an area under the ROC curve (AUROC) of 0.85 (95% CI: 0.81\u0026ndash;0.89), while the Sheng model achieved an AUROC of 0.80 (95% CI: 0.77\u0026ndash;0.85). These findings indicate that both models effectively distinguished between women with and without CIN2+/HSIL in an external population, with the Xue model demonstrating superior overall discrimination.\u003c/p\u003e\n\u003cp\u003eThe diagnostic performances of the predictive models in the external validation cohort is summarized in Tables 3 and 4. At model-specific optimal cutoff thresholds determined by the Youden index (Table 3), the Xue model demonstrated a sensitivity of 74.7% and specificity of 84.0%, yielding a Youden\u0026rsquo;s J index of 0.59. In comparison, the Sheng model showed a sensitivity of 72.7% and specificity of 75.9%, with a Youden\u0026rsquo;s J index of 0.49. These findings indicate a more favorable balance between sensitivity and specificity for the Xue model at data-driven optimal thresholds.\u003c/p\u003e\n\u003cp\u003eA marked trade-off between sensitivity and specificity was observed for both models when predefined model-specific probability thresholds were applied (Table 4). The Xue model, which used a higher probability threshold, demonstrated reduced sensitivity (56.5%) but substantially increased specificity (93.7%), whereas the Sheng model, which used a lower threshold, achieved high sensitivity (93.5%) at the expense of low specificity (34.3%). These results highlight how threshold selection substantially influences diagnostic performance and underscore the importance of context-specific threshold choices when applying predictive models to external populations.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026bull;\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;Calibration and Overall Model Accuracy\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe calibration performance is shown in Figure 3. For both models, the predicted probabilities closely aligned with the observed event frequencies across the range of risk estimates. Quantitative calibration metrics demonstrated excellent agreement, with an observed-to-expected ratio of 1.00, a calibration-in-the-large of 0.00, and a calibration slope of 1.00 for both models, indicating no evidence of systematic overestimation or underestimation of risk.\u003c/p\u003e\n\u003cp\u003eThe overall predictive accuracy was further supported by Brier scores (Figure 4). The Brier score was 0.127 for the Xue model and 0.147 for the Sheng model, reflecting good overall accuracy of the predicted probabilities. Together, these findings indicate that both models provide reliable absolute risk estimates for CIN2+/HSIL in the external validation cohort.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026bull;\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;Clinical Utility Assessed by Decision Curve Analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe clinical utility of the predictive models was evaluated using a decision curve analysis (Figure 5). Both models demonstrated a positive net clinical benefit across a wide range of threshold probabilities compared to the default strategies for treating all patients. The net benefit was most pronounced across clinically relevant threshold probabilities, approximately between 0.05 and 0.35, corresponding to the decision thresholds commonly used in colposcopy management and excisional treatment.\u003c/p\u003e\n\u003cp\u003eThese findings suggest that the application of predictive models in clinical practice could reduce unnecessary procedures while maintaining appropriate detection of high-grade cervical lesions. Overall, the decision curve analysis supports the potential role of these models in individualized risk-based decision-making in Thai women undergoing colposcopy.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThis study validated two predictive models for high-grade cervical lesions (CIN2+/HSIL) in a Thai colposcopy population. Despite the demographic and clinical differences from the original cohorts, both models demonstrated good discrimination and clinical benefits, supporting their transportability to populations with distinct characteristics.\u003c/p\u003e\u003cp\u003eThe findings of this external validation study align with those of previous reports showing good discriminative performance of the prediction models for high-grade cervical lesions. In the development studies by Xue et al. and Sheng et al.\u003csup\u003e12,13\u003c/sup\u003e, both models achieved AUROC values of 0.80\u0026ndash;0.90, indicating strong ability to distinguish CIN2+/HSIL from lower grade lesions. The Xue model showed higher sensitivity, whereas the Sheng model demonstrated higher specificity, reflecting variations in the model structure and intended use. In the Thai validation cohort, both models exhibited good discrimination, with AUROC values of 0.85 and 0.80. This consistency across distinct populations suggests that the core predictors - cytological severity, HPV genotype, and colposcopic features - capture the fundamental determinants of high-grade cervical disease.\u003c/p\u003e\u003cp\u003eCalibration is a critical and often underreported aspect of model validation. In this study, both models showed good agreement between predicted probabilities and observed outcomes, as shown by the calibration plots, calibration-in-the-large, calibration slope, and Brier score\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e. The absence of systematic overestimation or underestimation indicates that the models produced reliable risk estimates in the validation cohort. This finding is particularly relevant for clinical decision-making, as poorly calibrated models may misinform patient counseling, even when discrimination appears acceptable.\u003c/p\u003e\u003cp\u003eA key observation of this study is the strong dependency of diagnostic performance on the probability threshold selection. When model-specific optimal cutoffs were applied, a balanced trade-off between sensitivity and specificity was achieved, whereas predefined thresholds resulted in shifts that favored either sensitivity or specificity\u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e. These findings indicate that prediction models should be interpreted as continuous risk estimators rather than binary diagnostic tests. Therefore, threshold selection should be context-specific, considering disease prevalence, healthcare resources, and the consequences of false-positive and false-negative decisions. Rigid adoption of fixed thresholds across populations may undermine the benefits of risk-based approaches.\u003c/p\u003e\u003cp\u003eDecision curve analysis further supported both models' clinical utility, demonstrating positive net benefits across relevant threshold probabilities compared with the treat-all and treat-none strategies\u003csup\u003e\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e. This suggests that these models may improve decision-making by aligning interventions with individual risk profiles. This approach is particularly relevant in cervical cancer prevention, where balancing early detection while avoiding unnecessary procedures is central.\u003c/p\u003e\u003cp\u003eThese findings have implications for HPV-based cervical cancer screening, which is increasingly adopted in Thailand and other low- and middle-income countries\u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e. While HPV testing improves sensitivity, it also increases referral rates to colposcopy, straining healthcare systems and exposing women to potentially unnecessary procedures. By integrating clinical, cytological, HPV, and colposcopic variables into individualized risk estimates, validated prediction models may support refined colposcopy triage strategies. This aligns with risk-based management paradigms and may help optimize resource utilization while ensuring patient safety.\u003c/p\u003e\u003cp\u003eColposcopic assessment remains subjective in cervical cancer prevention, with variability related to operator experience and practice patterns. Notably, colposcopic features retained predictive relevance in both models despite inter-observer variability, suggesting that these features capture meaningful disease information in heterogeneous real-world settings\u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e. Future studies incorporating standardized colposcopy training, image-based assessments, or AI-assisted interpretation may further enhance colposcopic predictor reliability and improve the model performance.\u003c/p\u003e\u003cp\u003ePopulation characteristics vary across regions and healthcare systems and may influence model performance. Despite these variations, both models demonstrated good calibration and maintained clinical utility in the Thai population, thus underscoring their robustness. Similar observations have been reported in other external validation studies of cervical precancer prediction models\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e, where acceptable performance was preserved despite the differences in screening strategies. These findings reinforce the importance of external validation and suggest that prediction models based on routine clinical variables can be applied across diverse populations.\u003c/p\u003e\u003cp\u003eThis study had several strengths. It provides rigorous external validation of two predictive models in an independent cohort distinct from the development populations which is- essential for assessing model transportability. All outcomes used histologically confirmed diagnoses from the excisional specimens to ensure reliable reference standards. The evaluation incorporated discrimination, calibration, and clinical utility through decision curve analysis. The models' use of routine clinical, cytological, HPV, and colposcopic variables enhanced their real-world applicability.\u003c/p\u003e\u003cp\u003eThis study has limitations due to its retrospective design and potential selection bias from the tertiary referral center population with high CIN2+/HSIL prevalence compared with general screening. Some predictors were not uniformly available or had inter-observer variability, particularly in colposcopic assessment, affecting model performance. The study excluded emerging triage biomarkers such as dual-stained p16/Ki-67 cytology. While the overall calibration was acceptable, local recalibration may be required before implementation, and prospective validation is required to evaluate the model-guided decision-making impact.\u003c/p\u003e\u003cp\u003eThe findings of this external validation study have important clinical implications for the management of women with abnormal cervical cancer screening results. The validated predictive models support a shift from uniform strategies toward risk-based triage in colposcopy by providing individualized CIN2+/HSIL risk estimates. Women with low predicted risk may undergo surveillance, whereas those with a higher risk may need excisional treatment. This approach can reduce unnecessary procedures while maintaining the detection of significant disease. These implications are especially relevant in settings with high HPV-positive referrals and limited colposcopy resources, such as Thailand and other low-income countries. Risk-based decisions using validated prediction models may optimize resources and reduce patient anxiety and procedure-related morbidities. The models used routine clinical, cytological, HPV, and colposcopic variables for easy workflow integration. However, prospective studies are required to confirm improved outcomes in practice.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eIn conclusion, this external validation study demonstrated that the two predictive models for CIN2+/HSIL retain robust performance, good calibration, and meaningful clinical utility when applied to a real-world Thai colposcopy population. These findings support the transportability of risk-based prediction models across populations and highlight their potential role in advancing individualized, context-aware cervical cancer prevention strategies.\u003c/p\u003e "},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe data supporting the findings of this study are available from the corresponding author (T.\u0026nbsp;Atjimakul), upon reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe would like to sincerely thank the Department of Epidemiology, Songklanagarind Hospital, for assisting us with sample size calculation, data collection, and data analysis.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis research was supported by (REC.68-007-12-3) the Faculty of Medicine, Prince of Songkla University.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDeclaration of interest\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no conflicts of interest.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contribution Statement\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll authors met the authorship criteria established by the International Committee of Medical Journal Editors (ICMJE).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eA. Eidbur :\u003c/strong\u003e Conceptualization, Methodology, Formal analysis, Investigation, Resources, Data curation, Validation, Writing-original draft, Writing-review \u0026amp; Editing\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eT.\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003eAtjimakul\u003c/strong\u003e\u003cstrong\u003e:\u003c/strong\u003e Conceptualization, Methodology, Formal analysis, Investigation, Resources, Data curation, Supervision, Software, Validation, Visualization, Writing-original draft, Writing-review \u0026amp; Editing, Writing-original draft, Project administration, Funding acquisition\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eBray F, Laversanne M, Sung H, Ferlay J, Siegel RL, Soerjomataram I, et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. \u003cem\u003eCA Cancer J Clin.\u003c/em\u003e 2024;74(3):229\u0026ndash;263. doi:10.3322/caac.21834\u003c/li\u003e\n \u003cli\u003eSchiffman M, Castle PE, Jeronimo J, Rodriguez AC, Wacholder S. Human papillomavirus and cervical cancer. \u003cem\u003eLancet.\u003c/em\u003e 2007;370(9590):890\u0026ndash;907. doi:10.1016/S0140-6736(07)61416-0\u003c/li\u003e\n \u003cli\u003eWorld Health Organization. \u003cem\u003eWHO guideline for screening and treatment of cervical pre-cancer lesions for cervical cancer prevention\u003c/em\u003e. 2nd ed. Geneva: World Health Organization; 2021. PMID:34314129\u003c/li\u003e\n \u003cli\u003eSchiffman M, Wentzensen N, Perkins RB, Guido RS. An introduction to the 2019 ASCCP risk-based management consensus guidelines. \u003cem\u003eJ Low Genit Tract Dis.\u003c/em\u003e 2020;24(2):87\u0026ndash;89. doi:10.1097/LGT.0000000000000531\u003c/li\u003e\n \u003cli\u003eThai National Cancer Institute. \u003cem\u003eHospital-based cancer registry 2022\u003c/em\u003e. Medical Record and Database Cancer Unit; 2022 [cited 2025 Oct 14]. Available from: https://www.nci.go.th/e_book/hosbased_2565/index.html\u003c/li\u003e\n \u003cli\u003eOzturk E, Ince O, Karaca İ, Demirayak G, Comba C, Alay I, et al. The impact of concurrent HPV infections on the presentation of high-grade cervical intraepithelial lesions. \u003cem\u003eGinekol Pol.\u003c/em\u003e 2020;91(6):324\u0026ndash;330. doi:10.5603/GP.a2020.0055\u003c/li\u003e\n \u003cli\u003eCarrillo-Garc\u0026iacute;a A, Ponce-de-Le\u0026oacute;n-Rosales S, Cant\u0026uacute;-de-Le\u0026oacute;n D, Fragoso-Ontiveros V, Mart\u0026iacute;nez-Ram\u0026iacute;rez I, Orozco-Col\u0026iacute;n A, et al. Impact of human papillomavirus coinfections on the risk of high-grade squamous intraepithelial lesion and cervical cancer. \u003cem\u003eGynecol Oncol.\u003c/em\u003e 2014;134(3):534\u0026ndash;539. doi:10.1016/j.ygyno.2014.05.033\u003c/li\u003e\n \u003cli\u003eKyrgiou M, Athanasiou A, Kalliala IEJ, Paraskevaidi M, Mitra A, Martin-Hirsch PP, et al. Obstetric outcomes after conservative treatment for cervical intraepithelial lesions and early invasive disease. \u003cem\u003eCochrane Database Syst Rev.\u003c/em\u003e 2017;11(11):CD012847. doi:10.1002/14651858.CD012847\u003c/li\u003e\n \u003cli\u003eWittenborn J, Kupec T, Iborra S, Stickeler E, Najjari L, Kennes LN. HPV high-risk multiple infection is a key predictor of cervical dysplasia in diagnostic LEEPs: a retrospective cohort analysis. \u003cem\u003eGeburtshilfe Frauenheilkd.\u003c/em\u003e 2022;82(12):1387\u0026ndash;1396. doi:10.1055/a-1749-6420\u003c/li\u003e\n \u003cli\u003eArbyn M, Kyrgiou M, Simoens C, Raifu AO, Koliopoulos G, Martin-Hirsch P, et al. Perinatal mortality and other severe adverse pregnancy outcomes associated with treatment of cervical intraepithelial neoplasia: meta-analysis. \u003cem\u003eBMJ.\u003c/em\u003e 2008;337:a1284. doi:10.1136/bmj.a1284\u003c/li\u003e\n \u003cli\u003eKyrgiou M, Athanasiou A, Paraskevaidi M, Mitra A, Kalliala I, Martin-Hirsch P, et al. Adverse obstetric outcomes after local treatment for cervical preinvasive and early invasive disease according to cone depth: systematic review and meta-analysis. \u003cem\u003eBMJ.\u003c/em\u003e 2016;354:i3633. doi:10.1136/bmj.i3633\u003c/li\u003e\n \u003cli\u003eSheng B, Yao D, Du X, Chen D, Zhou L. Establishment and validation of a risk prediction model for high-grade cervical lesions. \u003cem\u003eEur J Obstet Gynecol Reprod Biol.\u003c/em\u003e 2023;281:1\u0026ndash;6. doi:10.1016/j.ejogrb.2023.01.003\u003c/li\u003e\n \u003cli\u003eXue P, Seery S, Wang S, Jiang Y, Qiao Y. Developing a predictive nomogram for colposcopists: a retrospective, multicenter study of cervical precancer identification in China. \u003cem\u003eBMC Cancer.\u003c/em\u003e 2023;23(1):163. doi:10.1186/s12885-023-09919-3\u003c/li\u003e\n \u003cli\u003eRamspek CL, Jager KJ, Dekker FW, Zoccali C, van Diepen M. External validation of prognostic models: what, why, how, when, and where? \u003cem\u003eClin Kidney J.\u003c/em\u003e 2020;14(1):49\u0026ndash;58. doi:10.1093/ckj/sfz123\u003c/li\u003e\n \u003cli\u003eCollins GS, Reitsma JB, Altman DG, Moons KGM. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. \u003cem\u003eBMJ.\u003c/em\u003e 2015;350:g7594. doi:10.1136/bmj.g7594\u003c/li\u003e\n \u003cli\u003ePrendiville W, Sankaranarayanan R. Colposcopy and treatment of cervical precancer. In: \u003cem\u003eColposcopic terminology: the 2011 IFCPC nomenclature\u003c/em\u003e. Lyon (FR): International Agency for Research on Cancer; 2017. p. 91\u0026ndash;115. (IARC Technical Report No. 45).\u003c/li\u003e\n \u003cli\u003eWolff RF, Moons KGM, Riley RD, Whiting PF, Westwood M, Collins GS, et al; PROBAST Group. PROBAST: a tool to assess the risk of bias and applicability of prediction model studies. \u003cem\u003eAnn Intern Med.\u003c/em\u003e 2019;170(1):51\u0026ndash;58. doi:10.7326/M18-1376\u003c/li\u003e\n \u003cli\u003eVan Calster B, Steyerberg EW, Van Smeden M, McLernon DJ, Wynants L. Calibration: the Achilles heel of predictive analytics. \u003cem\u003eBMC Med.\u003c/em\u003e 2019;17(1):230. doi:10.1186/s12916-019-1466-7\u003c/li\u003e\n \u003cli\u003eAkarachantachote N, Chadcham S, Saithanu K. Cutoff threshold of variable importance in projection for variable selection. \u003cem\u003eInt J Pure Appl Math.\u003c/em\u003e 2014;94(3):307\u0026ndash;322. doi:10.12732/ijpam.v94i3.2\u003c/li\u003e\n \u003cli\u003eZhang Z, Rousson V, Qian X, Chen M, Lee WC, Guo Y, et al. Decision curve analysis: a technical note. \u003cem\u003eAnn Transl Med.\u003c/em\u003e 2018;6(15):308. doi:10.21037/atm.2018.07.02\u003c/li\u003e\n \u003cli\u003eSahasrabuddhe VV, Parham GP, Vermund SH, Mwanahamuntu MH. Cervical cancer prevention in low- and middle-income countries: feasible, affordable, essential. \u003cem\u003eCancer Prev Res (Phila).\u003c/em\u003e 2012;5(1):11\u0026ndash;17. doi:10.1158/1940-6207.CAPR-11-0540\u003c/li\u003e\n \u003cli\u003eKhan MJ, Einstein MH, Huh WK, Conageski C, Moscicki AB, Waxman AG, et al. ASCCP colposcopy standards: role of colposcopy, benefits, potential harms, and terminology for colposcopic practice. \u003cem\u003eJ Low Genit Tract Dis.\u003c/em\u003e 2017;21(4):223\u0026ndash;229. doi:10.1097/LGT.0000000000000338\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003eTables 1 to 4 are available in the Supplementary Files section.\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-cancer","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bcan","sideBox":"Learn more about [BMC Cancer](http://bmccancer.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bcan/default.aspx","title":"BMC Cancer","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Cervical Intraepithelial Neoplasia, High-Grade Squamous Intraepithelial Lesion, Colposcopy, Risk Prediction Model, External Validation, Human Papillomavirus","lastPublishedDoi":"10.21203/rs.3.rs-8615288/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8615288/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003ePredictive models for high-grade cervical lesions (CIN2+/HSIL) have been developed in various populations; however, their generalizability across different healthcare settings is uncertain. External validation in specific populations is required before clinical implementation.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eThis retrospective external validation study included Thai women who underwent colposcopy between 2022 and 2024 at a tertiary referral center. The primary outcome was histologically confirmed CIN2+/HSIL. Two previously published predictive models were applied without modifications. Model performance was evaluated in terms of discrimination using the area under the receiver operating characteristic curve (AUROC), calibration using calibration plots, calibration metrics, Brier scores, and clinical utility using a decision curve analysis.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eA total of 536 women were included, of whom 28.7% were diagnosed with CIN2+/HSIL. Despite differences in demographic characteristics, HPV genotype distribution, cytological findings, and colposcopic features compared to the original development cohorts, both models demonstrated robust performance in the Thai population. The Xue model showed excellent discrimination, with an AUROC of 0.85 (95% CI: 0.81\u0026ndash;0.89), whereas the Sheng model demonstrated good discrimination, with an AUROC of 0.80 (95% CI: 0.77\u0026ndash;0.85). At the optimal thresholds, the Xue model achieved a sensitivity of 77% and specificity of 82%, whereas the Sheng model achieved a sensitivity of 63% and specificity of 84%. Both models showed good calibration, with calibration slopes of 1.00, calibration-in-the-large of 0.00, and Brier scores of 0.127 and 0.147. Decision curve analysis demonstrated a positive net clinical benefit across the clinically relevant threshold probabilities.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eBoth predictive models showed good external validity and clinical utility in Thai colposcopy patients. These findings support individualized risk stratification in women with abnormal screening results for guiding clinical decisions.\u003c/p\u003e","manuscriptTitle":"External Validation of Predictive Models for High-Grade Cervical Lesions in Thai Women Undergoing Colposcopy","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-02-12 13:47:16","doi":"10.21203/rs.3.rs-8615288/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-04-28T09:05:58+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-02-22T11:31:20+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-02-22T03:16:00+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-02-21T10:37:50+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-02-20T21:09:08+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"157077310322908099165992324936310523841","date":"2026-02-16T05:46:36+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"216665946860409514287134522061494589113","date":"2026-02-16T04:19:00+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"215143018262645495004446237318798181071","date":"2026-02-16T03:45:33+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"201814090865581710407339699214735142152","date":"2026-02-16T03:43:46+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"49967613203811178414726009474981555489","date":"2026-02-10T06:56:02+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-02-09T06:48:40+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-02-06T07:08:10+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-01-19T05:44:35+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-01-18T16:18:11+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Cancer","date":"2026-01-18T16:11:39+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-cancer","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bcan","sideBox":"Learn more about [BMC Cancer](http://bmccancer.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bcan/default.aspx","title":"BMC Cancer","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"1154b678-b515-423e-9c07-6ee6cb960c2f","owner":[],"postedDate":"February 12th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-05-18T12:38:32+00:00","versionOfRecord":[],"versionCreatedAt":"2026-02-12 13:47:16","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8615288","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8615288","identity":"rs-8615288","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}