In response to growing evidence that social determinants of health (SDH) impact outcomes, the Centers for Medicare and Medicaid Services evaluated the impact of dual eligibility (DE) and disability status (DIS) on Medicare Advantage (MA) performance measures and found little effect. This study evaluates whether adjusting for DE and DIS is sufficient to account for the full impact of SDH and compares additional SDH variables measured at 2 levels of granularity. The study sample included 1.7 million MA beneficiaries from a nationally representative sample of MA encounter data. Beneficiaries were matched to SDH variables from: 1) a commercial research database of the U.S. population aggregated at the 9-digit ZIP Code (ZIP9) level (>30M areas), and 2) the American Community Survey aggregated at the Census Block Group (CBG) level (approximately 220K areas). DE, DIS, and five additional SDH variables (race, education, marital status, home ownership, and income) were used as predictors in three generalized linear regression models. The response variables were two key performance measures, Medication Adherence for Diabetes Medications (MA-D) and Breast Cancer Screening (BCS). The models tested were: 1) only DE and DIS; 2) DE, DIS, +5 SDH at CBG level; 3) DE, DIS, +5 SDH at ZIP9 level. Models were compared using -2 residual log likelihood. For both outcomes, Model 2 (including SDH predictors at CBG level) had significantly better fit than Model 1 (DE and DIS, p <0.0001). Model 3 (SDH predictors at ZIP9 level) further improved model fit, by 42% for BCS and 25% for MA-D. Although DE and DIS are predictive and should be considered for inclusion in quality measure risk adjustment, significantly greater prediction can be obtained by including additional SDH. Predictors aggregated at a “near neighborhood” ZIP9 level provide even better differentiation of disparities compared to SDH measured at CBG level.