Incidence rate ratios (IRRs) and 95% CIs for incident sarcoidosis were estimated for categories of BMI at age 18 years, BMI at baseline, and weight gain from age 18 years to 1995 using Cox proportional hazard models in PROC PHREG of SAS 9.1 (SAS Institute, Inc). Person-time was calculated from baseline to year of sarcoidosis diagnosis, loss to follow-up, death, or end of follow-up, whichever occurred first. We constructed two analytic models to control for potential confounders. Model 1 adjusted for age (1-year intervals) and questionnaire cycle (2-year intervals). Model 2 adjusted for model 1 covariates plus education (≤ 12, 13-15, ≥ 16 years), geographic region (Northeast, South, Midwest, West), pack-years of smoking (never smoked, < 5, 5-14, 15-24, ≥ 25 years), alcohol consumption (never, < 1, 1-6, 7-13, ≥ 14 drinks/week), and hours of vigorous physical activity per week (none, < 5, and ≥ 5 h/week). All variables in the model, with the exception of BMI, were treated as time varying using the Anderson-Gill data structure.15 We did not treat BMI as time varying because of the possibility that symptoms of sarcoidosis would have affected weight during follow-up. In further analyses, we controlled for additional potential confounders, including prudent dietary pattern (quintiles), health insurance status (yes, no), recent medical visit (yes, no), hypertension with medication (yes, no), type 2 diabetes mellitus (yes, no), and asthma (yes, no). Tests of trend were performed by entering each variable in its ordinal form into a single term in the model.16 We assessed whether associations between BMI and sarcoidosis were consistent across age (< 45 years, ≥ 45 years) and smoking status (ever smoker, never smoker). We obtained P values for interaction by comparing models with and without cross-product terms between the covariate and the exposure variable.