Included children

When analyzing dental records from the Public Dental Service, 1048 children (510 girls, 538 boys) meeting the criteria were identified, all of whom had CRAs at 3 and 6 years of age. Data from SKaPa showed that 381 children were lacked an examination code and were consequently excluded from further analysis (Fig. 1). The remaining 667 children, comprising 330 (49%) girls and 337 (51%) boys, were included in the subsequent study. Among these, 435 children also underwent R24&5.

Fig. 1

Flow chart detailing the inclusion of children and drop-outs in the study. Reasons for the drop-out of 381 children were attributed to the absence of examination codes

Caries status

At 3 years of age, 41 children (6%), including 22 girls and 19 boys, exhibited signs of caries with deft3 > 0. By 6 years of age, deft6 > 0 was observed in 169 children (25%), comprising 81 girls and 88 boys.

The mean (± SD) for deft at 3 years (deft3) was 0.23 (± 1.16), and at 6 years, deft6 was 0.96 (± 2.17).

Caries risk assessment–oral factors

When analyzing R2, R23 showed 37 children (6%) at risk for caries, while R24&5 resulted in 79 children (13%) were at risk (Table 5).

Table 5 Results from caries risk assessment, proportion, and simple regression analyses of the associations between oral and social factors (from Table 2), and the prevalence of caries (deft > 0) at 6 years of age

The presence of risk at R23 or R24&5 was significantly associated with caries at 6 years of age (simple regression, p < 0.0001). The OR for developing caries at 6 years of age was calculated to be 9.6 for those identified at risk at 3 years and 36.7 for those at risk at 4&5 years of age.

Social factors

Data from SCB regarding social factors (Table 2) showed varying associations (OR) with caries. Key social factors found to be significantly associated with caries included having a foreign background, low disposable income, a low education level of mother, belonging to a single-parent family, and having ≥ 3 siblings (Table 5).

After analyzing data from dental records, SKaPa, and SCB, it was observed that oral factors, social factors, and a combination of both were associated with caries, with OR > 1 (Table 5). Considering a combination of oral and social factors, various caries risk models were developed based on oral factors and different social factors (Table 5). Models including a foreign background, single parent family, parents’ education level, and having ≥ three siblings were created and analyzed. The predictive ability increased when models included a foreign background, having ≥ three siblings, parents’ education level and single-parent family. In contrast, including economic factors in the created models did not result in any further difference (Tables 3, 4).

The best-fitting models, including both oral and social factors, are presented in Tables 3, 4, represented by the area under the Receiver Operating Characteristic (ROC) Curve (AUC).

Combined models, 3 years of age

In models utilizing data from 3 years age (B3 to J3), the AUC, including both oral and social factors, showed up to a 7% higher prediction compared to oral factors alone in model A3. All models showed a higher predictive value with a larger AUC in comparison to model A3 (Table 3). Pairwise comparisons of the ROC curves between model A (oral factors) and models B-J (oral and social factors) were conducted, revealing a significant difference except for model E3 (DeLong test p < 0.05). Pairwise comparisons of the ROC curves among all other models, excluding A, showed no significant difference (DeLong test p < 0.05). ROC curves for models A3, C3, and H3 (Fig. 2).

Fig. 2

Receiver operating characteristic (ROC) curve depicting the predictive ability of three caries risk models at 3 years of age. Model A3 (blue line) includes oral factors with an AUC of 0.63. Model C3 (orange line), including oral factors, foreign background, ≥ three siblings, and single-parent family shows an AUC of 0.70. Model H3 (green line), including oral factors, ≥ three siblings, single-parent family, and father and mother with low educational levels shows an AUC of 0.70

Combined models, 4&5 years of age

Analyzing models of data from 4 and 5 years of age (B4&5 to J4&5), including both oral and social factors with calculations of AUC, resulted in up to a 4% higher prediction compared to model A4&5, exclusively including oral factors. All models demonstrated a higher predictive value with a larger AUC compared to model A4&5 (Table 4). Pairwise comparisons of the ROC curves between model A (oral factors) and models B-J (oral and social factors) revealed a significant difference between all models (DeLong test p < 0.05). Pairwise comparisons of the ROC curves among all other models, excluding A, showed no significant difference (DeLong test p < 0.05). For the illustration of ROC curves of models A4&5 and J4&5 (Fig. 3).

Fig. 3

Receiver Operating Characteristic (ROC) curve, depicting the predictive ability of two caries risk models at 4&5 years of age. Model A4&5 includes oral factors AUC 0.78, while model J4&5 includes oral factors, foreign background, ≥ three siblings, and low educational level of father, AUC 0.82

Comparing R2 from different ages

To compare R23 to R24&5, the AUC of models A3 and A4&5 were calculated. The AUC for A3 was 0.63, and for A4&5, it was 0.78 (Tables 3, 4). The predictive ability increased 15% when performing R24&5 compared to R23 based on oral factors (Fig. 4).

Fig. 4

Receiver Operating Characteristic (ROC) curve, depicting the predictive capability of the caries risk model, including oral factors at 3 and 4&5 years of age, model A3 AUC 0.63 and model A4&5 AUC 0.78

Comparing models from different ages

When reviewing the models A to J at the various ages, the prediction increased by 8–15% when utilizing models from 4 and 5 years, comparing AUCs for the different models at different ages. The 15% difference was observed in AUC for A3 (0.63), compared to AUCs for models C4&5, I4&5, and J4&5 (0.82) (Tables 3, 4). Pairwise comparisons of the ROC curves of model A (oral factors) to B-J (oral and social factors) were conducted, indicating differences between age groups for all models except models C, D and I (DeLong test p < 0.05). This implies that the designed models A, B, E–H, and J increased the prediction for CRA when used at 4&5 years of age, compared to use at 3 years age.