Methods
For this study, 2006 BRFSS data were analyzed to examine if depression and rurality were important independent dimensions of the epidemiology of partial and/or full edentulism while controlling for other possible confounders such as SES, health behaviors, chronic diseases, and health service deficits. The BRFSS survey is comprised of both core questions and optional modules. We chose this year of data to analyze because the optional BRFSS adult depression module was used by 33 states and/or territories. In the subsequent years of available data, many fewer states chose this option. We analyzed data collected by questions from both the core survey as well as the optional adult depression module based on the Personal Health Questionnaire-8 (PHQ-8).
BRFFS data are collected using a random-digit dial telephone survey targeting adults 18 through 99 years of age. These data are collected under the guidance of the CDC in collaboration with all US states and most US territories. Once collected, the data are weighted by state or territory to represent the non-institutionalized US adult population based on the most recent census data available. BRFSS data are cross-sectional and are focused on health risk factors and behaviors as well as chronic disease. A detailed description of the survey design and sampling measures can be found elsewhere.
In the analyses presented here a number of variables were either re-coded or computed. Re-coding for the most part entailed collapsing response categories and removing the response categories of don't know and refused. The following variables were computed: chronic disease index, health service deficits, socioeconomic status and current depression.
Chronic disease index (CDI) entailed combining the variables of diabetes and cardiovascular disease. Anyone having one or both of the diseases was categorized as having at least one chronic condition related to edentulism.
Health service deficits, one of the independent variables in this analysis, was computed from the response categories of four separate variables (health insurance status, personal healthcare provider, deferment of medical care because of cost, routine medical exam). Health service deficits is a proxy for health care coverage and utilization since the BRFSS asks no questions about dental insurance. The response categories included in the computation of the variable were: did not have health insurance, did not have a healthcare provider, deferred medical care because of cost, and did not have a routine medical exam, all within the last 12 months. Together these four issues form a constellation of factors that can and often lead to deficits in care in the US health system. These four issues are interwoven and since health service deficits is an evolving concept they are given equal weight in this analysis. Having at least one of these constituted having a health service deficit.
SES was also one of the primary independent variables. SES is one of the strongest determinants of health. While it is a commonly used term in analyses across disciplines (e.g., sociology, social epidemiology, social psychology), there is no general consensus about how to either define or measure the construct. Typically SES refers to a combination of household income and other social measures such as attained educational level indexed into a single variable. The purpose of SES is to provide some means of comparing relative position with regard to others. Almost always, SES is computed as a three-level variable (i.e., low, middle and high). Various measures of SES are typically not interchangeable and reflect the intent and approach of the investigator. In our analyses, SES was a computed variable comprised of two categorical variables: attained education and median annual household income. In keeping with convention, data categories from each of these individual variables were coded as one of low, mid-range or high and numbered 1, 2 or 3 respectively. The variables with numbered factors or categories were then added together to create the composite variable of SES. For education, low was less than high school and was coded as 1, mid-range was high school graduate and was coded as 2, and high was at least some college and was coded as 3. For income, low referred to the category < $25,000 and was coded as 1, mid-range referred to $25,000 - < $50,000 and was coded as 2, and high equaled ≥ $50,000 and was coded as 3. When the individual variables were added together the possible computed range was 2–6 points. These points were then indexed in the following manner: low = 2–3 points, mid-range = 4–5 points and high = 6 points. These cut-points were purposive. For the lowest range of the index, 2 points were the floor (smallest possible point assignment), for the mid-range of the index, 4 points was the floor and likewise for the high range of the index, 6 points was the floor. Any points below the floor for the mid-range were assigned to the lowest index category just as any points below the floor for the highest index category were assigned to the mid-range index category.
The standardized and validated PHQ-8 was used to measure current depression. This validated instrument consists of eight of the nine criteria on which the Diagnostic and Statistical Manual of Mental Disorders 4 Edition Revised Text (DSM-IV-TR) diagnosis of depressive disorders is based. The ninth question in the DSM-IV-TR assesses suicidal or self-injurious thoughts. It is omitted because interviewers/researchers were not able to provide adequate intervention by telephone if a respondent indicates that they were having such thoughts. The PHQ-8 response set was standardized to make it similar to other BRFSS questions by asking the number of days in the past two weeks the respondent had experienced a particular depressive symptom. Similar to a methodology employed by other researchers, the modified response set was converted back to the original response set: 0 to 1 day = not at all, 2 to 6 days = several days, 7 to11 days = more than half the days, and 12 to 14 days = nearly every day, with points (0 to 3) assigned to each category, respectively. The scores for each item were summed to produce a total score between 0 and 24 points. A total score of 0 to 4 represents no significant depressive symptoms. A total score of 5 to 9 represents mild depressive symptoms; 10 to 14, moderate; 15 to 19, moderately severe; and 20 to 24, severe. This is summarized in Table 1. For our analyses, current depression was defined as: a PHQ-8 score of ≥ 10, which has 88% sensitivity and 88% specificity for major depression and, regardless of diagnostic status, typically represents clinically significant depression.
The Metropolitan Statistical Area (MSA) variable included in BRFSS was used to define place of residence as either rural or non-rural. Rural residents were defined as persons living either within an MSA that had no city center or outside an MSA. Non-rural residents included all respondents living in a city center of an MSA, outside the city center of an MSA but inside the county containing the city center, or inside a suburban county of the MSA.
Race and ethnicity was calculated from participant responses to two separate survey questions—one regarding race and the other regarding Latino/Hispanic ethnicity. All race/ethnicity categories were computed as mutually exclusive entities. For example, all respondents coded as Caucasian chose white as their racial classification, likewise, black for African American, etc. If a respondent identified themselves as Hispanic or Latino they were classified by that ethnic category regardless of any additional racial classification. The category of Other/Multiracial was also calculated.
Data analyses entailed both bivariate and multivariate techniques. Our population of interest was non-institutionalized adults completing the PHQ-8 depression screening tool as part of the 2006 BRFSS survey. Two logistic regression models were performed, one using partial edentulism as the dependent variable and the second using full edentulism as the dependent variable. SPSS (IBM, Chicago, Illinois) version 21.0 was used for all of the analyses. Alpha was set a p > =.05. This was a database study; as such human subjects' approval was not necessary.