Data source
We used data from the National Health and Nutrition Examination Survey (NHANES) 2003–04, which is part of a series of surveys conducted by the Center for Disease Control and Prevention (CDC) to assess the health and nutritional status of the non-institutionalized civilian population aged 2 years and older residing in the USA [20]. The 2003–04 NHANES is the only survey that included an expanded oral health section providing information on oral health conditions that were never assessed previously and have not been assessed again since then.
The selection of survey participants was conducted by stratified, multistage probability sampling to recruit a representative sample. Oversampling was conducted on certain subgroups which included non-Hispanic blacks, Mexican Americans, and participants aged 12 to 19 years and 60 years or more. Data were collected through phone interviews, face-to-face interviews, and clinical examinations. A total of 10,122 individuals of the eligible population agreed to take part in the survey (79% response rate). The expanded oral health examination was conducted on participants aged 13 years and older and included 8272 participants (76%) [20]. The NHANES 2003–04 survey included 4308 participants aged 18 years and over with complete tooth wear. We excluded 767 participants because of missing data on obesity, dietary data, and confounders. Therefore, the study sample included 3541 participants.
Variables selection
The outcome variable for this study was tooth wear, which was assessed using the modified tooth wear index (TWI), has been used in the UK Adult Dental Health Survey [21]. The clinical assessment included visual examination of the facial, lingual, and incisal surfaces of the 12 anterior teeth as well as occlusal surfaces of the first molars (40 tooth surfaces overall). Tooth surfaces were scored as sound (any wear restricted to the enamel); mild tooth wear (loss of enamel just exposing dentine); moderate tooth wear (dentine exposure greater than one-third of the surface); or severe tooth wear (complete loss of enamel, pulp exposure, or exposure of secondary dentine). Oral clinical examinations were conducted by two trained and calibrated examiners. Inter-examiner reliability for the tooth wear examination gave a percent agreement of 87.7% and a kappa of 0.80 [20]. The number of surfaces with moderate-to-severe tooth wear was the outcome for this study.
The exposure was obesity measured using the body mass index (BMI). Trained health technicians measured participants’ weight and height using standard protocols and equipment [22]. BMI was calculated as body weight (in kilograms) divided by the square of height (in meters) and used to classify participants as normal (BMI < 25), overweight (25 < BMI < 29.9), or obese (BMI > 30).
Several confounders were also selected for analysis based on previous studies [23, 24]. A basic set of confounders included demographic characteristics (sex, age, race/ethnicity), socioeconomic position (education and poverty income ratio), acid reflux medication (antacids and anti-reflux medication, as a proxy for gastroesophageal reflux symptoms), and dental insurance coverage. The key confounder in our analysis was the intake of sugar-sweetened acidic drinks, which was estimated from two non-consecutive 24-h recalls [25]. The first 24-h dietary recall was performed in person by trained interviewers in the mobile examination center, while the second 24-h dietary recall was conducted by telephone 3 to 10 days after the examination. Dietary recall data were used to measure an average intake (drinks/day) across the 2 days of four different categories of drinks: sugar-sweetened acidic drinks (SSA: sweetened soda, non-100% fruit juice, sweetened sports drinks), sugar-sweetened non-acidic drinks (SSNA: sugar-sweetened coffee, tea- and milk-based products with sugar), non-sugar-sweetened acidic drinks (NSSA: diet soft drinks, 100% fruit juice or flavored sparkling water), non-sugar-sweetened non-acidic drinks (NSSNA: non-sugar-sweetened coffee, tea- and milk-based products). The specific USDA food codes included in each of the four categories of drinks are listed in the Appendix.
Data analysis
Data were analyzed using Stata/SE version 14.1 (Stata Corp., College Station TX, USA). All analyses were weighted to produce a representative sample and to take account of the variation in cluster size and different sampling rates for clusters within strata.
The distribution of the number of surfaces with moderate-to-severe tooth wear was positively skewed, with many participants having zero counts. Therefore, a hurdle model was preferred to model associations with the number of surfaces with moderate-to-severe tooth wear. Hurdle models account for the excessive proportion of zero values and over-dispersion [26, 27]. They include two components. The first component estimates the probability of being a non-zero (positive) count from a binary logit model. Odds ratios (OR) were reported for this component of the model, which represent the probability of having moderate-to-severe tooth wear. The second component estimates the mean number of surfaces with moderate-to-severe tooth wear among those with positive counts only (those with tooth wear) using a truncated negative binomial model. Rate ratios (RRs) were reported for this component, which represent how higher/lower the mean number of surfaces with moderate-to-severe tooth wear was in one group compared to the reference group, among those with the condition [26, 27].
We first examined the association between each type of drink (SSA, SSNA, NSSA, NSSNA) and the number of surfaces with moderate-to-severe tooth wear through a series of regression models: Model 1A was unadjusted; model 1B was adjusted for confounders (sex, age groups, race/ethnicity, education, poverty income ratio, dental insurance, and acid reflux medication); and model 2C was additionally adjusted for the remaining types of drinks. We then examined the association between BMI and the number of surfaces with moderate-to-severe tooth wear through a series of regression models: model 2A was unadjusted; model 2B was adjusted for confounders (sex, age groups, race/ethnicity, education, poverty income ratio, dental insurance, and acid reflux medication); models 2C to 2F were additionally adjusted for one type of drink at a time (model 2C: SSA, model 2D: SSNA, model 2E: NSSA, model 2F: NSSNA); and model 3F was adjusted for confounders and all four types of drinks simultaneously.