Predicting early childhood caries at birth using machine learning: a nationwide Finnish cohort study

Background

Early childhood caries (ECC) is a common and largely preventable oral condition in children, reflecting broader social and family-level differences in health. Identifying children at higher risk early can support targeted prevention.

Methods

We conducted a nationwide cohort study using linked Finnish registers including all children born in 2012–2017. Children born in 2012–2015 were used for model development and etiologic analyses, with temporal validation in 2016–2017 births. ECC was defined as first diagnosis before age 6. Prediction models were developed using elastic net, ridge, lasso, and gradient boosting based on variables available at birth. Logistic regression models were used to estimate adjusted associations.

Results

Prediction models showed moderate discrimination (AUC 0.73–0.74), with the highest 20% risk group accounting for approximately 50% of cases. Socioeconomic factors were the strongest determinants. Family clustering was substantial: parental oral disease and older sibling ECC were strongly associated with higher risk. Maternal smoking remained an independent predictor, and gestational diabetes showed a persistent effect. Maternal antibiotic and opioid use during pregnancy were associated with higher ECC risk after adjustment, whereas associations for benzodiazepines attenuated after socioeconomic adjustment. Neonatal antibiotic exposure was not independently associated. A gestational age–birth weight mismatch (late preterm with high birth weight) was associated with markedly elevated risk. Modest municipality-level differences accounted for 5–7% of variation.

Conclusions

These findings suggest that early-life factors are associated with differences in ECC risk and may inform targeted preventive strategies. Family context, maternal health, and specific perinatal patterns are important components of risk, supporting early, family-centered and proportionate prevention strategies.

Key messages

• ECC risk differences are evident from birth and are shaped by socioeconomic and family factors, highlighting opportunities for early prevention.

• Early-life data can help identify children at higher risk of ECC, supporting targeted and proportionate prevention strategies.