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

Abstract

A systematic review and meta-analysis were conducted to synthesize results from cohort studies on initiation into drug injection among vulnerable populations, to quantify heterogeneity in the estimated incidence rates of drug injection and to identify potential sources of heterogeneity and bias. MEDLINE, EMBASE, PsycINFO and LILACS were searched for relevant studies published between 1980 and 2012. Investigators independently reviewed studies for inclusion, retrieved information on baseline population characteristics and follow-up features and assessed study quality. Study-specific incidence rates of drug injection were calculated as the number of new injectors divided by the person-years at risk. The I(2) statistic was used to quantify heterogeneity in incidence rates across studies, and random-effects meta-regression models were used to identify determinants of heterogeneity and bias. Nine cohorts totalling 1843 participants met the inclusion criteria, with individual sample sizes of 70-415 participants and follow-up lengths of 6 months-3.4 years. The incidence of drug injection varied widely, from 2.1 to 24.2 cases per 100 person-years. The strong between-study heterogeneity (I(2)  = 90%, P<0.001) was reduced significantly after accounting for the different follow-up lengths (I(2)  = 17%, P = 0.30), with a 57% (95% confidence interval 46-66%) decrease in the pooled incidence of drug injection per 1-year increase in average follow-up. The incidence of drug injection decreases sharply with increasing follow-up length in cohort studies on drug injection initiation. Low retention rates and potential for downward selection bias in cohort studies on drug injection initiation are caused primarily by greater loss to follow-up among individuals at higher risk of starting injection, compared with other participants.