Causal Inference in Observational Studies with Outcome-Dependent Sampling