Study design and location
We carried out a prospective cohort study using data from the BRISA Cohort Study, conducted in two cities in Brazil: São Luís (MA) and Ribeirão Preto (SP)14. The two cities are located in different regions of the country and display widely varying cultural, socioeconomic, and demographic characteristics. So, the present study focused on only one of the cities (São Luís) to provide a more homogeneous dataset and minimize confounding bias. Besides, In addition, there were no data for some important variables in these studies collected in a standardized way in both municipalities. The study was conducted following the guidelines of the STROBE Statement.
São Luís is the capital of Maranhão, a state in the northeast of Brazil with one of the lowest human development indices (0.639) in the country. The estimated population of the city in 2010 was 1,014,83715. The average annual number of live births between 2010 and 2013 was 21,42816. A study carried out using data from 2009 showed that while many of the 480 water samples had optimal fluoride levels, access to fluoridated water was unequal, with poorer areas having lower levels of fluoride17.
The data were collected at three separate intervals between January 2010 and March 2013: during antenatal care (T1); first follow-up at birth (T2); and second follow-up at age 12–30 months (T3).
Sample
The study sample consisted of liveborn singleton children who participated in the baseline study, were followed up at T2 and T3, and underwent a dental examination (n = 865) and their mothers.
The sample of 865 children was estimated to have a power of 95.76% to detect associations between SGA and S-ECC, calculated based on the following parameters: alpha = 0.05 (bilateral); mean (± standard deviation) number of carious surfaces equal to 0.40 (± 1.54) in the exposed group and 1.69 (± 7.63) in the unexposed group; sample of 46 exposed and 819 unexposed individuals.
The study sample was selected using convenience sampling because it was not possible to obtain a representative random sample of pregnant women since there is not a reliable record of pregnant women and/or women receiving antenatal care in the state of Maranhão. Available data is neither updated nor validated. So, the women were selected during visits to public and private health centers for antenatal/follow-up care. Inclusion criteria included having performed an ultrasound scan before 20 weeks of gestation and not having completed more than 25 weeks of gestation at the time of baseline data collection. The baseline data were collected at the Clinical Research Center (CEPEC) between January 2010 and June 2011.
Data collection procedures
The data were collected using clinical examinations and three questionnaires: the mother’s questionnaire at T1, including questions on socioeconomic and demographic characteristics; the newborn questionnaire at T2, containing questions on weight and gestational age at birth; and the child questionnaire at T3, to obtain data on age, diet, and consumption of sugar-sweetened foods.
A dental examination was performed at T3 to assess DDE, S-ECC, VPI (used as a proxy measure of oral hygiene), and the number of teeth. At the end of the examination, the dentist brushed the child’s teeth to check for bleeding gums during brushing. Six examiners were trained to perform the examination (inter- and intra-rater Kappa ≥ 0.80). The data were recorded on a specific individual form for each child. The dental examinations were performed at the Maternal and Infant Health Unit (HUUMI) of the Federal University of Maranhão University Hospital (HUUFMA) following World Health Organization (WHO) recommendations: under artificial lighting, using mouth mirrors, periodontal probes, compressed air, distilled water, and a three-way syringe18.
Variables
The study outcome was S-ECC, characterized by the ratio between the number of decayed tooth surfaces and the total number of tooth surfaces. The presence of the disease was determined using the Nyvad codes and criteria: 1—Active caries without surface discontinuity (intact surface); 2—active caries (microcavity); 3—active caries (cavity); 5—inactive caries (microcavity); and 6—inactive caries (cavity)19. Nyvad is a validated classification system19, has been shown to have high reproducibility for lesion severity assessment when compared to the International System for the Detection and Evaluation of Caries (ICDAS)20, and has been used in international studies21, 22.
The main independent variable was SGA23, determined based on Kramer et al.24 and the INTERGROWTH-21st standard25. According to Kramer, SGA is defined by dividing birthweight (in g) by the weight on the 50th percentile line of the birthweight for gestational age curve based on the Canadian birthweight standards24.
According to the INTERGROWTH-21st standard, SGA babies have birthweight below the 10th percentile, meaning they are smaller than 90% of (most) other babies of the same gestational age. The INTERGROWTH-21st standards were produced using data from a multicenter multiethnic population-based project conducted between 2009 and 2014 in eight developed and developing countries, including Brazil. The primary aim of the project was to study growth, health, nutrition, and neurodevelopment from 14 weeks of gestation to 2 years of age25.
The potential mediating variables tested by this study were DDE, number of erupted teeth, and visible plaque index (VPI). DDE was diagnosed using the Fédération Dentaire Internationale Defects of Enamel index (1982)26, which categorizes DDE as diffuse opacities, demarcated opacities, and hypoplasia. Tooth surfaces with any of the three types of DDE were counted. The analysis considered the total number of teeth present in the oral cavity. The tooth was considered present in the oral cavity when any part of the tooth was visible during the examination. The VPI was also assessed during the dental examination.
The confounding variables were socioeconomic status, oral hygiene, and eating habits. Socioeconomic status was a latent variable measured using the following variables: family income (in minimum wages for the baseline year 2010); occupation of the head of the family (from unskilled job to manager/business owner); economic classification based on the criteria proposed by the Brazilian Market Research Association (ABEP)—(A/B; C; or D/E)27; and maternal education (≤ 4 to ≥ 12 years). These variables are important indicators of caries28. Oral hygiene and eating habits were assessed by asking the child’s mother/caregiver the following questions: Do you brush your child’s teeth every night before they go to bed? (yes/no); Does your child eat sugar-sweetened foods? (yes/no).
These variables were chosen based on the assumption that teeth with enamel defects are likely to have higher levels of cariogenic biofilm formation (measured by the VPI) when fermentable sugars are consumed, increasing the incidence of S-ECC. The consumption of sugar-sweetened foods was chosen as a measure of eating habits because sugar is the main cause of caries29. The VPI was used because this measure indicates severe chronic plaque buildup, which, in the presence of sugar, can lead to caries30. Nighttime brushing was chosen as a measure of oral hygiene habits because evidence suggests that this method is more effective for controlling cariogenic activity than daytime brushing10.
Statistical analysis
The association between SGA, DDE, and other covariates and S-ECC was assessed using a theoretical model (Fig. 1) tested using SEM. The statistical analyses were performed using Stata/SE 12.0 and Mplus 7.3 software. Absolute and relative frequencies were presented for the three study intervals to identify follow-up losses. Differences were estimated using chi-squared or Fisher’s exact test.
The latent variable was modeled using variable indicators. A good latent variable should have adequate convergent validity, showing that the indicator correlates with other indicators used to measure the same construct. Standardized factor loadings (SFL) greater than 0.60 indicate adequate convergent validity. A latent variable must also have adequate discriminant validity, when correlations between indicators are not excessively high (> 0.95), showing that each indicator measures different aspects of the construct. Negative loadings indicate an inverse association and positive loadings indicate a direct association31.
We adopted the SFL thresholds proposed by Kline31. Coefficients with values close to 0.10 indicate a small effect; around 0.30, a medium effect; and greater than 0.50, a strong effect. The tested models were evaluated using fit indices. The following values were considered acceptable: RMSEA (root mean square error of approximation) < 0.05; CFI (comparative fit index) and TLI (Tucker–Lewis index) > 0.95; and WRMR (weighted root mean square residual) < 0.95. Chi-squared, degrees of freedom, and P-values were calculated, but were not adopted as parameters to determine model fit due to the large sample size, which could influence the results of these tests.
The models were evaluated using mean- and variance-adjusted weighted least squares (WLSMV), indicated for the analysis of categorical data31. Theta parameterization was used to control for differences in residual variance. The automatic command “MODINDICES” was used to suggest modifications to the initial model. When the proposed modifications are considered theoretically plausible and the modification index is greater than 10,000, a new model can be developed and analyzed.
Ethical considerations
This study protocol was approved by the HUUFMA Research Ethics Committee (Reference Code 4771/2008-30). All mothers signed an informed consent form explaining the nature of the study. Thus, all methods were performed following relevant guidelines and regulations.