Interest of large electronic health care databases in addictovigilance: Lessons from 15 years of pharmacoepidemiological contribution.