Design and study population
Between 2002 and 2009, the SCCS enrolled approximately 86,000 adults (over two-thirds black) aged 40–79 living in 12 southeastern states. Approximately 86 % of participants were recruited at community health centers (CHC), which provide primary health and preventive care services for low-income populations [11, 23], while the remaining 14 % were recruited via mail-based general population sampling. Data on socioeconomic, demographic, lifestyle, and anthropometric characteristics, as well as personal medical history, were ascertained at cohort enrollment via standardized computer-assisted personal interviews for CHC participants, and via self-administered mailed questionnaire for general population participants. Detailed description of SCCS methods has been previously published [11, 23].
The study population for the current nested case–control study was restricted to CHC-enrollees, which ensured that participants were of similar socioeconomic status and had generally equal access to health care at cohort entry regardless of race. Incident ESRD cases were ascertained by linkage of the cohort, using date of birth, Social Security number, and first and last name, with the US Renal Data System (USRDS) from January 1, 2002 to September 1, 2009, the latest date for which data were available. The USRDS registers ESRD cases certified by a physician diagnosis and filed using a medical evidence report form (to the Medicare ESRD program) or when there is other evidence of chronic dialysis or a kidney transplant irrespective of the glomerular filtration rate (GFR) [1, 24]. SCCS participants (n = 404) who had a diagnosis of ESRD recorded in the USRDS prior to SCCS enrollment were excluded from our analyses. Three controls were individually matched to each case based on age (±5-year categories), sex and race. Therefore, the study population for the current analysis comprised all 631 ESRD cases identified during the study period and 1,897 matched controls; and with this sample size we had over 82 % power to detect an odds ratio of 1.48.
Assessment of BMI and covariates
The main exposure variable in this study is BMI, defined as weight (kg)/height2 (m2), calculated from weight and height self-reported by participants at cohort entry. Participants also reported their weight at age 21. Since BMI at age 21 can indicate long-term exposure to obesity and, given the relatively short follow-up period (median 2.6 years, range: 0–7.2 years) of the cohort, it may be more closely associated than current BMI with an insidious outcome such as ESRD. BMI at age 21was thus considered our primary exposure of interest. We categorized BMI at age 21 using the World Health Organization (WHO) classification as: underweight: <18.5 kg/m2; normal weight: 18.5–24.9; overweight: 25.0–29.9; obese: ≥ 30.0 , and for BMI at cohort entry, further divided the obese category into class I (30.0–34.9), class II (35.0–39.9) and class III (≥40.0).
Univariate case–control comparisons were performed using T-tests for continuous variables and chi-square tests for categorical variables; 2-sided p-values were presented. The Pearson’s correlation coefficient between BMI at enrollment and BMI at age 21 was computed.
Using normal BMI as the referent, we used conditional logistic regression to estimate odds ratios (OR) and corresponding 95 % CI for ESRD associated with the other BMI categories, overall and stratified by race. We performed the main analyses using BMI at age 21 and, in separate models, considered BMI at enrollment. We investigated race × BMI interactions using a Wald test for the race × continuous BMI interaction term. In addition to the matching variables, covariates included in the main analyses were education (<high school, high school/vocational training/junior college, ≥college), and cigarette smoking (never, former, current). Differences between the crude and adjusted estimates were minimal but we present the adjusted estimates. We did not include history of diabetes or hypertension in our main models given that these may be on the causal pathway between elevated BMI and ESRD. However, in sensitivity analyses, we adjusted additionally for diabetes and hypertension at baseline, in order to examine the magnitude of the association between BMI at age 21 and ESRD independent of these two main intermediates. Also, in order to investigate the influence of short follow-up on the BMI at enrollment-ESRD association, we performed additional sensitivity analyses by successively excluding cases diagnosed within 12, 24 and 36 months after enrollment.
In subsequent analysis, BMI at 21 was modeled using restricted cubic splines with 5 equally spaced knots chosen according to Harrell’s percentile distribution: 0.05, 0.275, 0.5, 0.725 and 0.95 corresponding to BMI values of 17.2, 20.7, 22.9, 25.7, 33.8 kg/m2 respectively . Predicted probabilities of incident ESRD were then computed from the multivariable logistic models and plotted against BMI at 21. Given that potential outliers at the tail ends of the distribution—BMI at 21 < 18.5 kg/m2 and BMI > 35 kg/m2—could influence the shape of the curve, we performed sensitivity analyses excluding persons with these values and plotted the curves using splines with knots placed at 20th, 40th, 60th and 80th percentiles of BMI at age 21.
All analyses were performed using STATA (version 12.1, Stata Corp, College Station, Texas, USA) and R version 3.1.1 (R Core Team 2014). For all analyses, p-values < 0.05 were considered statistically significant.
SCCS participants provided written informed consent, and protocols were approved by the Institutional Review Boards of Vanderbilt University Medical Center and Meharry Medical College.