BMI AND EARLY CHILDHOOD CARIES IN HIGH CARIES RISK CHILDREN: A NESTED CASE-CONTROL METHODOLOGICAL INVESTIGATION

Abstract

Purpose:

Inconsistently-associated, this study evaluated body mass index (BMI)-early childhood caries (ECC) relationships with various BMI expressions.

Methods:

Healthy 8–18 month children with un-erupted molar(s) were conveniently sampled from Uniontown, Alabama, a high caries risk community (i.e., rural, poor, African American). Staff-measured height/weight checks, dentist-conducted oral exams, and parent-completed questionnaires were performed annually (2008–2014) for BMI, ECC (decayed, missing due to caries, filled primary surfaces (dmfs) score), and socio-demographic values, respectively. Utilizing a nested case-control design (N=127), cases experienced ECC between annual visits; controls did not. Nationally-recognized standard (underweight-normal-overweight-obese), crude (overweight/obese-not), and continuous BMI variables were evaluated. Logistic regressions (with restricted cubic splines) assessed BMI-ECC relationships, producing odds ratios (OR) and 95% confidence intervals (CI).

Results:

Male and female ECC OR supported positive and negative parabolic functions, respectively, for increasing standard BMI categories; underweight males were associated with ECC (OR: 4.59; 95% CI: 1.06–19.85). Crudely expressed, overweight/obese males and females had lower and slightly increased odds of ECC, respectively. A continuous BMI produced similar OR across genders while spline models suggested non-linearity for each.

Conclusions:

BMI-ECC associations might be nonlinear; being underweight could be a male ECC risk factor. Studies should include extreme BMI values without collapsing BMI categories.

INTRODUCTION

Obesity and dental caries are two very common chronic childhood conditions in the United States (U.S.). Preschool-aged (i.e., two to five-year-old) childhood obesity rates have been on a significant upward trend since 2011 with 41.7% of these young children measuring at least overweight in 2016¹. Two out of every ten U.S. preschoolers have early childhood caries (ECC)². According to national epidemiological trends, the proportion of children with obesity and caries increases as children grow older^{1, 2}. Both chronic conditions disproportionately affect non-Hispanic Black, socioeconomically-disadvantaged, and/or rural residing U.S. children compared to their respective counterparts^1–4. Alongside minority and poorer socio-economic statuses, unwholesome dietary choices, sedentary behavior, and barriers to (quality) care are also linked to both childhood obesity and dental caries^5–8. Both are, among other factors, diet dependent and related to excessive, often refined, carbohydrates, leading to possibly overlapping etiologies for which biological plausibility has only been inferred in research thus far^9–11. Given the extensive overlap in risk factors and interest in identifying a potential causal mechanism, studies continue to hypothesize that obesity and dental caries are related. Yet, results of an association between the chronic conditions have been ambiguous, finding no, positive, and negative relationships⁸. A relationship could exist and the ambiguity might be methodologically-driven. Obesity-caries review articles often cite weaker (i.e., mostly cross-sectional vs. temporality/directionality data producing case-control/longitudinal) designs and a lack in range of BMI values as potential reasons for varying results^{6, 8, 12}. BMI is frequently investigated as a continuous variable, assuming a linear relationship with caries¹². If BMI is categorized, it is usually crudely dichotomized into vague groups (e.g., normal/non-normal BMI, obese/not obese)^{6, 8, 12–16}. The standard 4-category BMI variable is used infrequently^16–22.

Without a universal definition, overweight/obesity groupings vary by population, meaning inconsistencies in the obesity-caries relationship could be related to the chosen BMI variable type^{6, 8, 12}. Epidemiological studies are reliant on accurate quantification and expression of exposure measurements (e.g., BMI)²³. A continuous variable is preferred to a categorical one, when available, to maximize data usage. Variable expression can be statistical model-determined; categorical variables can be appropriate under certain study conditions. If linearity is assumed, but a nonlinear relationship exists, then variables may need to be expressed categorically to meet model assumptions and validate results¹². Smaller, more precisely refined categories function better in statistical analyses, strengthening detectable associations and minimizing measurement error and biased findings^{12, 23}. Inaccurately identified relationships can have far-reaching implications. Providers rely on epidemiological data to facilitate educational discussions with patients to reduce disease risks on an individual level. For communities, public health professionals analyze epidemiological data to evaluate current (and future) programs and target health messages accordingly. If the BMI-ECC relationship remains ambiguous because of BMI’s misrepresentation in caries research, dental provider and public health efforts will likely continue to be misinformed and, therefore, less effective at reducing ECC burdens.

Given that obesity and ECC are prevalent in our youngest ages^{1, 2}, sharing epidemiological trends and risk factors^5–8, researchers must better understand if/how childhood obesity and ECC are related so their disease rates can be reduced. A BMI-ECC relationship may exist, but the unsystematic quantification of BMI in prior research could be contributing to the current ambiguity in obesity-ECC findings¹². Higher quality study design choices (e.g., precise BMI measurement, diversity in participant BMIs, longer observation times) can address some BMI-caries epidemiological study weaknesses, allowing any association(s) between the chronic diseases to be unmasked^{6, 8, 12}. Therefore, the study objective was to evaluate relationships between various BMI variable types (i.e., continuous, dichotomous, 4-category) and incident ECC in a high caries risk population, using a nested case-control design, to determine if BMI quantification choices are influencing BMI-ECC analyses.

METHODS

This epidemiological methods-based investigation was ancillary to the Pediatric Dentistry Department of The University of Alabama (UAB) School of Dentistry’s longitudinal study entitled “Epidemiology of Dental Caries and Immunity in Children (Alabama)”. Children between eight and 18 months with un-erupted primary molars, no systemic diseases, and residence in Uniontown (Perry County), Alabama, were eligible for study enrollment. A sample size of 100 children would adequately power (75–80%) the longitudinal study to detect ECC risk factors as determined a priori with conservative participant attrition rate (i.e., 5%/year) and population estimates(e.g., r=0.3, caries prevalence >0.4, HR=2). Participants were recruited through convenience sampling (e.g., fliers, community center events, community advocates), including local daycares and the Berean Head Start program. Legal guardians provided informed consent and completed waivers of assent for their children due to age and maturity. Study protocols aligned with Helsinki Declaration guiding principles. UAB Institutional Review Board (Birmingham, AL) approved the study.

Between July 2008 and December 2009, ninety-three children enrolled in the study. Children were high caries risk due to Uniontown being rural, poor, and minority-populated as well as having limited access to dental care and minimally-fluoridated water systems. Because of study eligibility, children were homogeneous on some obesity-caries risk factors (i.e., age, socioeconomic status, urbanization level, minority status). Prior research also indicated that two-thirds of these children developed ECC after three years— twice that of national preschool-aged childhood caries prevalence^{2, 24}. Study visits occurred at baseline and annually for four additional years, collecting oral examination, height/weight measurement (follow-ups only), and questionnaire data at each visit. Appointment reminders were utilized to minimize participant attrition.

The original cohort study monitored caries development over time. A nested case-control study design was utilized for this methodological investigation to promote data usage in identifying (intermediate) BMI-ECC relationships. Described in detail below, cases (“new caries”) and controls (“no new caries”) were retrospectively identified. Caries status was investigated annually, making the outcome recurrent. Children reentered the study population after each annual interval for consideration within the succeeding interval, allowing participant BMI-ECC observations to be counted separately for each interval and maximizing the dataset. Back-dated exposure status was also retrospectively collected.

Routinely study-trained and calibrated (κ=0.93 for inter-examiner agreement) dentists conducted visual (i.e., no radiographic) oral examinations to determine ECC status annually²⁴. Only decayed teeth with cavitation were recorded as carious in this study. Children’s decayed, missing due to caries, filled primary tooth surfaces were summed to produce a dmfs score. Cases were children with annual changes in dmfs scores greater than zero between follow-up years one through four, indicating new disease was present by next study visit. Controls were children who had an annual dmfs change score of zero, meaning no new caries developed within the previous year. Annual dmfs scores were converted to a dichotomized outcome of “new caries” (incidence >0) vs. “no new caries” (incidence=0) for cases and controls, respectively. With the nested case-control study design, children could transition between case and control status during visit intervals. A baseline-year one dmfs change score was calculated, but was excluded because the parent study did not measure height and weight until age two (Fig. 1)²⁵. Negative dmfs change scores were possible outcomes. Because re-mineralization of cavitated surfaces is unlikely, ‘caries reversal’ scores were excluded to retain maximum internal validity.

Fig 1.

Schematic of nested case-control BMI-early childhood caries (ECC) association study methodology (N = 127). *Change in decayed, missing due to caries, filled surface (dmfs) scores between final and first oral exam visits per annual interval determined case/control status for high caries risk children; BMI was determined from first visit’s height (Ht)/weight (Wt) check per annual interval.

Study-trained staff used study-provided anthropometric protocols and equipment (i.e., stadiometer, scale) to measure height and weight annually. Values were directly entered onto questionnaires or in the database. Children’s annual BMI values were obtained from inputting their study-measured height (centimeters) and weight (kilograms) data into the traditional BMI formula. The Centers for Disease Prevention and Control (CDC)-supported age and gender specific percentiles were used in conjunction with widely research-accepted classification cut points to categorize BMI values^{8, 25}. BMI was independently analyzed continuously (traditional BMI formula numeric output without age/gender adjustments), dichotomously (overweight/obese (≥ 85^th percentile) vs. not (< 85^th percentile)), and standardly (underweight (< 5^th percentile), normal (5^th to < 85^th percentile), overweight (85^th to < 95^th percentile), obese (≥ 95^th percentile))⁷. Each child’s exposure status was determined from the BMI value at the earliest visit of a two visit interval (Fig. 1).

Guardians completed baseline questionnaires, providing children’s demographic (e.g., sex, age), birth history (e.g., full term birth status, childbirth methodology, birthweight), and dietary (e.g., breastfeeding status) and oral hygiene (e.g., toothbrushing status) habit data. Follow-up questionnaire data were also collected, but not utilized herein due to this study’s methodological focus and the uncertainty in timing for sociodemographic influences on children’s growth/development and oral health. Age (years) and birthweight (kilograms) were reported as continuous values. Sex (i.e., male/female), full term birth status (premature/full term), childbirth delivery (vaginal/cesarean), infancy breastfeeding status (yes/no), and toothbrushing status (yes/no) were captured with pre-specified, categorical responses.

Case-control baseline sociodemographic differences were evaluated by Chi-square/Fisher’s exact and t-tests (α=0.05 & 2-tailed P-values) for categorical and continuous variables, respectively. Potential associations between each BMI variable (i.e., continuous, dichotomous, standard) and ECC status (i.e., “new caries” (incident caries)/cases vs. “no new caries”/controls) were analyzed by three independent logistic regression models. Models were adjusted for full term birth status and stratified by sex to account for any growth and development differences. Adjusted odds ratios (OR) and 95% confidence intervals (CI) were reported for each model. To assess the potential for a dose-response relationship, restricted cubic splines (RCS), specifying quintile-based spline knots, were also incorporated in an additional adjusted regression model with the continuous BMI variable. “Not overweight/obese” and “normal” BMI values were the referent groups for the dichotomous and standard BMI logistic regression models, respectively. If BMI and/or caries status was missing for a child, the participant remained uninvestigated only for the associated annual time interval. SAS 9.3 (SAS Institute Inc., Cary, NC, USA) was utilized to conduct all statistical analyses.

RESULTS

Eighty-one children had three annual visit intervals each to contribute to this analysis whereby their case-control status could change annually. Relocation and no-shows resulted in non-participation. After removing children who had missing or negative dmfs scores and/or missing BMI data, 64 (34 male) ECC incident cases developed throughout the study. There were 63 (44 male) one-year intervals when no (new) caries developed for these children (Fig. 2, Table 1). Almost equally distributed in sample size for cases and controls, there were more male controls than male cases, trending towards statistical significance (P=0.05). Otherwise, the two groups were socio-demographically similar (Table 1). All parents/legal guardians reported children’s teeth being brushed at least once daily at baseline.

Fig. 2.

Schematic of study population for nested case-control study of high caries risk children. *dmfs = decayed, missing due to caries, filled surfaces. **Cases: annual early childhood caries (ECC) incidence > 0 (i.e., annual dmfs change score > 0); ^†controls: annual ECC incidence = 0 (i.e., annual dmfs change score = 0).

Table 1.

Sociodemographic characteristics of cases and controls (N=127) from a nested case-control study

Characteristic	Cases* (N=64)	Controls** (N=63)	p-value
Male†	34 (53.13%)	44 (69.84%)	0.05
BMI Continuous‡	16.11 (11.09–35.88)	16.89 (11.26–26.77)	0.31
BMI Crude Dichotomous†   Not Overweight/Obese	24 (70.59%)	26 (59.10%)	0.29
Overweight/Obese	10 (29.41%)	18 (40.90%)
BMI Standard 4-category†   Underweight	9 (26.47%)	3 (6.82%)	0.04 §§
Normal	15 (44.12%)	23 (52.27%)
Overweight	1 (2.94%)	7 (15.91%)
Obese	9 (26.47%)	11 (25.00%)
Female	30 (46.88%)	19 (30.16%)
BMI Continuous‡	17.56 (13.54–21.95)	17.54 (12.13–21.70)	0.65
BMI Crude Dichotomous†   Not Overweight/Obese	12 (40.00%)	8 (42.11%)	0.88
Overweight/Obese	18 (60.00%)	11 (57.89%)
BMI Standard 4-category†   Underweight	2 (6.67%)	3 (15.79%)	0.79§§
Normal	10 (33.33%)	6 (31.58%)
Overweight	8 (26.67%)	4 (21.05%)
Obese	10 (33.33%)	6 (31.58%)
Age‡ (years)
Follow-Up Year 1	1.77 (1.55–2.64)	1.87 (1.55–2.55 )	0.41
Follow-Up Year 2	2.95 (2.45–3.75)	3.09 (2.13–4.33)	0.15§
Follow-Up Year 3	3.94 (3.45–4.90)	3.97 (3.13–4.80)	0.37
Full Term Birth† (yes)	57 (89.06%)	51 (80.95%)	0.20
Vaginal Child Birth† (yes)	30 (46.88%)	28 (44.44%)	0.78
Birth weight‡ (kilograms)	3.20 (1.42–4.54)	3.06 (1.36–4.06)	0.46
Infant Breastfed† (yes)	1 (1.56%)	2 (3.17%)	0.62§§

^*cases = “new caries” or decayed, missing, filled surfaces (dmfs) change score (dmfs_{follow-up year #} - dmfs_{follow-up year # - 1}) >0

^**controls = “no new caries” or dmfs_{follow-up year #} - dmfs_{follow-up year # - 1}=0

^†categorical variable: frequency (percentage), Chi-square test

^‡continuous variable: median (minimum–maximum), t-test

^§unequal variance

^§§Fisher’s exact test

Bold=(trending towards) statistical significance (α ≤0.05)

Overall, approximately 13.3% of children were underweight (< 5^th percentile), 42.5% were normal (5^th to < 85^th percentile), 16.5% were overweight (85^th to < 95^th percentile), and 27.6% were obese (≥ 95^th percentile), equating to 44% being overweight/obese. Most male cases and controls had similar BMI values that were not overweight/obese. With about half of each group having normal BMIs, the remaining males were classified as underweight, overweight, or obese cases and controls, resulting in a significant relationship between case-control status and the standard 4-level BMI variable (P=0.04). Contrarily, most female cases and controls had similar BMI values which fell within the overweight/obese category with as many as a third of each group being obese and another third of each having normal BMIs, attenuating any female case-control and BMI association (Table 1).

When BMI was investigated standardly with four categories (referent=“normal”), an underweight male had a 4.5 times greater odds of having incident ECC than a normal weight male (OR: 4.59, 95% CI: 1.06–19.85). Remaining male ORs were much lower and non-significant as BMI groups became greater in BMI value. In other words, male ECC ORs were a parabolic function of increasing BMI categories. For females, underweight and obese categories implied a lesser odds for incident ECC while the overweight group exhibited a slightly higher OR. In essence, female ECC ORs were a negative parabolic function of increasing BMI categories. None of these relationships were significant (Table 2). Being overweight/obese (vs. not) suggested a non-significant, slightly reduced and increased odds of ECC for males and females, respectively (Table 2). Considering BMI continuously (i.e., traditional BMI formula numeric output unadjusted for age/gender), each additional BMI unit increase did not significantly increase or decrease a male or female child’s odds for incident ECC development. Males potentially saw a minor, non-significant reduction in disease odds as their BMI increased (Table 2). In the adjusted RCS model, the continuous BMI variable was not associated with ECC development, but suggested dose-response relationships parabolic in nature with u-shape and inverted u-shape curves for males and females, respectively (data not shown).

Table 2.

Odds ratio estimates^* of early childhood caries by BMI variable type in young, high caries risk children (N=127).

BMI	Males (N=78)	Females (N=49)
Continuous	0.94 (0.82–1.07)	1.00 (0.76–1.31)
Crude dichotomization (ref=not, nmale=50, nfemale=21)  Overweight/Obese (nmale=28; nfemale=28)	0.61 (0.24–1.59)	1.06 (0.31–3.63)
Standard 4-category (ref=normal, nmale=38, nfemale=16)  Underweight (nmale=12, nfemale=5)	4.59 (1.06–19.85)	0.41 (0.05–3.51)
Overweight (nmale=9, nfemale=12)	0.41 (0.08–2.28)	1.12 (0.22–5.71)
Obese (nmale=19, nfemale=16)	1.18 (0.38–3.69)	0.67 (0.14–3.19)

^*Logistic regression models stratified by sex and adjusted for full-term birth status only

Bold=statistical significance (α ≤0.05).

DISCUSSION

This study’s findings suggested that BMI’s association with ECC might be sex and measurement dependent. The standard (4-category) BMI variable indicated nonlinear relationships--opposite in direction for each sex. Underweight males had a significantly greater odds for ECC development than “normal” males. When the BMI variable was dichotomized, resulting in broader BMI categorization, the underweight BMI-ECC relationship was attenuated.

The parabolic/nonlinear relationship demonstrated here with the four-category standard BMI variable is not seen elsewhere¹², but a few recent childhood caries studies use this BMI variable. Their findings are mixed—positive^17–19, negative²⁰, and no^{21, 22} associations. None occur within the U.S.^17–22, and most are cross-sectional^{17, 18, 20–22}. The one case-control study involves Canadian children and severe ECC only¹⁹. Unlike the current study, a majority of these 4-category BMI studies include older children^{17, 20–22} and do not adjust for potential confounders in statistical analyses^17–19. These studies also have fewer children in the extreme BMI categories (i.e., underweight, obese), containing mostly “normal” BMI-valued children^{12, 17–22}; other studies only investigate a tri-category BMI variable due to a lack of children spanning the entire BMI continuum^{13, 15, 16, 26, 27}.

Other studies also find underweight children to be at greater risk for or experience more disease8, ^{12, 14, 16, 18, 20, 28-30} when compared to others. Most are in non-U.S. children^{16, 18, 20, 28–30}, cross-sectional^{16, 18, 20, 28, 29}, and measure BMI as a tri-category variable^{14, 16, 28, 29}. One case-control study occurs in the U.S., but it is restricted to older (6–11-year-old) children, fewer (underweight/healthy combined) BMI categories, and decayed teeth¹⁴. While the other case-control study finds significantly more underweight children in the ECC vs. caries-free group, the study occurs outside of the U.S. and measures BMI continuously, assuming a linear association with ECC and finding no overall BMI-ECC relationship³⁰.

In general, underweight children can have additional medical issues which likely contribute to their increased risk for worse health outcomes overall, potentially including, but not limited to, dental caries³¹. Specific to caries, children’s lower BMI values can stem from chewing difficulties, infrequent/smaller feedings, less nutritious diets, and/or malnourishment which may lead to vitamin deficiencies that, in turn, weaken tooth development and restrict saliva flow—a caries-susceptible environment^{8, 12}. Untreated caries in underweight children can cause tooth pain which contributes to poorer eating habits and sleep quality, further impeding proper growth and development and enabling caries to progress^{8, 16}. Having a lower household income increases one’s risk for both underweight and caries diagnoses^{8, 12}.

As in the current study, recent investigations measuring BMI continuously or crudely (dichotomously) do not find associations with dental caries^{7, 16, 22, 30, 32–37}. This methodological investigation sought to investigate how BMI should be expressed in ECC research studies so that any potential BMI-ECC relationships can be identified. Our findings support the idea that model selection can mask the true association simply because of model-variable incompatibility¹². For example, the current study’s standard (4-category) BMI variable, which can be linearly or nonlinearly related to the outcome, demonstrated a nonlinear association with ECC while the continuous BMI variable, which assumes a linear relationship, did not suggest any significant association. When splines were added to the continuous model, which BMI-caries studies typically do not do, non-linearity was further suggested when probabilities of ECC development were plotted against BMI values, demonstrating parabolic functions for each sex. Variable dichotomization can lose some important data with broader groups and attenuate existing relationships^{12, 16, 23} as was the case when comparing the current study’s crude/dichotomized and standard (4-category) BMI model findings. In other words, poor variable expression choices can invalidate statistical tests which impedes the identification of any association between an exposure and outcome of interest (e.g., BMI and ECC)²³.

The current study serves as a strong preliminary investigation into a potential methodological issue related to BMI-ECC association determination. This study’s nested case-control design provided repeated BMI and ECC measurements, potential establishment of temporality between the chronic conditions, and determination of intermediate BMI-ECC associations occurring during children’s growth/development. Annual (vs. global) dmfs change scores provided a more accurate measure of caries development over time. BMI’s precise measurement allowed the exposure to be expressed in alternative forms. BMI percentiles (vs. Z-scores) facilitated results interpretation for the research’s intended audience (e.g., providers, public health, parents). Use of highly reliable (κ=0.93)²⁴, National Institutes of Health-funded prospective data further strengthened this methodological investigation. Including children who had BMI values that spanned the entire BMI continuum was another study strength.

Homogeneity on known obesity-caries risk factors (i.e., age, socioeconomic status, urbanization level, minority status, diet/breastfeeding, oral hygiene/toothbrushing) minimized the need for additional adjustment for these covariates/confounders in statistical modelling, maximizing study power. Sex stratification identified BMI-ECC differences in males and females. Sexes can have varying growth/development (e.g., body fat/structure, tooth eruption time) and dietary (e.g., type, amount, frequency) patterns—factors relating to both BMI and ECC⁸. Fullterm birth status adjustment accounted for pre-birth growth/development effects. Premature birth can be linked to both childhood obesity (e.g., “catch-up” growth), and poor tooth development with enamel deficiencies (higher caries risk) and/or delayed eruption (lower exposure time)^{38, 39}. Childbirth methodology was also included since C-section babies can test positive for Streptococcus mutans much earlier than vaginally delivered children; likely a dietary (i.e., sugar) initiated disease, earlier bacteria attainment can lead to more (severe) ECC^40–42.

This study also had limitations. Only baseline questionnaire data were utilized, omitting additional oral hygiene data due to infancy age and changes in dietary habits over time. The parent study protocol did not begin height and weight measurements before age two; therefore, assessment of earlier BMI values were unavailable for incorporation in this methodological investigation, resulting in some caries data going unused. The dmfs score may have underestimated the ECC burden, ignoring non-cavitated lesions. The International Caries Detection and Assessment System is a more sensitive measure, but it is a more time-intensive and intricate (data collection, examiner calibration) analytical approach that can inaccurately increase measured disease amounts^{43, 44}. The dmfs score’s more conservative ECC estimate was chosen in exchange for assurance in true disease measurement, ease of use and interpretation, validity and reliability, and alignment with current ECC literature and the World Health Organization⁴⁴. The study had moderate participant attrition, impacting statistical significance during analyses, despite implementation of advanced scheduling/reminder efforts and a nested case-control study design to maximize annual observations. The study included a minimal number of underweight study participants which is reflective of the amount of obesity in Alabama. In addition to convenience sampling efforts for recruitment, the reduced and specific study population may restrict the study’s generalizability to like cohorts, necessitating future studies in differing populations.

These contradictions in the childhood obesity-ECC relationship are problematic with heightened rates of both chronic diseases in the youngest US population^{1, 2}. To clarify any BMI-ECC association, future ECC methodological research studies should determine if childhood BMI is nonlinearly-related to the dental disease in larger and more diverse epidemiological studies¹². BMI data should be collected as a continuous variable to allow for investigation of various configurations of BMI values (e.g., continuous, dichotomous, standard 4-category) with ECC data to determine the true population-specific exposure-outcome relationship. Accurate population level data will provide a better understanding of the current BMI-ECC association which, in turn, can be utilized in various public health efforts to decrease both childhood disease risks.

CONCLUSIONS

1. Dichotomized or (non-spline incorporating) continuous BMI values could be obscuring the BMI-ECC relationship.

2. BMI and ECC might be nonlinearly associated with “underweight” males having an increased risk for ECC.