Risk of WHO 4 events and death by current CD4 on ART in the DART trial: the impact of CD4-dependent reporting bias.
Background: Antiretroviral therapy is often managed without routine laboratory monitoring in Africa. The Development of AntiRetroviral Therapy in Africa (DART) trial showed that this had a small statistically significant impact on WHO 4 (AIDS) events; but randomization was open and CD4-dependent reporting bias may have affected findings. Methods: In Uganda and Zimbabwe, 3316 ART-naive adults (median CD4 86 cells/mm<sup>3</sup>) were randomized to routine laboratory and clinical monitoring (LCM) vs clinically driven monitoring (CDM). Participants initiated zidovudine (ZDV)/lamivudine (3TC) plus tenofovir (TDF) (74%), abacavir (ABC) (9%), or nevirapine (NVP) (16%), and were switched to second-line therapy after new/recurrent WHO stage 4 events and (in the LCM group only) CD4 count 3. In the LCM group, 12-weekly haematology, biochemistry, and CD4 results were returned; in the CDM group, results (except CD4) could be requested and grade 4 toxicities were returned. Deaths and WHO 4 events were independently adjudicated. Risk according to latest CD4 count was estimated using Poisson models, including multiple events. Results: Median follow-up was 4.9 years to December 2008. During this period, 382 participants died: 129 (8%) CDM vs 119 (7%) LCM in the first 2 years, and 89 (5%) vs 45 (3%) subsequently. In all, 992 new or recurrent WHO 4 events were reported, 780 meeting protocol criteria. Latest CD4 explained 65% and 66% of the difference between CDM and LCM in time to first new WHO 4 or death (HR = 1.31; 95%CI, 1.14 to 1.51; adjusted aHR = 1.10, 0.96 to 1.27), and death (HR = 1.35, 1.10 to 1.65; aHR = 1.11, 0.90 to 1.36), respectively. There was no difference between CDM and LCM in the risk of death at a given CD4 (P = 0.54), which was 12.6, 2.0, 1.0, 0.9 and 0.6/100 person-years for latest CD4 counts of 0 to 99, 100 to 199, 200 to 349, 350 to 499, and >500 cells/mm<sup>3</sup>, respectively. Most deaths were judged HIV-related (111 CDM vs 83 LCM), with the most frequent primary causes being septicemia/neutropenia (44), neurological events (39), cryptococcosis (32) and tuberculosis (31). In contrast, risks of reported and accepted WHO 4 events for a given CD4 count differed significantly between CDM and LCM (P = 0.002 and P = 0.02, respectively), being greater in the CDM group only at higher CD4 counts where most time was spent (e.g., for reported WHO 4 events, 3.0 vs 1.8/100 person-years with CD4 count 200 to 349 cells/mm<sup>3</sup>; 2.0 vs 1.1/100 person-years with CD4 count 350 to 499 cells/mm<sup>3</sup>; 2.0 vs 0.6/100 person-years with CD4 count 500+ cells/mm<sup>3</sup>). Predicting the expected number of events given observed CD4 counts in the LCM group suggested 40 missed WHO 4 events at CD4 >200. Conclusions: HIV-related deaths after 2 years on ART drove differences between monitoring strategies. Open randomization led to underreporting, and likely under-investigation, of potential WHO 4 endpoints at higher CD4 counts in the LCM group (doctors would assume such events to be unlikely when the CD4 count was high). True differences between CDM and LCM in time to WHO 4 events, but not death, may be smaller than estimated.
Presentation from 17th Conference on Retroviruses and Opportunistic Infections, San Francisco, California, USA, 16-19 February 2010. Abstract K-104; Paper #56.