A Novel Approach to Accounting for Loss to Follow-Up when Estimating the Relationship between CD4 Count at ART Initiation and Mortality

BACKGROUND While CD4 strongly predicts mortality on antiretroviral therapy (ART), estimates from programmatic data suffer from incomplete patient outcomes. METHODS We conducted a pooled analysis of one-year mortality data on ART accounting for lost patients. We identified articles reporting one-year mortality by ART initiation CD4 count. We estimated the average mortality among those lost as the value that maximizes the fit of a regression of the natural log of mortality on the natural log of the imputed mean CD4 count in each band. RESULTS We found 20 studies representing 64,426 subjects and 51 CD4 observations. Without correcting for losses, one-year mortality was >4.8% for all CD4 counts <200 cells/mm(3). When searching over different values for mortality among those lost, the best fitting model occurs at 60% mortality. In this model, those with a CD4≤200 had a one-year mortality above 8.7, while those with a CD4>500 had a one-year mortality <6.8%. Comparing those starting ART at 500 vs. 50, one-year mortality risk was reduced by 54% (6.8 vs. 12.5%). Regardless of CD4 count, mortality was substantially higher than when assuming no mortality among those lost, ranging from a 23-94% increase. CONCLUSIONS Our best fitting regression estimates that every 10% increase in CD4 count at initiation is associated with a 2.8% decline in one-year mortality, including those lost. Our study supports the health benefits of higher thresholds for CD4 count initiation and suggests that reports of programmatic ART outcomes can and should adjust results for mortality among those lost.