Paper # 556|
A Meta-analytic Approach to Estimating a Smooth Parametric Relationship between Initiating CD4 Count and Mortality Using a Novel Method of Adjusting for Loss to Follow Up
Matthew Fox*1, O McCarthy2, and M Over2
1Boston Univ, MA, US and 2Ctr for Global Devt, Washington, DC, US
Background: Studies reporting 1-year mortality of
patients starting ART often report mortality by CD4 intervals without adjusting
for loss to follow up (LTFU). This prevents incorporation into models
projecting multi-year, population-level impacts of policies on what CD4 count
to begin ART. We propose a new meta-analysis approach of grouped mortality data
to estimate a smooth parametric relationship to predict 1-year mortality among
all starting ART, including the LTFU, at any CD4 count 10 to 400.
Methods: We searched public databases for
reports of mortality on ART by initiating CD4 count. For each reported CD4
strata we recorded the 1-year mortality and LTFU rate and N. We used multiple
imputation to overcome patient grouping by CD4 intervals and a 2-stage maximum
likelihood (ML) approach to estimate unobserved mortality among those LTFU. We
regressed logged mortality (adjusted for LTFU) on logged initiating CD4 count
weighted by sample size. We varied the value of mortality among those lost from
0 to 100% and identified the ML estimate as our point estimate of 1-year
mortality among the LTFU.
Results: We included 21 reports of data on 58,058
patients from 14 countries and 57 CD4 count observations. One-year mortality
ranged 1.3 to 40% and 1-year LTFU from 0.3 to 24%. Without adjusting for LTFU,
we estimate those initiating ART at a CD4 of 200 have a predicted 1-year
mortality of 5% and that prediction increases sharply with lower CD4 count.
After searching values of unobserved mortality (range 0 to 100%), the best fit
occurs when mortality among those lost is 40% (F stat 20.8). Adjusting for
LTFU, predicted mortality was higher at all CD4 counts, while the slope of
mortality with respect to initiation CD4 is flatter. Incorporating this
adjustment, predicted 1-year mortality among those with a CD4 <200 cells/mm3
and >500 was 7.5% to 12.1% and 5.5%, respectively. Comparing those starting
ART <500 to those at 50 cells/mm3, 1-year ART mortality was
halved (12.13% vs 5.5). Comparison of our curve with survival data from HIV+
not on ART reveals our approach may overestimate 1-year mortality at the
highest CD4 counts.
Conclusions: Our findings confirm patients initiated
at lower CD4 counts are at significantly greater risk of death during the first
year on ART, but suggest the benefit to earlier initiation is not as great as
would appear in data not adjusted for LTFU.