The Center for Disease Control and Prevention's BRFSS collects state-based data on preventive health practices and risk behaviors in the non-institutionalized civilian population aged ≥ 18 years . The analyses we report are derived from the 2001 BRFSS. Information on BRFSS design and sampling methods are reported elsewhere [7, 8].
Body mass index (BMI; kg/m2), calculated from self-reported weight and height, was the predictor.
Outcomes were interval since the most recent use of fecal occult blood test (FOBT), and sigmoidoscopy (SIG) in adults aged ≥ 50 years who reported ever having had the respective screening examination. BRFSS codes FOBT responses as: 'within past year', 'within past 2 years', 'within past 5 years', '5 or more years ago', 'don't know/not sure', or 'refused'. SIG is coded as: 'within past year', 'within past 2 years', 'within past 5 years', 'within past 10 years', '10 or more years ago', 'don't know/not sure', or 'refused'. Consistent with screening recommendations , we dichotomized FOBT as 0 = > 1 year since last screening vs. 1 = screened within the past year. For SIG, the American Cancer Society recommends screening every 5 years for adults aged ≥ 501. Thus, SIG was dichotomized as 0 = > 5 years since last screening vs. 1 = screened within the past 5 years.
We included age, education, race, income, self-reported general health status, smoking, employment, and health insurance as covariates.
We grouped respondents into 5 BMI-defined categories (18.5–<25 "normal weight", 25–<30 "overweight", 30–<35 "obesity class I", 35–<40 "obesity class II", and ≥ 40 "obesity class III"). Respondents (n = 250; .3%) with BMI's <18.5 ("underweight") were omitted from the analyses.
We used multivariate logistic regression to estimate BMI-screening associations by entering the BMI-defined categories and potential confounders into the model as either continuous (e.g., age [including polynomials up to the third order]) or dichotomous variables (e.g., health insurance). Using the guidelines proposed by Greenland , we retained covariates that were statistically significant at the two-sided 0.20 alpha level or caused a ≥ 10% change in any of the BMI-defined categories when deleted. As a result, education, income, self-reported general health, and employment were omitted. Responses coded as 'don't know/not sure', or 'refused' were treated as missing variables and excluded from analyses, as were respondents with missing data on any covariates. To ensure unbiased general population estimates, we used sample weights provided by the BRFSS. BMI categories were investigated as 4 contrasts with the normal weight category serving as the referent.
To evaluate whether sex moderated the BMI-screening association, we ran adjusted logistic models that also included BMI × sex interaction terms. Finally, because we observed a significant BMI × sex interaction, we then analyzed the data for men and women separately. Analyses were performed with SPSS 11.5.