The BRFSS is a state-based telephone (random-digit-dialed) survey of the noninstitutionalized U.S. population aged 18 years of age and older that provides data related to chronic diseases and their risk factors [21
]. The BRFSS uses a Disproportionate Stratified Sample (DSS) method, where phone numbers are randomly selected throughout the state, business and nonworking numbers are omitted, and individuals aged 18 years and older are randomly selected from each household called. Data are subsequently weighted to reflect the complex sampling methods and nonresponse bias of the final sample [23
]. This survey provides annual population-based cross-sectional data that can be used to analyze self-reported risks and health conditions. The BRFSS includes national "core" questions and modules, and state-added modules on special topics of interest to specific states. The BRFSS has previously been used to identify prevalence and correlates of general health among people with disabilities [5
The present study analyzed data from the seven states and the District of Columbia that used both the core BRFSS Healthy Days measures (CDC HRQOL-4) and the HRQOL/Disability module each year from 1998–2000. The states (Arkansas, Iowa, Kansas, New York, North Carolina, Rhode Island, and South Carolina) and the District of Columbia represent a wide selection of the U.S. population. The total BRFSS sample size across all three years was 73,867. For the eight sites used in the study, response rates ranged from 52.2% (New York) to 75.1% (Kansas) with a median of 61.3% in 1998; in 1999, the range was 45.0% (New York) to 66.3% (Kansas) with a median of 48.7; in 2000, response rates ranged from 32.9% (New York) to 59.3% (North Carolina) with a median of 40.8% (see 2000 Behavioral Risk Factor Surveillance System Report [25
] for response rates for each state in each year). For these analyses, we limited the sample to respondents who were classified as having a disability and answered a question on their disability duration (n = 11,905). Our working definition of disability was based on respondents saying "yes" to either of two questions: "Are you limited in any way in any activities because of any impairment of health problem?" or "Do you now have any health problem that requires you to use special equipment, such as a cane, a wheelchair, a special bed, or a special telephone?"
For this study, the dependent variable of self-reported general health status was classified as a dichotomous variable (fair/poor vs. excellent/very good/good), based on answers to the question "Would you say that in general your health is excellent, very good, good, fair, or poor?" This question is included in the CDC HRQOL-4, along with three questions about the number of recent days of the last 30 when physical health was not good, mental health was not good, and activities were limited because of poor health. The CDC HRQOL-4 questions have been demonstrated to predict morbidity, health care use, and mortality and are associated with chronic diseases and disability [26
]. The retest reliability of HRQOL questions is moderate to excellent [4
]. The questions also have demonstrated reliability and validity for population health surveillance [4
] and people with disability [30
The primary predictor variable for these analyses was age at disability onset, computed from questions on self-reported disability duration and current age. Respondents who answered yes to one or more of the disability screener questions above were asked how long their activities had been limited. Responses were given in days, weeks, months, or years. These responses were recoded by the authors to indicate the number of years, or portions of a year (in decimal format), activities had been limited. This disability duration variable was subtracted from the respondent's current age on the date of the interview to determine the respondent's age at the time of disability onset. Age at disability onset was then categorized into four groups: birth through 21 years of age, ages 22 to 44, ages 45–64, and age 65 years or older. We classified early disability onset as birth through age 21 years based on federal laws designating developmental disability services for individuals aged 0–21 years. Adult onset was considered age 22 years and older; the additional divisions within this age range were made to allow examination of potential effects of early adult versus older adult disability onset. Thus, there were four groups classified by differing ages at disability onset, based on information reported at time of interview.
In addition, we included a set of potential confounding variables: age (as a continuous variable); gender; race/ethnicity (white non-Hispanic, African American non-Hispanic, other non-Hispanic groups, and all Hispanics); current employment status (employed, unemployed, student/homemaker, retired, unable to work); education (< high school graduate vs. ≥ high school graduate); marital status (married, separated/divorced, widowed, never married); and disability duration (years limited as a continuous variable).
Descriptive analyses compared characteristics of all four disability onset groups and respondents who were classified as not having a disability. Logistic regression models with health status as the outcome report odds ratios (OR) and 95% confidence intervals (95% CI) for people with disabilities only. Models were constructed by forcing in age at onset as the primary predictor variable, and including additional variables if they had a meaningful effect on the odds ratio of age at onset (10% or more change in OR) or if the variable itself was a significant predictor of general health status. We performed an exploratory analysis to investigate whether the relationship of age at onset and health status might be different for men versus women by adding an interaction term between gender and age at onset, but the interaction was non-significant. An interaction of age and disability duration was also tested; there was no significant interaction (data available on request). Thus, results are provided for main effects only. Descriptive results were analyzed using SUDAAN 9.0.0 (Research Triangle Institute, Research Triangle Park, NC, 2004) for weighted data, and logistic regression was conducted with SPSS Complex Samples 14.0 for Windows (SPSS, Inc., Chicago, IL, 2005). Stratification and weighting variables related to the BRFSS sampling and weighting strategy were included in the analyses as design variables. This study was approved by the Institutional Review Boards at the University of Florida and Oregon Health & Science University.