Data source
Data from the 2011–2014 National Health and Nutrition Examination Survey (NHANES) were used for this study. NHANES is a cross-sectional survey conducted by the National Center for Health Statistics (NCHS) of the Centers for Disease Control and Prevention (CDC) that uses a stratified, multistage sampling design to obtain a representative probability sample of non-institutionalized, civilian population of all ages in the US. NHANES data are used to assess the health, nutritional status and health behaviors of the eligible population using self-reported responses, standardized physical examinations and laboratory tests. The NHANES protocol was developed and reviewed to be in compliance with the HHS Policy for Protection of Human Research Subjects (45 CFR part 46). The protocol was approved by the NCHS Research Ethical Review Board and underwent annual review. Sample persons were informed of the survey process and their rights as a participant by interviewers and by written materials. Written informed consent was freely obtained from each individual participant. Only de-identified observations were used and presented in aggregated summary form. The current study combined the 2011–2012 and 2013–2014 NHANES cycles to derive a study sample covering the 2011–2014 period. Data from both cycles were collected via in-home interviews, with health examinations and laboratory tests conducted in mobile examination centers (MEC). Data for this analysis were obtained from responses to the home interviews and results from the dental examinations. The home interviews used a structured questionnaire that assessed demographic and socioeconomic characteristics and various health-related issues, including oral health. The dental examinations were conducted by trained dentists in the MEC on all eligible participants aged 1 year and older. The 2011–2012 and 2013–2014 NHANES data sets followed the same protocols and are in the public domain, analyzed using only de-identified observations and presented in aggregated summary form. Further details on NHANES survey sample design are provided elsewhere [20].
Study population
The available data set of the 2011–2014 NHANES comprised 19,931 participants of all ages. From this study population, 8602 participants who were younger than 20 years were excluded. Additionally, all participants aged 20 years and over with no dental examination information (422) and those with missing observations (1661) were excluded. A complete flowchart of participants is available in Additional file 1.
Published response rates were 73% and 70% respectively for the interview and examination samples in the 2011–2012 cycle and 71% and 69% respectively for the 2013–2014 cycle [21, 22]. To increase the precision of estimates, NHANES oversamples some population subgroups. For NHANES 2011–2014, the primary change was the addition of an oversample of Asian persons. Oversampling was also carried out for Hispanic, non-Hispanic black, low-income white persons, and older white adults aged 80 years and over [20]. Sampling weights used in this data analysis are provided in the data files which are in public domain.
Variables
The outcome was derived from the question “About how long has it been since you last visited a dentist?” Participants who responded “6 months or less” or “more than 6 months but not more than 12 months” were defined as having a dentist visit in the past 12 months. Participants who responded “more than 12 months” or “never have been” were classified as not having a dentist visit in the past 12 months.
The income variable was based on the poverty income ratio (PIR) calculated by dividing family income to the poverty level threshold specific to family size and survey year. The PIR followed the HHS federal poverty guidelines (FPG) which are issued each year, in the Federal Register, for determining financial eligibility for certain federal programs such as Head Start, Supplemental Nutrition Assistance Program (SNAP) and the National School Lunch Programs. Additional information can be located at http://aspe.hhs.gov/poverty/11poverty.shtml. Ratios below 1.00 indicate that family income is below the official definition of poverty and ratios of 1.00 or greater indicate family income at or above the poverty level. Income was categorized into four PIR groups [23]: PIR greater or equal to 3.00 (high income); PIR greater or equal to 2.00 and less than 3.00 (middle income); PIR greater or equal to 1.00 and less than 2.00 (near poor); and PIR less than 1.00 (poor).
The wealth variable was constructed using a combination of family monthly income and home ownership variables. We chose to use wealth as a socioeconomic variable in addition to income because some authors have argued that income in combination with assets such as housing is a better indicator of health than income alone [24]. Wealth was categorized into three groups – high, middle and low wealth groups. High wealth corresponds to participants in households with more than $2900 USD of monthly income and who are home owners. Middle wealth corresponds to participants with either more than $2900 USD of monthly income and who are non-home owners or are home owners with less than $2900 USD of monthly income. Low wealth corresponds to participants in households with less than $2900 USD of monthly income and who are non-home owners. The choice of $2900 USD cut-off point was influenced by NHANES data reporting. The monthly income variable made available to the public is not presented as a continuous variable but rather as a range value in dollars. For example, $0 - $399 coded as 1, $400 - $799 coded as 2, $800 - $1249 coded as 3 etc. Therefore, our final binary categorization ensured participants’ income is as close to the US median value as possible and are equally divided.
Other variables used in the analyses were chosen based on Andersen’s behavioral model of health service use [25]. Predisposing factors included sex, age, race/ethnicity, country of birth, marital status, education, and smoking status. Age was categorized as 20–44 years, 45–64 years, and 65 years and over. Race/ethnicity was recoded as non-Hispanic white, Hispanic, non-Hispanic black, and non-Hispanic Asian. Country of birth was dichotomized as born in the US or born in another country. Marital status was recoded as married or cohabiting, widowed or separated, and never married. Educational attainment was classified as having more than 12 years, 12 years, and less than 12 years of school. Smoking status was categorized as current, former, and never smokers. Job status was an enabling factor being expressed as having a job versus not having a job. Need factors were self-rated oral health and the presence of untreated caries. Self-rated oral health was based on the question “Overall, how would you rate the health of your teeth and gums?” Participants were defined as satisfied if they responded excellent, very good or good, and not satisfied if they responded fair or poor. Presence of untreated dental caries (yes/no) was derived from the NHANES clinical dental examination.
Statistical analysis
Only records with complete data on all study variables were analyzed. The analyses were carried out using STATA 13 software (StataCorp, 2013). Sample weights and survey design variables were used to account for unequal probability of participant selection, nonresponse, and sampling error. Absolute numbers, weighted percentages, and standard errors (SE) were estimated to assess the prevalence of not having a dentist visit in the past 12 months. Differences between percentages were calculated by using two-sided t-tests at the α = 0.05 level following recommended NHANES analytical and tutorial guidelines. Because education is associated with income and wealth in some populations, we assessed for collinearity between income and wealth with education and found weak correlation: 0.39 (income) and 0.25 (wealth).
Unadjusted (univariable) and parsimonious Poisson regression models were derived separately for income and wealth using generalized linear models with long link and Poisson distribution. The models described associations between income, wealth, socio-demographic factors, health behaviors and need factors and not having a dentist visit. The Poisson regression models were first adjusted for age, sex, race/ethnicity, country of birth, marital status, education, smoking status, job status, self-rated oral health and presence of untreated caries. The criterion for inclusion in the parsimonious models was p < 0.05 in the full adjusted models. Associations are presented as prevalence ratios (RRs) with 95% confidence intervals (CIs). The income and wealth models were compared using the Akaike Information Criterion (AIC) as a means of model selection. The AIC estimates the relative information lost when a given model is used to approximate reality. It selects the better model given a set of available data. A preferred model is the one with a minimum AIC valuee [26, 27]. Due to significant interactions by race/ethnicity, the results in the parsimonious models were stratified by race/ethnicity.