Causal Inference with Data Subject to Covariate-Dependent Observation Times: An Application to a Cohort of New Users of Antidepressants