Health-maximizing and welfare-maximizing behaviors can be at odds, especially among disadvantaged groups, generating health disparities. We estimate a lifecycle model of medication and labor supply decisions using data on HIV-positive men. We evaluate an effective HIV treatment innovation that had harsh side effects: HAART. Measured in lifetime utility gains, HAART disproportionately benefitted higher-education men. Lower-education men were more likely to avoid the side effects of HAART that interfered with work. A counterfactual mandate to use HAART improves health but increases inequality because low-education individuals work less. A counterfactual non-labor income subsidy increases HAART adoption and improves health among lower-education individuals.