DEVELOPMENT OF XGBOOST MODELS FOR PREDICTING DEVELOPMENT OF CERVICAL DENTAL LESIONS BASED ON INDEPENDENT RISK PREDICTORS
DOI:
https://doi.org/10.11603/2311-9624.2026.3.16677Keywords:
erosion, wedge-shaped defect, cervical caries, predictors, oral fluid, preventionAbstract
Addressing the issue of significant prevalence of cervical dental lesions involves screening and prescribing preventive measures for individuals being at risk of their development. The aim of the study is to develop computer models for predicting risk of development of cervical dental lesions in young patients and evaluate their quality. Materials and methods. Sample included 272 patients (mean age 24.3 ± 6.9 years), in whom independent predictors for occurrence of wedgeshaped defect (WSD), cervical caries (CC) and enamel erosion (E) were identified that became input data for computer models. There was applied Extreme Gradient Boosting (XGBoost) tree-based machine learning method implemented in Python programming language with the use of scikit-learn and XGBoost librarie and SHAP. For WSD model, class imbalance was taken into account by introducing a class weight (scale_pos_weight), for CC model by a narrowed hyperparameter space and repeated stratified cross-validation, for enamel E model by Synthetic Minority Over-sampling Technique (SMOTE). The models were evaluated on out-of-sample (out-of-fold) predictions with calculation of Area under the curve-Receiver Operating Characteristic curve (AUC-ROC), sensitivity (recall), specificity, accuracy and NPV. Results. Priority was given to recall over indicators of accuracy and specificity that resulted in decreasing the number of missed cases for each pathology. The highest recall = 0.90 combined with moderate discrimination ability (AUC-ROC = 0.702) was consistent with CC prediction model. WSD prediction model also had a high recall = 0.833 and a slightly higher discrimination ability compared to CC (AUC-ROC = 0.714). This confirmed feasibility of using XGBoost algorithm to determine influence of predictors on development of dental lesions. Enamel E prediction model was developed under conditions of pronounced class imbalance (n = 15), therefore it is presented as a continuous evaluation of relative risk. Its discrimination ability (AUCROC = 0.638) confirmed presence of a useful prognostic signal when using SMOTE. Conclusions. Presented models can be used to identify individuals at increased risk of developing WSD, CC, and E of tooth enamel in order to reduce their prevalence and justify the prescribing therapeutic and preventive measures.
References
1. Alsubai S. Enhancing prediction of tooth caries using significant features and multi-model classifier. PeerJ Comput Sci. 2023. Vol. 9. Р. e1631. https://doi.org/10.7717/peerj-cs.1631. DOI: https://doi.org/10.7717/peerj-cs.1631
2. Клітинська О. В., Лешко М. M. Метаболом слини як інструмент діагностики в стоматології. Intermedical journal. 2025. Вип. 1. С. 36–40. 10.32782/2786-7684/2025-1-6.
3. Contemporary diagnosis and management of dental erosion / T. Donovan et al. J Esthet Restor Dent. 2021. Vol. 33 (1). Р. 78–87. https://doi.org/10.1111/jerd.12706. DOI: https://doi.org/10.1111/jerd.12706
4. Associated risk factors with quantitative erosive tooth wear progression / F. Marro et al. J Dent. 2022. Vol. 123. Р. 104179. https://doi.org/10.1016/j.jdent.2022.104179. DOI: https://doi.org/10.1016/j.jdent.2022.104179
5. A systematic literature review: exploring the challenges of ensemble model for medical imaging / M. R. Supriyadi et al. BMC Med Imaging. 2025. Vol. 25 (1). Р. 128. https://doi.org/10.1186/s12880-025-01667-4. DOI: https://doi.org/10.1186/s12880-025-01667-4
6. Teeth segmentation and carious lesions segmentation in panoramic X-ray images using CariSeg, a networks’ ensemble / A. C. Mărginean et al. Heliyon. 2024. Vol. 10 (10). Р. e30836. https://doi.org/10.1016/j.heliyon.2024.e30836. DOI: https://doi.org/10.1016/j.heliyon.2024.e30836
7. Salivary Biomarkers for Dental Caries Detection and Personalized Monitoring / P. N. Paqué et al. J Pers Med. 2021. Vol. 11 (3). Р. 235. https://doi.org/10.3390/jpm11030235. DOI: https://doi.org/10.3390/jpm11030235
8. Одноралов А. І. Біомаркери порожнини рота у діагностиці запалення пародонта у підлітків та осіб молодого віку: критичний огляд доказової бази та концептуальна схема клініко-лабораторного скринінгу. Intermedical Journal. 2026. Вип. 2. С. 70–77. https://doi.org/10.32782/2786-7684/2026-2-11. DOI: https://doi.org/10.32782/2786-7684/2026-2-11
9. Огляд сучасних методик оцінки ризику появи карієсу / Я. Ю. Водоріз та ін. Актуальні проблеми сучасної медицини: Вісник Української медичної стоматологічної академії. 2024. Т. 24, вип. 4. С. 277–283. https://doi.org/10.31718/2077–1096.24.4.277. DOI: https://doi.org/10.31718/2077-1096.24.4.277
10. Screening machine learning model for predicting the occurrence of cervical dental lesions / I. I. Zabolotna et al. Clinical and Preventive Medicine. 2026. Vol. 2. Р. 89–98. DOI: https://doi.org/10.31612/2616-4868.2.2026.11
11. Predicting dental caries outcomes in young adults using machine learning approach / C. Ogwo et al. BMC Oral Health. 2024. Vol. 24 (1). Р. 529. https://doi.org/10.31612/2616-4868.2.2026.11. DOI: https://doi.org/10.1186/s12903-024-04294-7
12. Latent variables and structural equation models for longitudinal relationships: an illustration in nutritional epidemiology / M. Chavance et al. BMC Med Res Methodol. 2010. Vol. 10. Р. 37. https://doi.org/10.1186/s12903-024-04294-7. DOI: https://doi.org/10.1186/1471-2288-10-37
13. Diagnostic accuracy of dental caries detection using ensemble techniques in deep learning with intraoral camera images / S. Kang et al. PLoS One. 2024. Vol. 19 (9). Р. e0310004. DOI: https://doi.org/10.1371/journal.pone.0310004
14. Mehrabanian M., Kavousinejad S., Dastouri E. Interpretable machine learning modeling of treatment outcomes for silver and fluoride therapy in early childhood caries. Evid Based Dent. 2025. Vol. 26 (4). Р. 168–169. DOI: https://doi.org/10.1038/s41432-025-01195-8
15. Risk Factors associated with noncarious cervical lesions / G. A. Fontelle et al. J Clin Exp Dent. 2024. Vol. 16 (7). Р. e873–878. DOI: https://doi.org/10.4317/jced.61349
16. Non-carious cervical lesions (NCCLs) and associated factors: A multilevel analysis in a cohort study in southern Brazil / F. F. Demarco et al. J Clin Periodontol. 2022. Vol. 49 (1). Р. 48–58. DOI: https://doi.org/10.1111/jcpe.13549
17. Worawongvasu R. Scanning electron microscope characterization of noncarious cervical lesions in human teeth. J Oral Maxillofac Pathol. 2021. Vol. 25 (1). Р. 202. DOI: https://doi.org/10.4103/jomfp.JOMFP_232_20
18. Non-carious cervical lesions and risk factors: A case-control study / A. Alvarez-Arenal et al. J Oral Rehabil. 2019. Vol. 46 (1). Р. 65–75. DOI: https://doi.org/10.1111/joor.12721
19. The Role of Occlusal Factors in the Presence of Noncarious Cervical Lesions in Young People: A Case-Control Study study / A. Alvarez-Arenal et al. Oper Dent. 2019. Vol. 44 (1). Р. E12–22. DOI: https://doi.org/10.2341/18-028-C
20. Tooth wear and the role of salivary measures in general practice patients / D. S. Ramsay et al. Clin Oral Investig. 2015. Vol. 19 (1). Р. 85–95. DOI: https://doi.org/10.1007/s00784-014-1223-4
21. Multivariable prediction models of caries increment: a systematic review and critical appraisal / K. Havsed et al. Syst Rev. 2023. Vol. 12 (1). Р. 202. DOI: https://doi.org/10.1186/s13643-023-02298-y
22. Дмитришин, Т. М. Аналіз взаємозв’язків між мікробіологічними та біохімічними, біофізичними показниками у пацієнтів, які користуються знімними протезами. Світ медицини та біології. 2018. № 2 (64). С. 44–48.
