The primary outcome variable was receipt of a DNR order within the first 24 hours after hospital admission (early DNR). Independent variables included age (in 5-year intervals), sex, race/ethnicity (non-Latino white, non-Latino black, Latino, Asian/Pacific Islander, and other), type of insurance (Medicare, Medi-Cal, private, other, or indigent), admission source (home, nursing home, or other), and diagnosis-related group (DRG). Patient illness was measured by comorbidity index and illness severity score. Comorbid medical conditions were identified from the Patient Discharge Database, and a Charlson Comorbidity Index was calculated for each hospitalization.16,17 Illness severity scores derived from multivariate logistic regression models of hospital death were categorized by quartile of the estimated probability of hospital death.18 Proxy measure of income was estimated from household income by residence ZIP code. To account for patient referral, we calculated the distance between the patient’s residence and the hospital using ZIP code centroids. For comparability between institutions, percentile distances traveled by patients treated within the same hospital were calculated. Distances were categorized as 75th percentile, above the 75th to the 95th percentile, and above the 95th percentile. For patients without valid ZIP codes (<3%), hot-deck imputation (random sampling of observed characteristics) was used to impute income, rural residence location, and distance traveled.19 Hospital-level predictors included hospital size (number of beds), control/profit status (private nonprofit, district, county, or for-profit), and academic hospital status.