Cox proportional hazards models of factors associated with loss to follow-up before and after correction for mortality.

ART, antiretroviral treatment; WHO, World Health Organization; HR, hazard ratio; CI, confidence interval; PMTCT, prevention of mother to child transmission; TB, tuberculosis; AZT, zidovudine; 3TC, lamivudine; NVP, nevirapine; D4T, stavudine; EFV, efavirenz. P-values for CD4 count, WHO stage, year of initiation on ART and antiretroviral regimen are from Wald test for linear hypothesis after estimation. After correction for mortality, loss to follow-up was not associated with male gender or weight anymore, while new associations emerged with higher baseline CD4 counts and pregnancy at initiation of ART.