Bivariate and multivariate techniques were used to analyze a multiyear Behavioral Risk Factor Surveillance Survey (BRFSS) heart attack and stroke module database. BRFSS data are collected using a random-digit dial telephone survey targeting adults 18–99 years of age. These data are collected under the guidance of the Centers for Disease Control and Prevention (CDC) in collaboration with all U.S. states and most U.S. territories. Once collected, BRFSS data are weighted such that they are representative of the non-institutional U.S. population by surveyed state. The data are cross-sectional and are focused on health risk factors and behaviors. A detailed description of the survey design and sampling measures can be found elsewhere [23
Data from the BRFSS optional module on heart attack and stroke were used in these analyses. Because different states use this module in different years, we merged 2005, 2007, and 2009 data to include as many states and territories as possible. In 2005, 14 states, the District of Columbia and United States Virgin Islands (USVI) included a module in their BRFSS surveys regarding symptoms of heart attack and stroke. In 2007, 13 states, the District of Columbia and the USVI included the module. In 2009, it was included by 19 states and the District of Columbia. Data from 25 states, the USVI, and the District of Columbia were used in these analyses. If a state used the module more than once, only the data from the most recent year were used. For the years in question, the BRFSS heart and stroke module included 13 questions focused on ascertaining knowledge of early symptoms of heart attack and stroke. Of these 13 questions, six were on knowledge of stroke symptoms, six were on knowledge of heart attack symptoms, and one question was on proper first response to either stroke or heart attack.
Respondents were asked if the following were warning signs of stroke: sudden confusion; trouble speaking or understanding; sudden numbness or weakness of face, arm, or leg; sudden trouble seeing in one or both eyes; sudden trouble walking, dizziness, or loss of balance or coordination; or sudden, severe headache with no known cause. An incorrect sign (i.e., sudden chest pain) was included to examine the possibility that respondents would answer “yes” for all the symptoms. Likewise, respondents were asked if the following were warning signs of a heart attack: pain or discomfort in the jaw, neck, or back; feeling weak, lightheaded, or faint; chest pain or discomfort; pain or discomfort in the arms or shoulders; shortness of breath. As was the case with stroke symptoms, an incorrect sign (i.e., trouble seeing in one or both eyes) was included to examine the possibility that respondents would answer “yes” for all the symptoms.
We chose to group the questions for heart attack and stroke symptomology together for analysis because these disorders are both vascular events that require the need for prompt recognition of symptoms and pre-hospital action by either the patient or bystanders. Any costly public health campaign will likely need to address both these vascular diseases together, and strokes are often referred to as “brain attacks,” as many aspects of early stroke management mimic heart attack management [24
For analysis we computed a Heart Attack and Stroke Knowledge Score for each respondent. Correct answers received one point and were categorized according to the following scale: low score 0–5 points, midrange score 6–9 points and high score 10–13 points. Although this scale, like most, is somewhat arbitrary, we based the cut points on the actual range (0 – 13) derived from responses. This scale served the purpose of allowing for the standardized comparison of knowledge levels.
The covariates for the analysis were: sex, age, race/ethnicity, annual household income, education attained, marital status, geographic locale, timing of last routine medical checkup, having a personal health care provider (HCP), having health insurance, deferment of medical care because of cost, and self-defined health status.
The Metropolitan Statistical Area (MSA) variable included in BRFSS was used to define geographic locale and was recoded into the dichotomous categories of rural or non-rural. Rural residents were defined as people living either within an MSA that had no center city or outside an MSA. Non-rural residents included all respondents living in a center city of an MSA, outside the center city of an MSA but inside the county containing the center city, or inside a suburban county of an MSA.
Race and ethnicity was also a computed variable calculated from participant responses to two separate survey questions—one regarding race and the other regarding Latino/Hispanic ethnicity. Combining the responses to these two questions allowed for the derivation of the race and ethnicity variable used in the analyses. 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, they were classified by that ethnic category regardless of any additional racial classification. The category of Other/Multi-racial was also calculated. All racial categories were non-Hispanic.
A number of additional original BRFSS variables—age, education attained, marital status, and annual household income—were recoded for these analyses. Age was recoded from a continuous variable to a categorical one with three factors/levels (18–44 years, 45–64 years, and
=65 years). For education, marital status and household income, recoding entailed the collapsing of multiple response categories into fewer ones—three for education and two factors each for household income and marital status. The three factors for education were less than high school, at least high school, and university graduate. Annual household income was categorized into the categories of
$50,000. Marital status was collapsed into the categories of married or living with a partner and not married or living with a partner. Three health care access variables: having a health care provider, deferment of medical care because of cost, and health insurance status were also included in these analyses. The response categories of “don’t know” and “refused to answer” were recoded as missing data and removed from the analyses.
Bivariate analysis describing the three-level composite scores by each covariate was conducted as well as a description of correct and incorrect answers for each heart attack and stroke symptomology knowledge question by geographic locale (non-rural, rural and total U.S.).
Two logistic regression models were performed using low scores on the combined heart attack and stroke knowledge questions as the dependent variable. All U.S. adults were included in the first model and only rural adults>
18 years of age responding to the survey questions were included in the second. The covariates entered into the two models were: sex, age, race/ethnicity, education attained, marital status, annual household income, geographic locale (first model only), health status, having a personal HCP, health insurance status in the past 12 months, deferring medical care in the past 12 months because of cost, and timing of last routine medical check-up. Alpha was set at 0.05 for all tests of statistical significance. To further reduce bias, a constant was entered into the models. SPSS version 19.0 (SPSS, IBM, Chicago, IL) was used to complete the analyses to account for the complex survey design. Human subjects’ approval was sought and received from Essentia Health’s Institutional Review Board (IRB).