Demographic and socioeconomic covariates included age, sex, self-reported race/ethnicity, marital status, education, income in relation to the Federal poverty line, prior Social Security Disability Insurance status, census region, metropolitan status, and a mortality variable indicating whether participants died over the follow-up period. We predicted spending for each person-year using a generalized linear model with a gamma distribution and log link and used robust variance estimators to correct standard errors for repeated observations within individuals. Regression models predicted spending based on BMI categories, a continuous time term, and their interaction, controlling for demographic and economic covariates. Additional models controlled for 10 chronic conditions commonly linked to obesity (diabetes, hypertension, ischemic heart disease, hyperlipidemia, heart failure, chronic lung disease, osteoarthritis, hypothyroidism, gastroesophageal reflux disease, and sleep apnea) to explore whether changes in spending were accounted for by changes in comorbidity over time. All analyses were conducted using Stata statistical software (version 10; StataCorp).