Results
Participant characteristics
Participants’ characteristics are presented in Table 1. Out of 837 children who underwent skin examinations, 7.3% (61/837) presented with AD lesions at the time of clinic visits. Amongst 61 children diagnosed with AD, 23% (14/61) presented with moderate-to-severe AD lesions while 77% (47/61) had mild AD lesions. The mean (SD) SCORAD scores for mild and moderate-to-severe AD cases were 17.2 (4.6) and 37.86 (12.7), respectively.
Of these 837 participants, 181 children (21.6%) did not undergo oral examination at 3-years due to logistic/manpower constraints, with no significant differences in socio-demographic characteristics, except for monthly household income, between those who received oral examination vs those who did not (Appendix Table 1). Although the AD severity profile of children who underwent oral examination significantly differed from those who did not, no differences in SCORAD scores were observed for both moderate-to-severe AD and mild AD groups (p>0.05). Oral examinations for ECC detection were performed in 656 participants at the 3-year visit. ECC lesions were observed in 85.7% (6/7), 36.8% (14/38) and 42.8% (262/611) children in moderate-to-severe AD, mild AD and non-AD groups, respectively. Furthermore, the mean (SD) decayed surfaces in moderate-to-severe, mild and non-AD groups were 5.4 (4.6), 2.2 (4.1) and 2.4 (5.1), respectively. Three examiners performed oral examinations, with mean inter- and intra-examiner reliability scores of 0.80 and 0.80, respectively (for ICDAS code 2) and 0.80 and 0.90, respectively (for ICDAS codes 3-6).
AD severity and early childhood caries
A significant difference in number of decayed surfaces across the AD severity groups was observed (P=0.04). On subgroup analysis after Bonferroni correction, significantly higher decayed surfaces were found in moderate-to-severe AD group compared to those without AD (p=0.03), while no difference was found between mild AD and the reference group (p=0.62) (Table 1). After adjusting for potential confounders (Model-1), children in the moderate-to-severe AD group were at higher risk of ECC (aIRR 2.30 [95% confidence interval (CI) 1.04-5.06]; p=0.03), compared to non-AD group. However, no differences in ECC risk was observed between mild AD and reference group (aIRR 0.89 [95% CI 0.47-1.70]; p=0.74) (Table 2). Other significant risk factors for ECC included Chinese ethnicity (aIRR 2.91 [95%CI 1.82-4.66]; p<0.001), Malay ethnicity (aIRR 2.31 [95%CI 1.37-3.89]; p=0.002) and lower household income level (aIRR 1.94 [95%CI 1.15-3.26]; p=0.01), detailed in Table 2.
Potential confounding effect of postnatal dietary and oral hygiene factors
On adjusting for additional postnatal dietary factors such as duration of breastfeeding and frequency of child’s daily sweet snack intake (Model-2), children with moderate-to-severe AD also showed a 2-fold increase in ECC risk (aIRR 2.41 [95%CI 1.08-5.37]; p=0.03) compared to non-AD, similar to Model-1 (Table 3). On further inclusion of oral hygiene factors such as child’s daily tooth-brushing frequency (Model-3), strength of the association remained robust (aIRR 2.31 [95%CI 1.06-5.03]; p=0.03). For both Models-2 and 3, no association was seen between mild AD and ECC, similar to Model 1 where no dietary/oral hygiene factors were controlled in the regression model.
Comparative evaluation of different diagnostic systems for AD estimation
Using physician diagnosis as the gold standard, caregiver-report of doctor-diagnosed AD demonstrated high specificity (86.6%) and negative predictive value (NPV=95.5%), but low sensitivity (51.9%) and positive predictive value (PPV=24.8%), in identifying true AD cases (Table 4). Single use of SPT performed similar to the caregiver-report of doctor-diagnosed AD with sensitivity, specificity, PPV and NPV of 40.3%, 88.2%, 22.6% and 94.6%, respectively. In contrast, combined criteria of caregiver-reported doctor-diagnosed AD and a positive SPT demonstrated 2-fold higher PPV values (45.4%), compared to individual criteria, although other parameters such as sensitivity (44.1%), specificity (96.3%) and NPV (96.2%) remained similar.