Correcting for mortality among patients lost to follow up on antiretroviral therapy in South Africa: a cohort analysis

Abstract

BACKGROUND: Loss to follow-up (LTF) challenges the reporting of antiretroviral treatment (ART) programmes, since it encompasses patients alive but lost to programme and deaths misclassified as LTF. We describe LTF before and after correction for mortality in a primary care ART programme with linkages to the national vital registration system. Methods and FINDINGS: We included 6411 patients enrolled on ART between March 2001 and June 2007. Patients LTF with available civil identification numbers were matched with the national vital registration system to ascertain vital status. Corrected mortality and true LTF were determined by weighting these patients to represent all patients LTF. We used Kaplan-Meier estimates and Cox regression to describe LTF, mortality among those LTF, and true LTF. Of 627 patients LTF, 85 (28.8%) had died within 3 months after their last clinic visits. Respective estimates of LTF before and after correction for mortality were 6.9% (95% confidence interval [CI] 6.2-7.6) and 4.3% (95% CI 3.5-5.3) at one year on ART, and 23.9% (95% CI 21.0-27.2) and 19.7% (95% CI 16.1-23.7) at 5 years. After correction for mortality, the hazard of LTF was reversed from decreasing to increasing with time on ART. Younger age, higher baseline CD4 count, pregnancy and increasing calendar year were associated with higher true LTF. Mortality of patients LTF at 1, 12 and 24 months after their last visits was respectively 23.1%, 30.9% and 43.8%; 78.0% of deaths occurred during the first 3 months after last visit and 45.0% in patients on ART for 0 to 3 months. CONCLUSIONS: Mortality of patients LTF was high and occurred early after last clinic visit, especially in patients recently started on ART. Correction for these misclassified deaths revealed that the risk of true LTF increased over time. Research targeting groups at higher risk of LTF (youth, pregnant women and patients with higher CD4 counts) is needed.