Potential bias due to outcome-related loss to follow-up in cohort studies on incidence of drug injection: systematic review and meta-analysis.