Changes in BMI per year in the AHS-2 population by quartile of baseline BMI (interrupted lines) and over all subjects (solid line). The regression model was change in BMI per year = α + Σ(βj ⋅ agej) + Σ(βk ⋅ covariatek), where j represents the J-1 age range indicator variables, and k the K covariates. Data are the predicted values of change in BMI per year at a particular age, conditional on covariates at mean values (continuous variables) or reference values (categorical variables), with 95% confidence bands. AHS-2, Adventist Health Study 2.