Patterns and predictors of early and late mortality during HIV antiretroviral therapy in a military-based programme in South Africa : a cohort study

Abstract

Background The patterns and predictors of mortality in patients receiving combination ART (cART) in the South African military are unknown. Although predictors of early mortality have been described in southern Africa, over 95% of those patients had initiated 3TC + NNRTIbased regimens. In this secondary analysis of Phidisa II, a completed, factorial RCT comparing lopinavir/r vs efavirenz, and D4T+3TC vs ZDV+DDI, 75% of patients initiated other regimens. Few sub-Saharan studies have reported predictors of late mortality, and fewer have reported causes of mortality in cART-treated patients for periods longer than five years. We investigated the patterns and predictors of early and late mortality after extended follow-up of up to eight years. Methods The cohort initiated cART in the Phidisa II trial, and after 31 March 2008, a median follow-up of 24.7 months, transferred to an observational study. Randomisation was stratified by site. The primary end-points of the trial were all-cause mortality and progression of disease to AIDS. In the observational study, the follow-up schedule and event reporting continued as in the RCT. Randomised regimens were continued for as long as indicated. For the current study, Kaplan-Meier survival estimates and mortality rates were calculated. Follow-up was censored at: 365 days post-randomisation for early mortality; and 31st December 2011 for late and overall mortality. Baseline factors associated with mortality were assessed using Cox regression analyses, stratified by site. For late mortality, updated values of CD4+ count, haemoglobin, and BMI were considered instead of baseline values. The final reduced multivariate models were obtained with backward elimination. Results Between 2004 and 2007, 1771 patients enrolled (mean age 35 years, 68% male, median CD4+ count = 102 cells/mL). The total follow-up was 8921 person-years, with median follow-up of 5.4 person-years. Of the 279 deaths, 151 occurred in the 1st year (death rate 9.1/100 py). The rate after 1 year was 1.8/100 py. Tuberculosis was the most common cause of death, responsible for 12.6% of early and 22.7% of late mortality. Factors predicting early mortality were: BMI<18.5 kg/m2 (adjusted Hazard Ratio (aHR) 2.35, p-value <0.001), lower CD4+ count (p<0.001), WHO clinical stage 3/4 (aHR 1.84, p= 0.003), anaemia (aHR 2.87, <0.001), higher AST (aHR 2.06, p=0.006), and marriage (aHR 0.63, p= 0.009). Factors predicting late mortality were: male sex (aHR 2.84, p=0.001), initiating ZDV+DDI (aHR 1.63, p= 0.032), low BMI (aHR 3.55, p<0.001), haemoglobin (aHR 0.75 per g/dL increase, p < 0.001), low CD4+ count (p=0.007) and viral suppression at month 12 (aHR 0.49, p= 0.007). Conclusions and Recommendations Mortality is highest early in the course of cART. Advanced HIV disease, and not being married, predicted higher early mortality. Poor response to cART at one year, male sex, and initiating ZDV+DDI, predicted higher late mortality. The unfavourable effect of ZDV+DDI on mortality has not been described in southern Africa before. Closer follow-up and support of persons with advanced HIV disease, poor response to cART, males, and the unmarried, are recommended. The use of cART regimes recommended by WHO and national guidelines, is encouraged.