Utility of an algorithm of surrogate markers for Cd4 count to determine eligibility for haart among HIV infected pregnant women.
Abstract
Background: CD4 count is an important marker of disease progression in patients
with HIV. But CD4 count testing is not always readily available in developing
countries like Kenya. Studies have shown significant correlation between CD4 count
and Total Lymphocyte Count (TLC) including two studies done in Kenya among
children and non-pregnant adults. Various TLC cut-offs including WHO TLC cut-off of
1200 have had low predictive value for identifying subjects with low CD4 count.
Both WHO and Kenya PMTCT programme recommends HAART for CD4 count
<350celllmm3.There was need therefore to revise the TLC cut-off for CD4 Count
<350cell/mm3 and develop clinical algorithms using biomarkers like haemoglobin
level and BMI to raise the predictive value of TLC.
Methods and Data analysis: This was a retrospective analysis of cross-sectional
data from HIV infected pregnant, ARV naive women. Data was extracted from
patients' charts and entered into a data proforma. The relationship between CD4 and
TLC, BMI, HB and WHO Clinical Stage (WCS) was calculated using Pearson's
Correlation and linear regression. Two by two tables were constructed to determine
performance of various TLC thresholds using the Sensitivity, Specificity, PPV and
NPV. These were used to compute Receiver Operating Curves (ROC) to determine
the predictive accuracy for each of the biomarkers singly and in various
combinations. Data was be analysed by SPSS version 15 and Stata 10.
Results: Of 362 HIV positive pregnant women, 160(44.5%) had CD4 count <350
cells/rnrn". Using linear regression optimal cut-off points for TLC. HB, BMI were
850cell/mm3
. 8Ag/dl and 15.5kg/m2 respectively. These cut-off points were highly
specific but with very low sensitivity. The best cut-off point using generated sensitivity
and specificity values was TLC~2200 with Sensitivity of 68% and Specificity of 51%
A 3-step algorithm of WCS 11&111, TLC~1000 and HB~12g/dl; in that order was the
most optimal with a Sensitivity, Specificity, PPV, NPV . Youden's indexiJ) and ROC
A,UC of 86.21 %,9200%, 94.3%. 74.20%,7200% and 89% respectively
Conclusion: TLC, HB, WCS and BMI have low predictive accuracy for CD4 co nt
c 350cell/mm3 when used alone Our data suggests that HB, BM!. and WCS
increased the Sensitivity of TLC at all thresholds These markers combined in an
algorithm are useful surrogate markers for CD4 Count and can be usee in .esource
poor settings to determine eligibility for H.t,ART