Longitudinal Data Discontinuity in Electronic Health Records and Consequences for Medication Effectiveness Studies