can cohort studies show causation