Predicting early childhood caries risk using machine learning: a Finnish registry-based study

Abstract Background Early childhood caries (ECC) remains a significant public health challenge, affecting young children and contributing to long-term oral health disparities. Traditional risk assessment tools often lack scalability and predictive accuracy. Machine learning (ML) models trained on national child health and social registries offer a promising solution for early, scalable ECC risk prediction in public health systems. Methods We constructed a retrospective nationwide cohort using the FinRegistry project, which links electronic health, demographic, and family data across Finland. Due to pending access to confirmed ECC diagnoses, a simulated ECC outcome with 20% prevalence was generated. A Random Forest model was trained on a 70/30 training/testing split to assess technical feasibility. Predictor importance was extracted to evaluate variable relevance. The study population included registry-linked individuals with complete demographic and household information. Results Preliminary models showed strong discrimination between simulated ECC cases and controls. Key predictive variables included receipt of social assistance, number of children in the household, maternal marital status, and indicators from the infectious disease registry. Model performance, measured by AUC, reached 1.00 on the simulated dataset. These results demonstrate that registry data structure can support early risk stratification for ECC using ML. Conclusions Our findings support the technical feasibility of using registry data and machine learning for ECC risk prediction. While current results are based on simulated outcomes, they provide a foundation for future application with real diagnostic data. This approach can inform early preventive targeting and strengthen precision public health strategies for oral disease prevention. Key messages • Machine learning models can be applied to linked national registry data to assess feasibility for predicting early childhood caries risk. • Registry-based prediction models can support early identification of high-risk children and inform future preventive oral health strategies.