This study was based on the large population-based SCCS cohort, which has been described previously (www.southerncommunitystudy.org).¹³  In brief, at entry into the SCCS, trained interviewers administered a comprehensive in-person baseline interview that elicited information, such as demographics, anthropometry, medication use, and medical history. Participants were asked to donate a blood sample at study enrollment. A 20-mL venous blood sample was donated by 59% of those eligible to provide one. The blood was separated into its components and stored at −86°C. The 1996 SCCS subjects made available for inclusion in this study were drawn from a stratified random sample selected by the SCCS for another project of obesity biomarkers and reflect the combination of 2 previously sampled groups. The first group of 395 women (enrolled from 2002 to 2004) had been randomly selected within strata defined by race (black/white), smoking status (never, former, current), and body mass index (BMI, 18-24.99 kg/m², 25-29.99 kg/m², 30-45 kg/m²). The second group (n = 1605, enrolled from 2002 to 2006) represents a similarly stratified random sample by race (black/white), BMI (18.5-24.99 kg/m², 25.0-29.99 kg/m², 30.0-34.99 kg/m², 35.0-44.99 kg/m²), and menopausal status (premenopausal/postmenopausal). Four women were removed because they had a recorded preceding lymphoproliferative malignancy,¹⁴  resulting in our final study population of 1996 subjects.

All study subjects had information on self-reported weight and height. As stated in “Results and discussion,” for approximately 25% of the cohort, medical record-abstracted weight and height (measured on the day of study enrollment at the community health center) were also available. When we compared BMI values calculated from self-reported height and weight compared with BMI calculated from the medical record data, we found a very high concordance (Pearson correlation > 0.95). This provides the assurance about the reliability of the self-report.