As health care consumes a growing share of GDP, the demand for better evidence regarding the effects of health care treatments and how these vary across individuals is increasing. Estimating this with observational data is difficult given the endogeneity of treatment decisions. But because the random assignment clinical trials (RACTs) used in the FDA approval process only estimate average health effects and do not consider spending, there is no good alternative. In this study we use administrative data from California’s Medicaid program to estimate the impact of HIV antiretroviral treatments (ARVs). We use data on health care utilization to proxy for health status and exploit the rapid takeup of ARVs following their FDA approval. Our estimate of a 68 percent average mortality rate reduction is in line with the results from RACTs. We also find that the ARVs lowered short-term health care spending by reducing expenditures on other categories of medical care. Combining these two effects we estimate the cost per life year saved at $19,000. Our results suggest an alternative method for estimating the real-world effects of new treatments that is especially well-suited to those treatments that diffuse rapidly following their approval.