Fig. 3
![Fig. 3](http://media.springernature.com/full/springer-static/image/art%3A10.1186%2Fs12874-018-0484-z/MediaObjects/12874_2018_484_Fig3_HTML.gif)
The effect of follow-up time on three different effect estimates comparing ART initiation by IDU accounting for the competing event of death and loss-to-clinic. Panel a) displays the standardizeda difference (with 95% confidence interval) in restricted mean time in years spent not on ART while alive and in clinic by history of injection drug use over follow-up time. Panel b) displays the standardizeda cause-specific hazard ratio (HR) over follow-up time. Panel c) displays the standardizeda subdistribution mortality risk ratio (RR) over follow-up time. a Curves are standardized to the distribution of sex, race, AIDS diagnosis, prior ART mono- or dual-therapy, age, CD4 cell count, log10 HIV viral load, Hepatitis C, history of smoking, and hazardous alcohol use in the total sample at clinic enrollment using inverse probability of exposure weights. b Dashed grey lines represent the null hypothesis (i.e. difference in RMST = 0 and HR = 1) in each panel. c The solid black line on panels B and C is the lowess smoother to show the overall trend of the estimands