An Explainable and Transparent Machine Learning Approach for Predicting Dental Caries: A Cross-National Validation Study

An Explainable and Transparent Machine Learning Approach for Predicting Dental Caries: A Cross-National Validation Study | 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 An Explainable and Transparent Machine Learning Approach for Predicting Dental Caries: A Cross-National Validation Study Otso Tirkkonen, Henna Tiensuu, Elina Väyrynen, Jaakko Suutala, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6783190/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 17 Jan, 2026 Read the published version in BMC Oral Health → Version 1 posted 10 You are reading this latest preprint version Abstract BACKGROUND There has been a notable increase in artificial intelligence (AI) studies in dentistry. However, the inadequate use of proper validation methods has led to overly optimistic performance metrics of machine learning (ML) models. External validation provides evidence of a ML model’s performance with independent datasets and is crucial for generalizability. METHODS We developed Extreme Gradient Boosting (XGBoost) models to detect dental caries using easy-to-collect questionnaire data. ML model training was conducted using cross-validation nested resampling with a holdout test set, utilizing NHANES datasets (n = 6070). Performance of the trained model was tested using external data from the Northern Finland Birth Cohorts (NFBC1966 and NFBC1986; n = 3616). To enhance interpretability, beeswarm plots were constructed to visualize variable importance. RESULTS The ML model demonstrated solid performance in predicting dental caries on the internal dataset, with an area under the operating characteristics curve (AUC) of 0.821 (95% CI 0.795–0.846). However, the model encountered difficulties in identifying participants with dental caries, as shown by its poor sensitivity of 0.420, despite achieving a high specificity of 0.916. When applied to the external dataset, the ML model encountered significant challenges, with the AUC dropping to 0.550 (95% CI 0.532–0.569), sensitivity decreasing to 0.053, and specificity slightly improving to 0.974. Important variables identified by the model included were self-rated condition of teeth and gums, presence of missing teeth, financial status, and time since last dental visit. CONCLUSION The performance of our ML model during external validation degraded notably compared to the internal validation. However, the XAI methodology exhibited great potential to be used in the future for individualized dental caries risk assessment Oral Health Artificial Intelligence Dentistry Validation Study Risk Factors Cohorts Studies Figures Figure 1 Figure 2 Introduction Artificial intelligence (AI) refers to the theory and advancement of computational systems capable of executing functions and tasks typically necessitating human intelligence, including the ability to make decisions [ 1 ]. Machine learning (ML) is a subclass of AI which specifically involves training algorithms to learn from data and make predictions based on data without explicit guidance from a human. ML models use algorithms and statistical models to analyze data and identify patterns rather than just following a set of instructions. These patterns are later utilized to recognize trends to make predictions from unseen data [ 2 ]. This type of methodical approach is a highly efficient way of modeling large datasets (both linear and non-linear), with high-dimensional characteristics commonly observed in health data [ 3 , 4 ]. To generalize the output of ML models, the models must be tested with an external dataset. External validation refers to the process of evaluating a model’s predictions using data from a different source (data not used for training) to assess the performance of the ML model [ 5 ]. Due to various factors such as differences in healthcare systems and populations across countries, the cross-national aspect of external validation ensures that the ML model can function effectively without relying on country-specific characteristics present in the dataset used for model training [ 5 ]. Despite the importance of external validation, the insufficient use of adequate ML model validation techniques in dentistry has been identified as a cause leading to overly optimistic metrics [ 6 , 7 ]. Currently, ML models in dentistry frequently encounter challenges concerning interpretability and transparency, making it challenging to understand and explain its applicability in the decision-making-processes [ 7 ]. Explainable AI (XAI) is a subclass of AI designed to help humans understand the decision-making process of ML models by providing transparency to further trust their outcomes [ 8 ]. This is particularly crucial in fields such as medicine and dentistry, where outcomes directly affect human health [ 9 ]. By utilizing XAI, clinicians could potentially integrate ML into the clinical decision-making process in the future. As far as we know, there are no prior studies implementing external validation for ML models in detecting dental caries, using easy-to-collect questionnaire data. The objective of this study was to train an ML model in predicting dental caries and performing cross-national external validation. Another objective was to enhance the transparency of ML models by implementing novel XAI methods. Methods Study population and data sources This study utilized two cohorts: the National Health and Nutrition Examination Survey (NHANES) from the United States (U.S.) and the Northern Finland Birth Cohorts (NFBC) from Northern Finland. NHANES is a large-scale study conducted by the Centers for Disease Control and Prevention (CDC) to assess the health and nutritional status of the U.S. population through interviews, physical exams, and laboratory tests. These examinations also included dental examinations performed by licensed dentists. This study included NHANES participants aged 30–50 from four waves: 2013–2014, 2015–2016, 2017–2018, and 2019–2020 were included (n = 6,070) [ 10 ]. The NFBC cohorts consisted of participants who were born in Northern Finland between January 1st to December 31st, 1966 (NFBC1966) and those born between July 1st ,1985 and June 30th, 1986 (NFBC1986) [ 11 – 14 ]. The clinical oral health data was collected when the NFBC1966 participants were 45– to 47–year-olds (n = 1,927) and the NFBC1986 participants were 33– to 35–year-old (n = 1,689) respectively. The total study sample included in the study is depicted in Fig. 1 . More detailed information from both cohorts is described in the appendix. Study variables In this study, the NHANES dataset was utilized for model development and internal evaluation. Dental caries was selected as an outcome variable. The clinical oral examinations in the NHANES dataset were performed by licenced dentists following strict examination protocol and diagnostic criterion set by Radike and colleagues [ 15 ]. Any of the following findings were considered criteria for coronal caries: softness at the base of occlusal, facial and lingual surfaces; opacity compared to adjacent areas indicating demineralization; smooth facial and lingual surfaces with decalcification or white spots; penetrable or scrapable areas; shadowing or loss of translucency: and breaks in the enamel of the proximal surfaces of posterior teeth, as detected with an explorer were considered as criterion for coronal caries. Participants presenting with at least one coronal caries lesion were classified as having decayed teeth. As NHANES is a national study, radiographic imaging was not utilized for diagnostic purposes. Details of the examination protocol and diagnostic criteria are described elsewhere [ 16 ]. The predictor variables included in this study were selected based on their association with dental caries, their presence in the NHANES datasets and the ease of their collection through self-reported questionnaires. The NFBC dataset was used for the external validation in this study. Likewise, the dental caries status recorded by the trained and calibrated dentists was considered as an outcome variable. The caries lesions were registered utilizing the International Caries Detection and Assessment System (ICDAS) [ 17 ]. Lesions with a score of 4 or higher were considered to require restorative treatment and were therefore defined as coronal caries. Participants with at least one coronal caries lesion were classified as having decayed teeth. In both NFBC cohorts, radiographs were not utilized for diagnostic purposes. Both NFBC1966 and NFBC1986 were conducted following the same examination protocol, which is described in more detail in the appendix. The predictor variables were selected based on the presence in the NHANES dataset to enable data alignment between datasets. More information on selected predictive variables for both cohorts are described in the appendix. Datasets and data harmonisation First, all continuous and categorial predictive variables in the NHANES dataset were retained in their original forms (see appendix). Secondly, data harmonization was performed solely for the NHANES data to ensure data alignment, enabling external validation with the NFBC dataset. Data harmonization involves standardizing data to achieve data alignment. In this study, data standardization was accomplished through variable recategorization and the removal of non-matching variables between datasets. Likewise, data alignment ensures variable comparability across separate datasets. This process allows for the evaluation of model performance using unseen, representative data, providing a more realistic measure of its performance. Data alignment, including data harmonization has become a common method in ML for ensuring consistency and comparability across datasets [ 18 ]. The process of data harmonization done in this study is explained in the appendix. Model training and validation The NHANES dataset was randomly divided into a training set (80%, n = 4,506) and a test set (20%, n = 1,127). Internal validation was performed on the pre-harmonized data before proceeding with data harmonization. After harmonization, post-harmonized internal validation was performed and evaluated, using the same 80:20 ratio with the same training and test sets as in the pre-harmonized internal validation. Subsequently, the model developed with the pre-harmonized NHANES training set was employed for external validation, utilizing the NFBC dataset as an unseen holdout test set. Extreme gradient boosting (XGBoost) was selected as the ML algorithm for this study. A cross-validation nested resampling technique with a holdout test set was utilized, using 5-fold cross-validation to train and evaluate the XGBoost models. After the initial data split, the nested resampling technique was applied exclusively to the training data for model development. Subsequently, hold out testing data was only utilized during the model testing. The nested resampling technique consists of inner and outer loops. Each of the 5 outer-loop training folds was moved into the inner resampling loop to tune the model’s hyperparameters. The best combination of hyperparameters was selected using the grid search method, based on the highest area under the receiver operating characteristic curve (AUC). These optimized hyperparameters were then used in the outer loop to train and calibrate the model. Finally, the model was tested with the unseen holdout test data. Missing variables were included in both the model training and testing process. All results presented in this study are based on unseen holdout test sets. The model training protocol is illustrated in Fig. 1 , and the hyperparameters utilized are listed in the appendix. Model interpretation and visualization Due to the complexity of interpreting the nonlinear decision-making process of ML models, SHapely Additive exPlanations (SHAP) values were used to investigate the influence of specific variables and impacts on the outcome [ 19 ]. This was done using a beeswarm plot, which visualizes variable importance and effects in terms of SHAP values. In this plot, each data point represents the SHAP value of a specific variable for an individual instance in the training set. This information can be used to analyze specific feature values related to the prediction of dental caries. In Table 1 , feature values are presented in ascending order for each variable. Variable importance reflects the relative contribution of each feature to the ML model, with higher values indicating greater importance. In the beeswarm plot, variables are presented by their importance, and their corresponding SHAP values are displayed next to each variable, further demonstrating their impact on the model’s risk prediction. Statistical analysis The performance of the ML models was evaluated using the following metrics: AUC, accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and F1-score. AUC was calculated using the Delong method. A 95% confidence interval (CI) was determined for both AUC and accuracy to evaluate the precision of the estimates. Software All statistical analyses and data handling were performed with the use of the R software [ 20 ]. The R “caret” package was used to train and tune the ML models [ 21 ], while the R “shapviz” package was used to visualize the SHAP values in the beeswarm plot [ 22 ]. Results Almost one-thirds of the NHANES (32.8%) and nearly half of the of NFBC participants (49.9%) presented with dental caries. Mean daily added sugar consumption (79.2 g/day vs 18.6 g/day) was higher among the NHANES cohort compared to the NFBC cohort. The remaining predictive variables showed greater variability in their distributions between the NFBC and NHANES datasets (Table 1 ). Table 1 Study population characteristics. Study variables NHANES % (n) NFBC % (n) Gender Female 52.9% (3209) 56.1% (2029) Male 47.1% (2861) 43.9% (1587) Missing 0 0 Education No college degree 39.1% (2372) 41.5% (1397) College degree 60.9% (3696) 58.5% (1969) Missing 2 250 Marital status Married/living with partner 66.1% (1656) 78.7% (2693) Divorced/widowed/separated 13.3% (333) 7.45% (255) Never married 20.6% (516) 13.9% (474) Missing 3565 194 Time since latest dentist visit 2 years 33.1% (2003) 17.2% (605) Missing 11 92 Tobacco use Non-user 71.5% (4022) 81.1% (2668) User 28.5% (1601) 18.9% (621) Missing 447 327 Interdental cleaning frequency < 5 days a week 63.4% (3847) 92.5% (2906) At least 5 day a week 36.6% (2221) 7.54% (237) Missing 2 473 Fillings No fillings 22.1% (1340) 46.9% (1694) Fillings 77.9% (4730) 53.1% (1920) Missing 0 2 Missing tooth No missing 48.4% (2935) 58.8% (2126) Missing tooth 51.6% (3135) 41.2% (1490) Missing 0 0 Decayed teeth No 29.7% (1803) 50.1% (1810) Yes 70.3% (4267) 49.9% (1806) Missing 0 0 Financial status (mean (sd)) 2.37 (1.60) 2.82 (1.35) Missing 915 501 Sugar consumption (mean (sd)) 79.2 (76.0) 18.6 (15.1) Missing: 426 344 Body mass index (mean(sd)) 30.2 (7.76) 26.4 (4.67) Missing: 47 339 Study variables NHANES % (n) NFBC % (n) Fillings No fillings 22.1% (1340) 4.18% (151) Fillings 77.9% (4730) 95.8% (3465) Missing 0 0 Tobacco use Non-user 71.5% (4022) 81.1% (2668) User 28.5% (1601) 18.9% (621) Missing 447 327 Time since latest dentist visit 2 years 33.1% (2003) 17.2% (605) Missing 11 92 Interdental cleaning frequency < 5 days a week 63.4% (3847) 92.5% (2906) At least 5 day a week 36.6% (2221) 7.54% (237) Missing 2 473 Education No college degree 39.1% (2372) 41.5% (1397) College degree 60.9% (3696) 58.5% (1969) Missing 2 250 Missing tooth No missing 48.4% (2935) 58.8% (2126) Missing tooth 51.6% (3135) 41.2% (1490) Missing 0 0 Gender Female 52.9% (3209) 56.1% (2029) Male 47.1% (2861) 43.9% (1587) Missing 0 0 Marital status Married/living with partner 66.1% (1656) 78.7% (2693) Divorced/widowed/separated 13.3% (333) 7.45% (255) Never married 20.6% (516) 13.9% (474) Missing 3565 194 Decayed teeth No 29.7% (1803) 50.1% (1810) Yes 70.3% (4267) 49.9% (1806) Missing 0 0 Financial status (mean (sd)) 2.37 (1.60) 2.82 (1.35) Missing 915 501 Sugar consumption (mean (sd)) 79.2 (76.0) 18.6 (15.1) Missing: 426 344 Body mass index (mean(sd)) 30.2 (7.76) 26.4 (4.67) Missing: 47 339 Model performance metrics, including results from internal validation and external validation, are presented in Table 2 . The AUC values which indicate the discriminative ability of the ML model in detecting diseases-free and diseased cases, were good in both the pre-harmonized (AUC = 0.821) and post-harmonized internal validation (AUC = 0.785). The accuracy of the pre-harmonized and post-harmonized internal validation models was 0.773 and 0.767, respectively. However, both AUC and accuracy dropped drastically in the external validation, indicating poor performance of the ML model. A similar drop was observed in sensitivity, reflecting a reduced ability to correctly identify participants with the disease. In contrast, specificity, which measures the model’s ability to correctly identify disease-free participants showed a slight improvement in the external validation model. Table 2 Performance of predictive models for internal and external datasets. AUC (95% CI) ACCURACY (95% CI) NIF Sensitivity Specificity F1-score PPV NPV Pre-harmonized internal validation 0.821 (0.795–0.846) 0.773 (0.748–0.796) 0.712 0.420 0.916 0.516 0.668 0.796 Post-harmonized internal validation 0.785 (0.756–0.813) 0.767 (0.742–0.790) 0.712 0.391 0.919 0.492 0.662 0.789 External validation 0.550 (0.532–0.569) 0.514 (0.498–0.531) 0.501 0.053 0.974 0.099 0.671 0.508 For the visualization and interpretation of the ML-models, the XAI beeswarm plots were constructed by visualizing the variable importance based on the SHAP values (Fig. 2 ). In the pre-harmonized internal validation model, the most predictive variables were Self-rated condition of teeth and gums , Time since last dentist visit, Missing tooth and Financial status . The same was true for the post-harmonized internal validation and external validation models except for the Self-rated condition of teeth and gums , which was removed during the data harmonization phase. Surprisingly, sugar consumption and interdental cleaning were among the least important variables in the external validation model. Discussion This study aimed to develop ML models to predict dental caries in the adult population and to improve the transparency of these developed models through external validation. The ML models developed and tested using the internal dataset demonstrated sufficient performance in predicting dental caries. However, the ML model encountered notable challenges when applied to the external dataset when predicting dental caries. Moreover, this study used novel XAI methods to provide new insights of ML model explainability in the dental caries risk assessment. During the internal validation, ML models had acceptable performances in terms of AUC. Similar findings were reported in previous studies [ 26 , 27 ]. Specificity, which measures the model’s ability to correctly identify individuals without disease (those with sound teeth), was high in both pre-harmonized and post-harmonized models. However, the ML models struggled to identify participants with dental caries, as demonstrated by their poor sensitivity. This phenomenon is common in population-based studies, where the number of healthy participants typically exceeds those with the disease. This trend was evident in the confusion matrix of this study. Additionally, similar findings were noted in a previous study where there was a minority of participants with dental caries [ 27 ]. This phenomenon is called class imbalance. Future studies should consider implementing class balancing methods such as under or oversampling to enhance ML metrics when encountering similar challenges. In this study, despite using cross-validation with ahold out test set, the trained model performed poorly. This result was consistent with previous studies which tested periodontal outcomes using an external dataset [ 28 , 29 ]. It can be speculated that this poor performance is due to a dataset shift. Dataset shift is a phenomenon when even slight deviations between training and testing datasets degrades ML model performance [ 30 ]. For instance, differences in the distributions of predictor variables between the internal and external datasets in this study may have led to covariate shift. Furthermore, the beeswarm plots constructed in this study also show that the external dataset’s variables are systematically skewed towards feature values predicting those without disease, further supporting the presence of covariate shift [ 31 ]. Likewise, differences in the distribution of the outcome variable between the internal and external datasets may have resulted in a prior probability shift. As a result of prior probability shift, the ML model tends to overestimate the prediction of participants without disease [ 31 ]. In the present study, XAI, specifically beeswarm plots, was used to enhance the transparency and interpretability of the model’s predictions. XAI allows for visualization of individual predictions and helps illustrate the nature and values of the features contributing to those predictions. In contrast to this study, previous research has recognized past caries experience and sugar intake as the most important variables predicting dental caries in young adults and children [ 26 , 32 ]. In this study, there was high variance in the effects of certain values, particularly among tobacco users and individuals who visited a dental clinic over 2 years ago. Additionally, XAI can be a useful tool for gaining deeper insight into ML model performance, particularly when comparing internal and external validation results. To develop clinically useful ML models, emphasis should also be given not only to cross-national external validation but also to model transparency. This can be achieved through XAI methods as demonstrated in this study. [ 33 ]. The XAI techniques such as beeswarm plots leverage the possibility of creating personalized risk assessment by visualizing the individual contributions of predictive variables. In contrast, traditional statistical methods such as regression models evaluate risk factors at the population level. Therefore, while XAI offers promising potential for clinical application, it is essential that ML models first achieve strong performance on both internal and external datasets. To the best of our best knowledge, this is the first study in the field of dentistry to perform cross-national external validation of an ML model for predicting dental caries using easy-to-collect questionnaire-based data. This can be considered as one of the strengths of our study. Another strength of this study is the use of large datasets (NHANES and NFBC), that were supported by a robust methodological framework. Finally, the use of nested resampling technique with a holdout test set for cross-validation during ML model training can also be considered as another strength of this study. However, this study also has several limitations. First, self-reported data may be subjected to response bias. Second, the absence of radiographic images limited the assessment of cariological status to clinical examinations only. Additionally, selection bias may have been introduced due to the geographically restricted nature of the NFBC participants. Finally, differences in the data collection periods between the NFBC and NHANES participants may have led to temporal bias. While these limitations exist, such constraints are often challenging to avoid in population-based studies of this nature. Conclusion The performance of our ML model during external validation degraded notably compared to the internal validation. However, the XAI methodology exhibited great potential to be used in the future for individualized dental caries risk assessment. Declarations Ethics approval and consent to participate The National Center for Health Statistics (NCHS) Research Ethics Review Board approved the NHANES study, in addition to all the participants having given their written consent for their participation. The NHANES study protocol was approved by the National Center for Health Statistics Research Ethics Review Committee. Clinical trial number: not applicable. Consent for publication Participation in the NFBCs and their follow-ups was voluntary, and participants provided their informed consent. The Northern Ostrobothnia Hospital District Ethical Committee consented to NFBCs study protocols ( NFBC 1966 permission number: 94/2011 & NFBC 1986 permission number: 108/2017 ). This study adhered to the TRIPOD+AI statement and the Artificial Intelligence in Dental Research: Checklist for Authors, Reviewers, and Readers statement and the STROBE guidelines for human observational studies in addition to the checklist for AI-specific guidelines [23-25]. Availability of data and materials The NHANES data has been made publicly accessible. The NHANES data is obtainable at https://wwwn.cdc.gov/nchs/nhanes/. The NFBC data used in this study is available from the University of Oulu, Infrastructure for Population Studies (www.oulu.fi/nfbc). Permission to use the NFBC data can be applied for research purposes via the electronic material request portal ( [email protected] ). Competing interests The authors declare that they have no competing interests. Funding The NFBC1966 46-year follow-up study was financially supported by the University of Oulu (Grant no. 24000692), Oulu University Hospital (Grant no. 24301140), and the ERDF European Regional Development Fund (Grant no. 539/2010 A31592). The NFBC1986 33-35-year follow-up study received funding from the University of Oulu (Strategic funding from donations) and Oulu University Hospital (Grant no. K65760). The oral health study was partially funded by the Research Council of Finland (former Academy of Finland, Grant no. 326189). Additionally, O.T received a financial support from the Finnish Dental Society Apollonia. The funder had no role in study design, data collection, analyses, and interpretation. Author contributions O.T contributed to conception and design, acquisition, analysis, and interpretation, writing original draft. H.T contributed to conception and design, analysis and interpretation, review & editing, supervision. Elina Väyrynen contributed to conception and design, writing - original draft. J.S contributed to interpretation of data, review & editing. V.V contributed to interpretation of data, review & editing, and supervision. M.LL contributed to conception, review & editing. S.K contributed to conception and design, writing - original draft, review & editing, and supervision. All authors gave their final approval and agree to be accountable for all aspects of the work. Acknowledgements We are grateful to all cohort members, researchers, and NFBC project center personnel who contributed to the NFBC data collections. References Legg S, Hutter M. Universal intelligence: A definition of machine intelligence. Minds Mach. 2007;17:391–444. https://doi.org/10.1007/s11023-007-9079-x . Erickson BJ, Korfiatis P, Akkus Z, Kline TL. Machine Learning for Medical Imaging. Radiographics. 2017;37:505–15. https://doi.org/10.1148/rg.2017160130 . Chatterjee P, Cymberknop LJ, Armentano RL. (2019) Nonlinear Systems in Healthcare towards Intelligent Disease Prediction. Nonlinear Syst Theor Asp Recent Appl. https://doi.org/10.5772/intechopen.88163 Berisha V, Krantsevich C, Hahn PR, Hahn S, Dasarathy G, Turaga P, Liss J. Digital medicine and the curse of dimensionality. NPJ Digit Med. 2021;4:1–8. https://doi.org/10.1038/s41746-021-00521-5 . 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;14:49–58. https://doi.org/10.1093/ckj/sfaa188 . Chow DY, Tay JRH, Nascimento GG. Systematic Review of Prognosis Models in Predicting Tooth Loss in Periodontitis. J Dent Res. 2024;103:596–604. https://doi.org/10.1177/00220345241237448 . Shafi I, Fatima A, Afzal H, Díez IT, Lipari V, Breñosa J, Ashraf I. (2023) A Comprehensive Review of Recent Advances in Artificial Intelligence for Dentistry E-Health. Diagnostics 13. https://doi.org/10.3390/diagnostics13132196 Barredo Arrieta A, Díaz-Rodríguez N, Del Ser J, et al. Explainable Artificial Intelligence (XAI). Inf Fusion. 2020;58:82–115. https://doi.org/10.1016/j.inffus.2019.12.012 . Tjoa E, Guan C. A Survey on Explainable Artificial Intelligence (XAI): Toward Medical XAI. IEEE Trans Neural Netw Learn Syst. 2021;32:4793–813. https://doi.org/10.1109/TNNLS.2020.3027314 . Johnson C, Paulose-Ram R, Ogden CL, Carroll M, Kruszon-Moran D, Dohrmann SM, Curtin LR. (2013) National health and nutrition examination survey: analytic guidelines, 1999–2010. Vital Health Stat 2 Data Eval Methods Res. Järvelin MR, Hartikainen-Sorri AL, Rantakallio P. Labour induction policy in hospitals of different levels of specialisation. Br J Obstet Gynaecol. 1993;100:310–5. https://doi.org/10.1111/j.1471-0528.1993.tb12971.x . Nordström T, Miettunen J, Auvinen J, Ala-Mursula L, Keinänen-Kiukaanniemi S, Veijola J, Järvelin MR, Sebert S, Männikkö M. Cohort Profile: 46 years of follow-up of the Northern Finland Birth Cohort 1966 (NFBC1966). Int J Epidemiol. 2022;50:1786–7. https://doi.org/10.1093/ije/dyab109 . j. University of Oulu. (1966) Northern Finland Birth Cohort 1966. University of Oulu. http://urn.fi/urn:nbn:fi:att:bc1e5408-980e-4a62-b899-43bec3755243 . Accessed 2025 Apr 4. University of Oulu. (1986) Northern Finland Birth Cohort 1986. University of Oulu. http://urn.fi/urn:nbn:fi:att:f5c10eef-3d25-4bd0-beb8-f2d59df95b8e . Accessed 2025 Apr 4. Radike AW. (1972) Criteria for diagnosis of dental caries. In: Szoke Anti-caries Effect of After-meal Gum-chewing. Proceedings of the Conference on the Clinical Testing of Cariostatic Agents; October 14–16, 1968; Chicago, IL. American Dental Association. pp. 87–88. Therapeutics C. on D (1972) Proceedings of the conference on the clinical testing of cariostatic agents: Chicago, Illinois October 14–16, 1968. Council on Dental Research. Ismail AI, Sohn W, Tellez M, Amaya A, Sen A, Hasson H, Pitts NB. The International Caries Detection and Assessment System (ICDAS): an integrated system for measuring dental caries. Community Dent Oral Epidemiol. 2007;35:170–8. https://doi.org/10.1111/j.1600-0528.2007.00347.x . Cheng C, Messerschmidt L, Bravo I, Waldbauer M, Bhavikatti R, Schenk C, Grujic V, Model T, Kubinec R, Barceló J. A General Primer for Data Harmonization. Sci Data. 2024;11:152. https://doi.org/10.1038/s41597-024-02956-3 . Lundberg S, Lee SI. (2017) A unified approach to interpreting model predictions. Proceedings of the 31st International Conference on Neural Information Processing Systems 2017; pp 4768–4777. https://dl.acm.org/doi/10.5555/3295222.3295230 R Core Team. R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing; 2022. Kuhn M. Building Predictive Models in R Using the caret Package. J Stat Softw. 2008;28:1–26. https://doi.org/10.18637/jss.v028.i05 . Mayer M. (2025) SHAP Visualizations [R package shapviz version 0.9.7]. CRAN: Contributed Packages. https://doi.org/10.32614/cran.package.shapviz Collins GS, Dhiman P, Navarro CLA, Ma J, Hooft L, Reitsma JB, Logullo P, Beam AL, Peng L, Van Calster B, et al. Protocol for development of a reporting guideline (TRIPOD-AI) and risk of bias tool (PROBAST-AI) for diagnostic and prognostic prediction model studies based on artificial intelligence. BMJ Open. 2021;11:e048008. https://doi.org/10.1136/bmjopen-2020-048008 . Schwendicke F, Singh T, Lee JH, Gaudin R, Chaurasia A, Wiegand T, Uribe S, Krois J. Artificial intelligence in dental research: Checklist for Authors, Reviewers, Readers. J Dent. 2021;107:103610. https://doi.org/10.1016/j.jdent.2021.103610 . Cuschieri S. The STROBE guidelines. Saudi J Anaesth. 2019;13:31. https://doi.org/10.4103/sja.sja_543_18 . Toledo Reyes L, Knorst JK, Ortiz FR, Brondani B, Emmanuelli B, Saraiva Guedes R, Mendes FM, Ardenghi TM. Early Childhood Predictors for Dental Caries: A Machine Learning Approach. J Dent Res. 2023;102:999–1006. https://doi.org/10.1177/00220345231170535 . Bomfim RA. Machine learning to predict untreated dental caries in adolescents. BMC Oral Health. 2024;24:1. https://doi.org/10.1186/s12903-024-04073-4 . Bashir NZ, Rahman Z, Chen SLS. Systematic comparison of machine learning algorithms to develop and validate predictive models for periodontitis. J Clin Periodontol. 2022;49:958–69. https://doi.org/10.1111/jcpe.13692 . Enevold C, Nielsen CH, Christensen LB, Kongstad J, Fiehn NE, Hansen PR, Holmstrup P, Havemose-Poulsen A, Damgaard C. Suitability of machine learning models for prediction of clinically defined Stage III/IV periodontitis from questionnaires and demographic data in Danish cohorts. J Clin Periodontol. 2024;51. https://doi.org/10.1111/jcpe.13874 . Subbaswamy A, Saria S. From development to deployment: dataset shift, causality, and shift-stable models in health AI. Biostatistics. 2020;21:345–52. https://doi.org/10.1093/biostatistics/kxz041 . Kull M, Flach P. (2014) Patterns of dataset shift. First International Workshop on Learning over Multiple Contexts (LMCE) at ECML-PKDD. 5. Ogwo C, Brown G, Warren J, Caplan D, Levy S. Predicting dental caries outcomes in young adults using machine learning approach. BMC Oral Health. 2024;24:529. https://doi.org/10.1186/s12903-024-04294-7 . Futoma J, Simons M, Panch T, Doshi-Velez F, Celi LA. The myth of generalisability in clinical research and machine learning in health care. Lancet Digit Health. 2020;2:e489–92. https://doi.org/10.1016/s2589-7500(20)30186-2 . Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 17 Jan, 2026 Read the published version in BMC Oral Health → Version 1 posted Editorial decision: Revision requested 08 Jul, 2025 Reviews received at journal 07 Jul, 2025 Reviews received at journal 03 Jul, 2025 Reviewers agreed at journal 28 Jun, 2025 Reviewers agreed at journal 26 Jun, 2025 Reviewers invited by journal 26 Jun, 2025 Editor invited by journal 18 Jun, 2025 Editor assigned by journal 31 May, 2025 Submission checks completed at journal 31 May, 2025 First submitted to journal 30 May, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6783190","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":477809064,"identity":"1ab89977-ba39-4424-a39e-532ec9764bf4","order_by":0,"name":"Otso Tirkkonen","email":"","orcid":"","institution":"University of Oulu","correspondingAuthor":false,"prefix":"","firstName":"Otso","middleName":"","lastName":"Tirkkonen","suffix":""},{"id":477809065,"identity":"234e5d1b-7725-49ce-b263-638c34de29e8","order_by":1,"name":"Henna Tiensuu","email":"","orcid":"","institution":"University of Oulu","correspondingAuthor":false,"prefix":"","firstName":"Henna","middleName":"","lastName":"Tiensuu","suffix":""},{"id":477809067,"identity":"bd140487-e989-4cc4-95f6-8f2e03822c64","order_by":2,"name":"Elina Väyrynen","email":"","orcid":"","institution":"University of Oulu","correspondingAuthor":false,"prefix":"","firstName":"Elina","middleName":"","lastName":"Väyrynen","suffix":""},{"id":477809073,"identity":"2e120b01-4882-4a97-b461-8c73c8302a28","order_by":3,"name":"Jaakko Suutala","email":"","orcid":"","institution":"University of Oulu","correspondingAuthor":false,"prefix":"","firstName":"Jaakko","middleName":"","lastName":"Suutala","suffix":""},{"id":477809076,"identity":"bcec743a-1133-4e67-8c16-e3eae5133f24","order_by":4,"name":"Vuollo Ville","email":"","orcid":"","institution":"University of Oulu","correspondingAuthor":false,"prefix":"","firstName":"Vuollo","middleName":"","lastName":"Ville","suffix":""},{"id":477809077,"identity":"d4c346ee-ef1b-484e-a206-c9ae192a551b","order_by":5,"name":"Marja-Liisa Laitala","email":"","orcid":"","institution":"University of Oulu","correspondingAuthor":false,"prefix":"","firstName":"Marja-Liisa","middleName":"","lastName":"Laitala","suffix":""},{"id":477809080,"identity":"c5780b5d-aac6-47b9-9cc5-ae715cb817d6","order_by":6,"name":"Saujanya Karki","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA/UlEQVRIie2RMUsDMRTH/6FwXXJ2drD3Fd6RUVs/Szmwyw0BFycJFOLScmv9FlcK4hjI4BJ0LXS5qZOC4iRUMLSCTjlHofkND154v/f+ECAS+afQrnKAKToDur65or8rF0DHN65FwY8C+60EBrMbR/Lj3mbgdvM2l099spxgZCCUKymfOZurVIvbmtZirwSCEUqiVFumehCsofWo9krnNaBk1TPln9qeq1733SuPOyV4BauShL8yUulUsJpMu0KrjRQnelxo7i7ZnApxbBNpQkpWFXf5iz4dVHy8ZNPtsH/0MFk0ZhsIBiR+4wTJ7ycTFPzXNcB1y0wkEokcNF8CnVGEb40uYwAAAABJRU5ErkJggg==","orcid":"","institution":"University of Oulu","correspondingAuthor":true,"prefix":"","firstName":"Saujanya","middleName":"","lastName":"Karki","suffix":""}],"badges":[],"createdAt":"2025-05-30 09:23:25","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6783190/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6783190/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1186/s12903-026-07660-9","type":"published","date":"2026-01-17T16:28:43+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":85757380,"identity":"08d83719-df32-45c3-bed7-65e9a9c5771b","added_by":"auto","created_at":"2025-07-01 10:54:42","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":132879,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eFlowchart illustrating study methodology\u003c/em\u003e\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-6783190/v1/a5acdb308272aa13ea177692.png"},{"id":85755951,"identity":"30887418-f028-4edc-9a5f-be92bd00bd7e","added_by":"auto","created_at":"2025-07-01 10:46:42","extension":"jpeg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":429852,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eSHAP beeswarm plots from (A) post-harmonized internal validation, (B) pre-harmonized internal validation and (C) external validation. A wider horizontal distribution of data points along the SHAP value axis indicates a higher density of data points. Data points positioned along the positive SHAP value axis indicate predictions in favor of dental caries. Feature values are represented by the color gradient of the data points, with brighter colors indicating higher raw feature values and grey representing missing values.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6783190/v1/d834365d01012d579ad05116.jpeg"},{"id":100615923,"identity":"73ca03e1-ecc8-4e76-913a-4ecaf5404773","added_by":"auto","created_at":"2026-01-19 17:38:15","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1488773,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6783190/v1/73cc8594-cf71-43c7-aa78-4466127c26fa.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"An Explainable and Transparent Machine Learning Approach for Predicting Dental Caries: A Cross-National Validation Study","fulltext":[{"header":"Introduction","content":"\u003cp\u003eArtificial intelligence (AI) refers to the theory and advancement of computational systems capable of executing functions and tasks typically necessitating human intelligence, including the ability to make decisions [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Machine learning (ML) is a subclass of AI which specifically involves training algorithms to learn from data and make predictions based on data without explicit guidance from a human. ML models use algorithms and statistical models to analyze data and identify patterns rather than just following a set of instructions. These patterns are later utilized to recognize trends to make predictions from unseen data [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. This type of methodical approach is a highly efficient way of modeling large datasets (both linear and non-linear), with high-dimensional characteristics commonly observed in health data [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. To generalize the output of ML models, the models must be tested with an external dataset. External validation refers to the process of evaluating a model\u0026rsquo;s predictions using data from a different source (data not used for training) to assess the performance of the ML model [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. Due to various factors such as differences in healthcare systems and populations across countries, the cross-national aspect of external validation ensures that the ML model can function effectively without relying on country-specific characteristics present in the dataset used for model training [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. Despite the importance of external validation, the insufficient use of adequate ML model validation techniques in dentistry has been identified as a cause leading to overly optimistic metrics [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eCurrently, ML models in dentistry frequently encounter challenges concerning interpretability and transparency, making it challenging to understand and explain its applicability in the decision-making-processes [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Explainable AI (XAI) is a subclass of AI designed to help humans understand the decision-making process of ML models by providing transparency to further trust their outcomes [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. This is particularly crucial in fields such as medicine and dentistry, where outcomes directly affect human health [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. By utilizing XAI, clinicians could potentially integrate ML into the clinical decision-making process in the future. As far as we know, there are no prior studies implementing external validation for ML models in detecting dental caries, using easy-to-collect questionnaire data. The objective of this study was to train an ML model in predicting dental caries and performing cross-national external validation. Another objective was to enhance the transparency of ML models by implementing novel XAI methods.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eStudy population and data sources\u003c/h2\u003e \u003cp\u003eThis study utilized two cohorts: the National Health and Nutrition Examination Survey (NHANES) from the United States (U.S.) and the Northern Finland Birth Cohorts (NFBC) from Northern Finland. NHANES is a large-scale study conducted by the Centers for Disease Control and Prevention (CDC) to assess the health and nutritional status of the U.S. population through interviews, physical exams, and laboratory tests. These examinations also included dental examinations performed by licensed dentists. This study included NHANES participants aged 30\u0026ndash;50 from four waves: 2013\u0026ndash;2014, 2015\u0026ndash;2016, 2017\u0026ndash;2018, and 2019\u0026ndash;2020 were included (n\u0026thinsp;=\u0026thinsp;6,070) [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. The NFBC cohorts consisted of participants who were born in Northern Finland between January 1st to December 31st, 1966 (NFBC1966) and those born between July 1st ,1985 and June 30th, 1986 (NFBC1986) [\u003cspan additionalcitationids=\"CR12 CR13\" citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. The clinical oral health data was collected when the NFBC1966 participants were 45\u0026ndash; to 47\u0026ndash;year-olds (n\u0026thinsp;=\u0026thinsp;1,927) and the NFBC1986 participants were 33\u0026ndash; to 35\u0026ndash;year-old (n\u0026thinsp;=\u0026thinsp;1,689) respectively. The total study sample included in the study is depicted in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. More detailed information from both cohorts is described in the appendix.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eStudy variables\u003c/h3\u003e\n\u003cp\u003eIn this study, the NHANES dataset was utilized for model development and internal evaluation. Dental caries was selected as an outcome variable. The clinical oral examinations in the NHANES dataset were performed by licenced dentists following strict examination protocol and diagnostic criterion set by Radike and colleagues [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. Any of the following findings were considered criteria for coronal caries: softness at the base of occlusal, facial and lingual surfaces; opacity compared to adjacent areas indicating demineralization; smooth facial and lingual surfaces with decalcification or white spots; penetrable or scrapable areas; shadowing or loss of translucency: and breaks in the enamel of the proximal surfaces of posterior teeth, as detected with an explorer were considered as criterion for coronal caries. Participants presenting with at least one coronal caries lesion were classified as having decayed teeth. As NHANES is a national study, radiographic imaging was not utilized for diagnostic purposes. Details of the examination protocol and diagnostic criteria are described elsewhere [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. The predictor variables included in this study were selected based on their association with dental caries, their presence in the NHANES datasets and the ease of their collection through self-reported questionnaires.\u003c/p\u003e \u003cp\u003eThe NFBC dataset was used for the external validation in this study. Likewise, the dental caries status recorded by the trained and calibrated dentists was considered as an outcome variable. The caries lesions were registered utilizing the International Caries Detection and Assessment System (ICDAS) [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Lesions with a score of 4 or higher were considered to require restorative treatment and were therefore defined as coronal caries. Participants with at least one coronal caries lesion were classified as having decayed teeth. In both NFBC cohorts, radiographs were not utilized for diagnostic purposes. Both NFBC1966 and NFBC1986 were conducted following the same examination protocol, which is described in more detail in the appendix. The predictor variables were selected based on the presence in the NHANES dataset to enable data alignment between datasets. More information on selected predictive variables for both cohorts are described in the appendix.\u003c/p\u003e\n\u003ch3\u003eDatasets and data harmonisation\u003c/h3\u003e\n\u003cp\u003eFirst, all continuous and categorial predictive variables in the NHANES dataset were retained in their original forms (see appendix). Secondly, data harmonization was performed solely for the NHANES data to ensure data alignment, enabling external validation with the NFBC dataset. Data harmonization involves standardizing data to achieve data alignment. In this study, data standardization was accomplished through variable recategorization and the removal of non-matching variables between datasets. Likewise, data alignment ensures variable comparability across separate datasets. This process allows for the evaluation of model performance using unseen, representative data, providing a more realistic measure of its performance. Data alignment, including data harmonization has become a common method in ML for ensuring consistency and comparability across datasets [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. The process of data harmonization done in this study is explained in the appendix.\u003c/p\u003e\n\u003ch3\u003eModel training and validation\u003c/h3\u003e\n\u003cp\u003eThe NHANES dataset was randomly divided into a training set (80%, n\u0026thinsp;=\u0026thinsp;4,506) and a test set (20%, n\u0026thinsp;=\u0026thinsp;1,127). Internal validation was performed on the pre-harmonized data before proceeding with data harmonization. After harmonization, post-harmonized internal validation was performed and evaluated, using the same 80:20 ratio with the same training and test sets as in the pre-harmonized internal validation. Subsequently, the model developed with the pre-harmonized NHANES training set was employed for external validation, utilizing the NFBC dataset as an unseen holdout test set.\u003c/p\u003e \u003cp\u003eExtreme gradient boosting (XGBoost) was selected as the ML algorithm for this study. A cross-validation nested resampling technique with a holdout test set was utilized, using 5-fold cross-validation to train and evaluate the XGBoost models. After the initial data split, the nested resampling technique was applied exclusively to the training data for model development. Subsequently, hold out testing data was only utilized during the model testing. The nested resampling technique consists of inner and outer loops. Each of the 5 outer-loop training folds was moved into the inner resampling loop to tune the model\u0026rsquo;s hyperparameters. The best combination of hyperparameters was selected using the grid search method, based on the highest area under the receiver operating characteristic curve (AUC). These optimized hyperparameters were then used in the outer loop to train and calibrate the model. Finally, the model was tested with the unseen holdout test data. Missing variables were included in both the model training and testing process. All results presented in this study are based on unseen holdout test sets. The model training protocol is illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, and the hyperparameters utilized are listed in the appendix.\u003c/p\u003e\n\u003ch3\u003eModel interpretation and visualization\u003c/h3\u003e\n\u003cp\u003eDue to the complexity of interpreting the nonlinear decision-making process of ML models, SHapely Additive exPlanations (SHAP) values were used to investigate the influence of specific variables and impacts on the outcome [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. This was done using a beeswarm plot, which visualizes variable importance and effects in terms of SHAP values. In this plot, each data point represents the SHAP value of a specific variable for an individual instance in the training set. This information can be used to analyze specific feature values related to the prediction of dental caries. In Table \u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, feature values are presented in ascending order for each variable. Variable importance reflects the relative contribution of each feature to the ML model, with higher values indicating greater importance. In the beeswarm plot, variables are presented by their importance, and their corresponding SHAP values are displayed next to each variable, further demonstrating their impact on the model\u0026rsquo;s risk prediction.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eStatistical analysis\u003c/h2\u003e \u003cp\u003eThe performance of the ML models was evaluated using the following metrics: AUC, accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and F1-score. AUC was calculated using the Delong method. A 95% confidence interval (CI) was determined for both AUC and accuracy to evaluate the precision of the estimates.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eSoftware\u003c/h3\u003e\n\u003cp\u003eAll statistical analyses and data handling were performed with the use of the R software [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. The R \u0026ldquo;caret\u0026rdquo; package was used to train and tune the ML models [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e], while the R \u0026ldquo;shapviz\u0026rdquo; package was used to visualize the SHAP values in the beeswarm plot [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e].\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eAlmost one-thirds of the NHANES (32.8%) and nearly half of the of NFBC participants (49.9%) presented with dental caries. Mean daily added sugar consumption (79.2 g/day vs 18.6 g/day) was higher among the NHANES cohort compared to the NFBC cohort. The remaining predictive variables showed greater variability in their distributions between the NFBC and NHANES datasets (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eStudy population characteristics.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStudy variables\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNHANES\u003c/p\u003e \u003cp\u003e% (n)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNFBC\u003c/p\u003e \u003cp\u003e% (n)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003eGender\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFemale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e52.9% (3209)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e56.1% (2029)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e47.1% (2861)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e43.9% (1587)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMissing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eEducation\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo college degree\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e39.1% (2372)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e41.5% (1397)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCollege degree\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e60.9% (3696)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e58.5% (1969)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMissing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e250\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eMarital status\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMarried/living with partner\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e66.1% (1656)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e78.7% (2693)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDivorced/widowed/separated\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e13.3% (333)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7.45% (255)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNever married\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e20.6% (516)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e13.9% (474)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMissing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3565\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e194\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTime since latest dentist visit\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026lt; 1 year\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e53.6% (3247)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e51.3% (1807)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1\u0026ndash;2 years\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e13.4% (809)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e31.6% (1112)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026gt; 2 years\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e33.1% (2003)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e17.2% (605)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMissing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e11\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e92\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTobacco use\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNon-user\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e71.5% (4022)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e81.1% (2668)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUser\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e28.5% (1601)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e18.9% (621)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMissing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e447\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e327\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eInterdental cleaning frequency\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026lt; 5 days a week\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e63.4% (3847)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e92.5% (2906)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAt least 5 day a week\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e36.6% (2221)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7.54% (237)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMissing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e473\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eFillings\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo fillings\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e22.1% (1340)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e46.9% (1694)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFillings\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e77.9% (4730)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e53.1% (1920)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMissing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eMissing tooth\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo missing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e48.4% (2935)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e58.8% (2126)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMissing tooth\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e51.6% (3135)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e41.2% (1490)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMissing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eDecayed teeth\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e29.7% (1803)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e50.1% (1810)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e70.3% (4267)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e49.9% (1806)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMissing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eFinancial status\u003c/b\u003e (mean (sd))\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2.37 (1.60)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2.82 (1.35)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMissing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e915\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e501\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSugar consumption\u003c/b\u003e (mean (sd))\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e79.2 (76.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e18.6 (15.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMissing:\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e426\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e344\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eBody mass index\u003c/b\u003e (mean(sd))\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e30.2 (7.76)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e26.4 (4.67)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMissing:\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e47\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e339\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eStudy variables\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNHANES\u003c/p\u003e \u003cp\u003e% (n)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNFBC\u003c/p\u003e \u003cp\u003e% (n)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eFillings\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo fillings\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e22.1% (1340)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4.18% (151)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFillings\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e77.9% (4730)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e95.8% (3465)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMissing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTobacco use\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNon-user\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e71.5% (4022)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e81.1% (2668)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUser\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e28.5% (1601)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e18.9% (621)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMissing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e447\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e327\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTime since latest dentist visit\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026lt; 1 year\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e53.6% (3247)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e51.3% (1807)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1\u0026ndash;2 years\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e13.4% (809)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e31.6% (1112)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026gt; 2 years\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e33.1% (2003)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e17.2% (605)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMissing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e11\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e92\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eInterdental cleaning frequency\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026lt; 5 days a week\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e63.4% (3847)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e92.5% (2906)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAt least 5 day a week\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e36.6% (2221)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7.54% (237)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMissing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e473\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eEducation\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo college degree\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e39.1% (2372)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e41.5% (1397)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCollege degree\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e60.9% (3696)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e58.5% (1969)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMissing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e250\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eMissing tooth\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo missing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e48.4% (2935)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e58.8% (2126)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMissing tooth\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e51.6% (3135)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e41.2% (1490)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMissing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eGender\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFemale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e52.9% (3209)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e56.1% (2029)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e47.1% (2861)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e43.9% (1587)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMissing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eMarital status\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMarried/living with partner\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e66.1% (1656)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e78.7% (2693)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDivorced/widowed/separated\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e13.3% (333)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7.45% (255)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNever married\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e20.6% (516)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e13.9% (474)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMissing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3565\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e194\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eDecayed teeth\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e29.7% (1803)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e50.1% (1810)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e70.3% (4267)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e49.9% (1806)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMissing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eFinancial status\u003c/b\u003e (mean (sd))\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2.37 (1.60)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2.82 (1.35)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMissing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e915\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e501\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSugar consumption\u003c/b\u003e (mean (sd))\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e79.2 (76.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e18.6 (15.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMissing:\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e426\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e344\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eBody mass index\u003c/b\u003e (mean(sd))\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e30.2 (7.76)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e26.4 (4.67)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMissing:\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e47\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e339\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eModel performance metrics, including results from internal validation and external validation, are presented in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. The AUC values which indicate the discriminative ability of the ML model in detecting diseases-free and diseased cases, were good in both the pre-harmonized (AUC\u0026thinsp;=\u0026thinsp;0.821) and post-harmonized internal validation (AUC\u0026thinsp;=\u0026thinsp;0.785). The accuracy of the pre-harmonized and post-harmonized internal validation models was 0.773 and 0.767, respectively. However, both AUC and accuracy dropped drastically in the external validation, indicating poor performance of the ML model. A similar drop was observed in sensitivity, reflecting a reduced ability to correctly identify participants with the disease. In contrast, specificity, which measures the model\u0026rsquo;s ability to correctly identify disease-free participants showed a slight improvement in the external validation model.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePerformance of predictive models for internal and external datasets.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"9\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAUC (95% CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eACCURACY (95% CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNIF\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSensitivity\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eSpecificity\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eF1-score\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003ePPV\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003eNPV\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003ePre-harmonized internal validation\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.821 (0.795\u0026ndash;0.846)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.773 (0.748\u0026ndash;0.796)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.712\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.420\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.916\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.516\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.668\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.796\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003ePost-harmonized internal validation\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.785 (0.756\u0026ndash;0.813)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.767 (0.742\u0026ndash;0.790)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.712\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.391\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.919\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.492\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.662\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.789\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eExternal validation\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.550 (0.532\u0026ndash;0.569)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.514 (0.498\u0026ndash;0.531)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.501\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.053\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.974\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.099\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.671\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.508\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eFor the visualization and interpretation of the ML-models, the XAI beeswarm plots were constructed by visualizing the variable importance based on the SHAP values (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). In the pre-harmonized internal validation model, the most predictive variables \u003cem\u003ewere Self-rated condition of teeth and gums\u003c/em\u003e, \u003cem\u003eTime since last dentist visit, Missing tooth\u003c/em\u003e and \u003cem\u003eFinancial status\u003c/em\u003e. The same was true for the post-harmonized internal validation and external validation models except for the \u003cem\u003eSelf-rated condition of teeth and gums\u003c/em\u003e, which was removed during the data harmonization phase. Surprisingly, \u003cem\u003esugar consumption\u003c/em\u003e and \u003cem\u003einterdental cleaning\u003c/em\u003e were among the least important variables in the external validation model.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study aimed to develop ML models to predict dental caries in the adult population and to improve the transparency of these developed models through external validation. The ML models developed and tested using the internal dataset demonstrated sufficient performance in predicting dental caries. However, the ML model encountered notable challenges when applied to the external dataset when predicting dental caries. Moreover, this study used novel XAI methods to provide new insights of ML model explainability in the dental caries risk assessment.\u003c/p\u003e \u003cp\u003eDuring the internal validation, ML models had acceptable performances in terms of AUC. Similar findings were reported in previous studies [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. Specificity, which measures the model\u0026rsquo;s ability to correctly identify individuals without disease (those with sound teeth), was high in both pre-harmonized and post-harmonized models. However, the ML models struggled to identify participants with dental caries, as demonstrated by their poor sensitivity. This phenomenon is common in population-based studies, where the number of healthy participants typically exceeds those with the disease. This trend was evident in the confusion matrix of this study. Additionally, similar findings were noted in a previous study where there was a minority of participants with dental caries [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. This phenomenon is called class imbalance. Future studies should consider implementing class balancing methods such as under or oversampling to enhance ML metrics when encountering similar challenges.\u003c/p\u003e \u003cp\u003eIn this study, despite using cross-validation with ahold out test set, the trained model performed poorly. This result was consistent with previous studies which tested periodontal outcomes using an external dataset [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. It can be speculated that this poor performance is due to a dataset shift. Dataset shift is a phenomenon when even slight deviations between training and testing datasets degrades ML model performance [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. For instance, differences in the distributions of predictor variables between the internal and external datasets in this study may have led to covariate shift. Furthermore, the beeswarm plots constructed in this study also show that the external dataset\u0026rsquo;s variables are systematically skewed towards feature values predicting those without disease, further supporting the presence of covariate shift [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. Likewise, differences in the distribution of the outcome variable between the internal and external datasets may have resulted in a prior probability shift. As a result of prior probability shift, the ML model tends to overestimate the prediction of participants without disease [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIn the present study, XAI, specifically beeswarm plots, was used to enhance the transparency and interpretability of the model\u0026rsquo;s predictions. XAI allows for visualization of individual predictions and helps illustrate the nature and values of the features contributing to those predictions. In contrast to this study, previous research has recognized past caries experience and sugar intake as the most important variables predicting dental caries in young adults and children [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]. In this study, there was high variance in the effects of certain values, particularly among tobacco users and individuals who visited a dental clinic over 2 years ago. Additionally, XAI can be a useful tool for gaining deeper insight into ML model performance, particularly when comparing internal and external validation results. To develop clinically useful ML models, emphasis should also be given not only to cross-national external validation but also to model transparency. This can be achieved through XAI methods as demonstrated in this study. [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. The XAI techniques such as beeswarm plots leverage the possibility of creating personalized risk assessment by visualizing the individual contributions of predictive variables. In contrast, traditional statistical methods such as regression models evaluate risk factors at the population level. Therefore, while XAI offers promising potential for clinical application, it is essential that ML models first achieve strong performance on both internal and external datasets.\u003c/p\u003e \u003cp\u003eTo the best of our best knowledge, this is the first study in the field of dentistry to perform cross-national external validation of an ML model for predicting dental caries using easy-to-collect questionnaire-based data. This can be considered as one of the strengths of our study. Another strength of this study is the use of large datasets (NHANES and NFBC), that were supported by a robust methodological framework. Finally, the use of nested resampling technique with a holdout test set for cross-validation during ML model training can also be considered as another strength of this study. However, this study also has several limitations. First, self-reported data may be subjected to response bias. Second, the absence of radiographic images limited the assessment of cariological status to clinical examinations only. Additionally, selection bias may have been introduced due to the geographically restricted nature of the NFBC participants. Finally, differences in the data collection periods between the NFBC and NHANES participants may have led to temporal bias. While these limitations exist, such constraints are often challenging to avoid in population-based studies of this nature.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThe performance of our ML model during external validation degraded notably compared to the internal validation. However, the XAI methodology exhibited great potential to be used in the future for individualized dental caries risk assessment.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cem\u003eEthics approval and consent to participate\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThe National Center for Health Statistics (NCHS) Research Ethics Review Board approved the NHANES study, in addition to all the participants having given their written consent for their participation. The NHANES study protocol was approved by the National Center for Health Statistics Research Ethics Review Committee. Clinical trial number: not applicable.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eConsent for publication\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eParticipation in the NFBCs and their follow-ups was voluntary, and participants provided their informed consent. The Northern Ostrobothnia Hospital District Ethical Committee consented to NFBCs study protocols (\u003cem\u003eNFBC 1966 permission number: 94/2011 \u0026amp; NFBC 1986 permission number: 108/2017\u003c/em\u003e).\u003c/p\u003e\n\u003cp\u003eThis study adhered to the TRIPOD+AI statement and the \u003cem\u003eArtificial Intelligence in Dental Research: Checklist for Authors, Reviewers, and Readers\u003c/em\u003e statement and the STROBE guidelines for human observational studies in addition to the checklist for AI-specific guidelines [23-25].\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eAvailability of data and materials\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThe NHANES data has been made publicly accessible. The NHANES data is obtainable at \u0026nbsp;https://wwwn.cdc.gov/nchs/nhanes/. The NFBC data used in this study is available from the University of Oulu, Infrastructure for Population Studies (www.oulu.fi/nfbc). Permission to use the NFBC data can be applied for research purposes via the electronic material request portal ([email protected]).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eCompeting interests\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eFunding\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThe NFBC1966 46-year follow-up study was financially supported by the University of Oulu (Grant no. 24000692), Oulu University Hospital (Grant no. 24301140), and the ERDF European Regional Development Fund (Grant no. 539/2010 A31592). The NFBC1986 33-35-year follow-up study received funding from the University of Oulu (Strategic funding from donations) and Oulu University Hospital (Grant no. K65760). The oral health study was partially funded by the Research Council of Finland (former Academy of Finland, Grant no. 326189). Additionally, O.T received a financial support from the Finnish Dental Society Apollonia. The funder had no role in study design, data collection, analyses, and interpretation.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eAuthor contributions\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eO.T contributed to conception and design, acquisition, analysis, and interpretation, writing original draft. H.T contributed to conception and design, analysis and interpretation, review \u0026amp; editing, supervision. Elina V\u0026auml;yrynen contributed to conception and design, writing - original draft. J.S contributed to interpretation of data, review \u0026amp; editing. V.V contributed to interpretation of data, review \u0026amp; editing, and supervision. M.LL contributed to conception, review \u0026amp; editing. S.K contributed to conception and design, writing - original draft, review \u0026amp; editing, and supervision. All authors gave their final approval and agree to be accountable for all aspects of the work.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eAcknowledgements\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eWe are grateful to all cohort members, researchers, and NFBC project center personnel who contributed to the NFBC data collections.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eLegg S, Hutter M. Universal intelligence: A definition of machine intelligence. Minds Mach. 2007;17:391\u0026ndash;444. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11023-007-9079-x\u003c/span\u003e\u003cspan address=\"10.1007/s11023-007-9079-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eErickson BJ, Korfiatis P, Akkus Z, Kline TL. Machine Learning for Medical Imaging. Radiographics. 2017;37:505\u0026ndash;15. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1148/rg.2017160130\u003c/span\u003e\u003cspan address=\"10.1148/rg.2017160130\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChatterjee P, Cymberknop LJ, Armentano RL. (2019) Nonlinear Systems in Healthcare towards Intelligent Disease Prediction. Nonlinear Syst Theor Asp Recent Appl. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.5772/intechopen.88163\u003c/span\u003e\u003cspan address=\"10.5772/intechopen.88163\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBerisha V, Krantsevich C, Hahn PR, Hahn S, Dasarathy G, Turaga P, Liss J. Digital medicine and the curse of dimensionality. NPJ Digit Med. 2021;4:1\u0026ndash;8. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41746-021-00521-5\u003c/span\u003e\u003cspan address=\"10.1038/s41746-021-00521-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRamspek 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;14:49\u0026ndash;58. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/ckj/sfaa188\u003c/span\u003e\u003cspan address=\"10.1093/ckj/sfaa188\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChow DY, Tay JRH, Nascimento GG. Systematic Review of Prognosis Models in Predicting Tooth Loss in Periodontitis. J Dent Res. 2024;103:596\u0026ndash;604. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/00220345241237448\u003c/span\u003e\u003cspan address=\"10.1177/00220345241237448\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShafi I, Fatima A, Afzal H, D\u0026iacute;ez IT, Lipari V, Bre\u0026ntilde;osa J, Ashraf I. (2023) A Comprehensive Review of Recent Advances in Artificial Intelligence for Dentistry E-Health. Diagnostics 13. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/diagnostics13132196\u003c/span\u003e\u003cspan address=\"10.3390/diagnostics13132196\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBarredo Arrieta A, D\u0026iacute;az-Rodr\u0026iacute;guez N, Del Ser J, et al. Explainable Artificial Intelligence (XAI). Inf Fusion. 2020;58:82\u0026ndash;115. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.inffus.2019.12.012\u003c/span\u003e\u003cspan address=\"10.1016/j.inffus.2019.12.012\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTjoa E, Guan C. A Survey on Explainable Artificial Intelligence (XAI): Toward Medical XAI. IEEE Trans Neural Netw Learn Syst. 2021;32:4793\u0026ndash;813. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/TNNLS.2020.3027314\u003c/span\u003e\u003cspan address=\"10.1109/TNNLS.2020.3027314\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJohnson C, Paulose-Ram R, Ogden CL, Carroll M, Kruszon-Moran D, Dohrmann SM, Curtin LR. (2013) National health and nutrition examination survey: analytic guidelines, 1999\u0026ndash;2010. Vital Health Stat 2 Data Eval Methods Res.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJ\u0026auml;rvelin MR, Hartikainen-Sorri AL, Rantakallio P. Labour induction policy in hospitals of different levels of specialisation. Br J Obstet Gynaecol. 1993;100:310\u0026ndash;5. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/j.1471-0528.1993.tb12971.x\u003c/span\u003e\u003cspan address=\"10.1111/j.1471-0528.1993.tb12971.x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNordstr\u0026ouml;m T, Miettunen J, Auvinen J, Ala-Mursula L, Kein\u0026auml;nen-Kiukaanniemi S, Veijola J, J\u0026auml;rvelin MR, Sebert S, M\u0026auml;nnikk\u0026ouml; M. Cohort Profile: 46 years of follow-up of the Northern Finland Birth Cohort 1966 (NFBC1966). Int J Epidemiol. 2022;50:1786\u0026ndash;7. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/ije/dyab109\u003c/span\u003e\u003cspan address=\"10.1093/ije/dyab109\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. j.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUniversity of Oulu. (1966) Northern Finland Birth Cohort 1966. University of Oulu. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://urn.fi/urn:nbn:fi:att:bc1e5408-980e-4a62-b899-43bec3755243\u003c/span\u003e\u003cspan address=\"http://urn.fi/urn:nbn:fi:att:bc1e5408-980e-4a62-b899-43bec3755243\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Accessed 2025 Apr 4.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUniversity of Oulu. (1986) Northern Finland Birth Cohort 1986. University of Oulu. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://urn.fi/urn:nbn:fi:att:f5c10eef-3d25-4bd0-beb8-f2d59df95b8e\u003c/span\u003e\u003cspan address=\"http://urn.fi/urn:nbn:fi:att:f5c10eef-3d25-4bd0-beb8-f2d59df95b8e\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Accessed 2025 Apr 4.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRadike AW. (1972) Criteria for diagnosis of dental caries. In: Szoke Anti-caries Effect of After-meal Gum-chewing. Proceedings of the Conference on the Clinical Testing of Cariostatic Agents; October 14\u0026ndash;16, 1968; Chicago, IL. American Dental Association. pp. 87\u0026ndash;88.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTherapeutics C. on D (1972) Proceedings of the conference on the clinical testing of cariostatic agents: Chicago, Illinois October 14\u0026ndash;16, 1968. Council on Dental Research.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIsmail AI, Sohn W, Tellez M, Amaya A, Sen A, Hasson H, Pitts NB. The International Caries Detection and Assessment System (ICDAS): an integrated system for measuring dental caries. Community Dent Oral Epidemiol. 2007;35:170\u0026ndash;8. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/j.1600-0528.2007.00347.x\u003c/span\u003e\u003cspan address=\"10.1111/j.1600-0528.2007.00347.x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCheng C, Messerschmidt L, Bravo I, Waldbauer M, Bhavikatti R, Schenk C, Grujic V, Model T, Kubinec R, Barcel\u0026oacute; J. A General Primer for Data Harmonization. Sci Data. 2024;11:152. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41597-024-02956-3\u003c/span\u003e\u003cspan address=\"10.1038/s41597-024-02956-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLundberg S, Lee SI. (2017) A unified approach to interpreting model predictions. Proceedings of the 31st International Conference on Neural Information Processing Systems 2017; pp 4768\u0026ndash;4777. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://dl.acm.org/doi/10.5555/3295222.3295230\u003c/span\u003e\u003cspan address=\"https://dl.acm.doi/10.5555/3295222.3295230\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eR Core Team. R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing; 2022.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKuhn M. Building Predictive Models in R Using the caret Package. J Stat Softw. 2008;28:1\u0026ndash;26. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.18637/jss.v028.i05\u003c/span\u003e\u003cspan address=\"10.18637/jss.v028.i05\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMayer M. (2025) SHAP Visualizations [R package shapviz version 0.9.7]. CRAN: Contributed Packages. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.32614/cran.package.shapviz\u003c/span\u003e\u003cspan address=\"10.32614/cran.package.shapviz\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCollins GS, Dhiman P, Navarro CLA, Ma J, Hooft L, Reitsma JB, Logullo P, Beam AL, Peng L, Van Calster B, et al. Protocol for development of a reporting guideline (TRIPOD-AI) and risk of bias tool (PROBAST-AI) for diagnostic and prognostic prediction model studies based on artificial intelligence. BMJ Open. 2021;11:e048008. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1136/bmjopen-2020-048008\u003c/span\u003e\u003cspan address=\"10.1136/bmjopen-2020-048008\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchwendicke F, Singh T, Lee JH, Gaudin R, Chaurasia A, Wiegand T, Uribe S, Krois J. Artificial intelligence in dental research: Checklist for Authors, Reviewers, Readers. J Dent. 2021;107:103610. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jdent.2021.103610\u003c/span\u003e\u003cspan address=\"10.1016/j.jdent.2021.103610\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCuschieri S. The STROBE guidelines. Saudi J Anaesth. 2019;13:31. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.4103/sja.sja_543_18\u003c/span\u003e\u003cspan address=\"10.4103/sja.sja_543_18\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eToledo Reyes L, Knorst JK, Ortiz FR, Brondani B, Emmanuelli B, Saraiva Guedes R, Mendes FM, Ardenghi TM. Early Childhood Predictors for Dental Caries: A Machine Learning Approach. J Dent Res. 2023;102:999\u0026ndash;1006. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/00220345231170535\u003c/span\u003e\u003cspan address=\"10.1177/00220345231170535\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBomfim RA. Machine learning to predict untreated dental caries in adolescents. BMC Oral Health. 2024;24:1. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12903-024-04073-4\u003c/span\u003e\u003cspan address=\"10.1186/s12903-024-04073-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBashir NZ, Rahman Z, Chen SLS. Systematic comparison of machine learning algorithms to develop and validate predictive models for periodontitis. J Clin Periodontol. 2022;49:958\u0026ndash;69. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/jcpe.13692\u003c/span\u003e\u003cspan address=\"10.1111/jcpe.13692\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEnevold C, Nielsen CH, Christensen LB, Kongstad J, Fiehn NE, Hansen PR, Holmstrup P, Havemose-Poulsen A, Damgaard C. Suitability of machine learning models for prediction of clinically defined Stage III/IV periodontitis from questionnaires and demographic data in Danish cohorts. J Clin Periodontol. 2024;51. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/jcpe.13874\u003c/span\u003e\u003cspan address=\"10.1111/jcpe.13874\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSubbaswamy A, Saria S. From development to deployment: dataset shift, causality, and shift-stable models in health AI. Biostatistics. 2020;21:345\u0026ndash;52. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/biostatistics/kxz041\u003c/span\u003e\u003cspan address=\"10.1093/biostatistics/kxz041\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKull M, Flach P. (2014) Patterns of dataset shift. First International Workshop on Learning over Multiple Contexts (LMCE) at ECML-PKDD. 5.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOgwo C, Brown G, Warren J, Caplan D, Levy S. Predicting dental caries outcomes in young adults using machine learning approach. BMC Oral Health. 2024;24:529. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12903-024-04294-7\u003c/span\u003e\u003cspan address=\"10.1186/s12903-024-04294-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFutoma J, Simons M, Panch T, Doshi-Velez F, Celi LA. The myth of generalisability in clinical research and machine learning in health care. Lancet Digit Health. 2020;2:e489\u0026ndash;92. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/s2589-7500(20)30186-2\u003c/span\u003e\u003cspan address=\"10.1016/s2589-7500(20)30186-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-oral-health","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"ohea","sideBox":"Learn more about [BMC Oral Health](http://bmcoralhealth.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/ohea/default.aspx","title":"BMC Oral Health","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Oral Health, Artificial Intelligence, Dentistry, Validation Study, Risk Factors, Cohorts Studies","lastPublishedDoi":"10.21203/rs.3.rs-6783190/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6783190/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBACKGROUND\u003c/h2\u003e \u003cp\u003eThere has been a notable increase in artificial intelligence (AI) studies in dentistry. However, the inadequate use of proper validation methods has led to overly optimistic performance metrics of machine learning (ML) models. External validation provides evidence of a ML model\u0026rsquo;s performance with independent datasets and is crucial for generalizability.\u003c/p\u003e\u003ch2\u003eMETHODS\u003c/h2\u003e \u003cp\u003eWe developed Extreme Gradient Boosting (XGBoost) models to detect dental caries using easy-to-collect questionnaire data. ML model training was conducted using cross-validation nested resampling with a holdout test set, utilizing NHANES datasets (n\u0026thinsp;=\u0026thinsp;6070). Performance of the trained model was tested using external data from the Northern Finland Birth Cohorts (NFBC1966 and NFBC1986; n\u0026thinsp;=\u0026thinsp;3616). To enhance interpretability, beeswarm plots were constructed to visualize variable importance.\u003c/p\u003e\u003ch2\u003eRESULTS\u003c/h2\u003e \u003cp\u003eThe ML model demonstrated solid performance in predicting dental caries on the internal dataset, with an area under the operating characteristics curve (AUC) of 0.821 (95% CI 0.795\u0026ndash;0.846). However, the model encountered difficulties in identifying participants with dental caries, as shown by its poor sensitivity of 0.420, despite achieving a high specificity of 0.916. When applied to the external dataset, the ML model encountered significant challenges, with the AUC dropping to 0.550 (95% CI 0.532\u0026ndash;0.569), sensitivity decreasing to 0.053, and specificity slightly improving to 0.974. Important variables identified by the model included were self-rated condition of teeth and gums, presence of missing teeth, financial status, and time since last dental visit.\u003c/p\u003e\u003ch2\u003eCONCLUSION\u003c/h2\u003e \u003cp\u003eThe performance of our ML model during external validation degraded notably compared to the internal validation. However, the XAI methodology exhibited great potential to be used in the future for individualized dental caries risk assessment\u003c/p\u003e","manuscriptTitle":"An Explainable and Transparent Machine Learning Approach for Predicting Dental Caries: A Cross-National Validation Study","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-07-01 10:46:37","doi":"10.21203/rs.3.rs-6783190/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-07-08T14:59:24+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-07-07T23:47:58+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-07-03T13:33:30+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"16033514770229268353239747943331961969","date":"2025-06-28T18:52:19+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"140770901309537485840434931142132059636","date":"2025-06-26T10:58:42+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-06-26T07:34:39+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-06-18T07:16:55+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-05-31T09:24:26+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-05-31T09:21:46+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Oral Health","date":"2025-05-30T09:15:32+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-oral-health","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"ohea","sideBox":"Learn more about [BMC Oral Health](http://bmcoralhealth.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/ohea/default.aspx","title":"BMC Oral Health","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"965ed61f-761c-487b-8dc8-5cafd20d5367","owner":[],"postedDate":"July 1st, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2026-01-19T17:02:10+00:00","versionOfRecord":{"articleIdentity":"rs-6783190","link":"https://doi.org/10.1186/s12903-026-07660-9","journal":{"identity":"bmc-oral-health","isVorOnly":false,"title":"BMC Oral Health"},"publishedOn":"2026-01-17 16:28:43","publishedOnDateReadable":"January 17th, 2026"},"versionCreatedAt":"2025-07-01 10:46:37","video":"","vorDoi":"10.1186/s12903-026-07660-9","vorDoiUrl":"https://doi.org/10.1186/s12903-026-07660-9","workflowStages":[]},"version":"v1","identity":"rs-6783190","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6783190","identity":"rs-6783190","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}