Search tips
Search criteria 


Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
J Infect Dis. Author manuscript; available in PMC 2010 May 1.
Published in final edited form as:
PMCID: PMC2758232

Comparison of CD4 Cell Count, Viral Load, and Other Markers for the Prediction of Mortality among HIV-1–Infected Kenyan Pregnant Women



There are limited data regarding the relative merits of biomarkers as predictors of mortality or time to initiation of antiretroviral therapy (ART).


We evaluated the usefulness of the CD4 cell count, CD4 cell percentage (CD4%), human immunodeficiency virus type 1 (HIV-1) load, total lymphocyte count (TLC), body mass index (BMI), and hemoglobin measured at 32 weeks’ gestation as predictors of mortality in a cohort of HIV-1–infected women in Nairobi, Kenya. Sensitivity, specificity, positive predictive value (PPV), and area under the receiver operating characteristic (ROC) curve (AUC) were determined for each biomarker separately, as well as for the CD4 cell count and the HIV-1 load combined.


Among 489 women with 10,150 person-months of follow-up, mortality rates at 1 and 2 years postpartum were 2.1% (95% confidence interval [CI], 0.7%–3.4%) and 5.5% (95% CI, 3.0%–8.0%), respectively. CD4 cell count and CD4% had the highest AUC value (>0.9). BMI, TLC, and hemoglobin were each associated with but poorly predictive of mortality (PPV, <7%). The HIV-1 load did not predict mortality beyond the CD4 cell count.


The CD4 cell count and CD4% measured during pregnancy were both useful predictors of mortality among pregnant women. TLC, BMI, and hemoglobin had a limited predictive value, and the HIV-1 load did not predict mortality any better than did the CD4 cell count alone.

In 2007, >85% of the world’s estimated 2 million HIV-infected pregnant women lived in sub-Saharan Africa [1]. Many women in this region have HIV-1 infection diagnosed during pregnancy as part of programs designed to prevent mother-to-child transmission of HIV-1 (PMTCT). Within these programs for the prevention of mother-to-child transmission of HIV-1, women receive a short course of antiretroviral therapy (ART) to prevent HIV-1 infection in infants, and those who meet World Health Organization (WHO) eligibility guidelines for ART initiate ART. Because the CD4 cell count decreases during pregnancy as a result of hemodilution, it is not clear whether it is appropriate to extrapolate to pregnant women the current CD4 cell count–associated guidelines for initiating ART in resource-poor settings [2, 3]. Several studies have suggested that the CD4 cell percentage (CD4%) may be a better marker for initiation of ART during pregnancy to benefit health of the mother [36].

Less expensive biological markers may have prognostic potential in sub-Saharan Africa. For example, hemoglobin has been shown to be associated with mortality among nonpregnant adults [7] and pregnant women [8], and body mass index (BMI) has been shown to be associated with mortality among men but not among women [7]. However, associations of markers with mortality do not mean that these markers necessarily have a high predictive value for mortality. Previous studies have neither assessed predictive value nor compared the relative predictive value of these markers with that of the CD4 cell count or HIV-1 load. The total lymphocyte count (TLC) has been considered as a surrogate for the CD4 cell count [9], but it has low sensitivity for predicting a CD4 cell count of <200 cells/µL. A few studies have also examined the combination of less expensive biomarkers, such as BMI, hemoglobin, and TLC, as surrogates for a CD4 cell count of <200 cells/µL in cohorts in the United States and Africa [10, 11]. A limitation of these studies is that another surrogate marker (a CD4 cell count of <200 cells/µL) was used as the outcome, rather than such clinical outcomes as AIDS or death. To date, studies have not compared these biomarkers (CD4, HIV-1 load, BMI, hemoglobin, and TLC) in terms of their relative predictive value for death or ART-free survival (which includes both mortality and initiation of ART) either in general or in a specific group, such as HIV-1–infected pregnant women.

In the present study, we determined and compared conventional markers of HIV-1 disease progression, such as the HIV-1 load, CD4 cell count, and CD4%, as well as such alternative potential markers as BMI, hemoglobin, and TLC (measured at 32 weeks’ gestation), for predicting death occurring within 1 and 2 years after delivery.


Study setting and population

Data were collected as part of a prospective cohort study conducted in Nairobi, Kenya, from July 1999 through May 2005 [12, 13]. In brief, pregnant HIV-1–seropositive women were referred from Nairobi City Council clinics to the study clinic at Kenyatta National Hospital (Kenyatta, Nairobi). The first 216 women, who were enrolled between 1999 and 2002, had a follow-up of 12 months; the subsequent 319 women, who were enrolled between 2002 and 2005, had a longer predefined follow-up of 24 months. In addition to mortality, the same immunologic and viral markers and morbidities were evaluated in both groups. Data from both cohorts were used. All subjects provided written, informed consent. Human experimentation guidelines of the US Department of Health and Human Services were followed, and the institutional review board of the University of Washington and the ethics review committee of the Kenya Medical Research Institute and Kenyatta National Hospital approved the study.

Clinical procedures

Baseline medical information was collected using a standardized questionnaire at the time of enrollment. At 32 weeks’ gestation, women were examined, and blood samples were collected for T cell subset analysis and determination of hemoglobin and HIV-1 RNA levels. Women received standard antenatal care, including short-course zidovudine [14] for the prevention of perinatal HIV-1 transmission, and they self-selected the infant feeding modality after receiving counseling. Women received iron and multivitamin supplementation in the first 6 postnatal months. Women with immunosuppression (CD4 cell count, <200 cells/µL) received cotrimoxazole prophylaxis and were referred to HIV treatment programs. After 2003, when ART became highly subsidized or free, women who had been referred accessed ART more readily. Dates of ART initiation were self-reported by participants. All women were followed monthly for 12 months after delivery. Women enrolled after 2001 were also followed trimonthly for an additional 12 months. On the basis of clinical information and interviews with relatives or neighbors, deaths were classified as non–HIV related, HIV related, or of unknown cause.

Laboratory procedures

TLC and CD4% measurements were conducted at the University of Nairobi, by use of a FACScan flow cytometer (Becton Dickinson) [15], with semiannual proficiency testing also performed. Plasma HIV-1 RNA levels were quantified in Seattle by use of a transcription-mediated amplification assay (Gen-Probe), which has been shown to quantify prevalent HIV-1 subtypes in Kenya [16].

Identification of predictor cutoffs

Various cutoffs were assessed and compared for their sensitivity, specificity, and predictive value for death during follow-up. CD4 cell count cutoffs of <200, <250, and <350 cells/µL, which were based on current WHO and US Department of Health and Human Services guidelines [17, 18] for initiation of HAART, were examined. The WHO cutoff for the TLC is 1200 cells/µL for adults (WHO stage 2+). The Centers for Disease Control and Prevention (CDC) [19] recommends a CD4% cutoff of 14%. HIV-1 load cutoffs were selected based on the US Department of Health and Human Services 2002 [20] and 2004 [21] recommendations to initiate ART at 55,000 and 100,000 copies/mL, respectively. We established a hemoglobin cutoff of 11 g/dL (i.e., the CDC definition of anemia at 32 weeks’ gestation) [22]. In addition, we examined the cutoff of 8.5 g/dL established by O’Brien et al. [8]. BMI cutoffs are based on nonpregnant populations and could not be applied to the present cohort. Therefore, we used a cohort-specific cutoff corresponding to the 25th percentile of the BMI distribution measured at 32 weeks’ gestation.

Statistical methods

Descriptive statistics, including counts and frequencies for categorical variables and median values and interquartile ranges for continuous variables, were computed for baseline measurements. Correlations between continuous variables were assessed using Spearman’s rank correlation coefficient. Associations between categorical variables were assessed using Fisher’s exact test.

Time to death was censored at the time of a second pregnancy, ART initiation, or last contact with clinic staff. Because ART initiation may be informative of a high risk for death, we re-ran time-to-event analyses, using ART-free survival (composite ART initiation/death) as the end point. Kaplan-Meier curves were calculated to estimate the cumulative risk. Cox proportional hazards models estimated associations between markers at baseline and the time to death and ART-free survival, allowing for different associations within the first and second years after delivery.

The sensitivities and specificities of the markers, as well as their associated receiver operating characteristic (ROC) curves [23] (including area under the ROC curves [AUCs]), were calculated to assess the usefulness of the markers as predictors of death within 1 and 2 years after delivery. The positive predictive values (PPVs) were calculated by stratifying the data on the basis of a cutoff level and obtaining the corresponding Kaplan-Meier estimate and 95% confidence interval (CI) at 1 and 2 years postpartum. Several biomarker cutoffs were examined to determine how traditional and novel cutoffs compared in terms of sensitivity and specificity. CIs for sensitivities and specificities were based on quantiles of 10,000 samples from the bootstrap procedure. To determine whether linear combinations of markers might provide better prediction than biomarkers used independently, we used as the marker of interest for calculating ROC curves the linear predictors from Cox proportional hazards models with multiple biomarkers as covariates.


Of 535 women enrolled in the study, 501 women (93.6%) were followed to delivery, 489 (97.6%) of whom had follow-up information beyond delivery available and were included in this analysis (180 women in cohort 1 and 309 women in cohort 2). Table 1 provides distributional summaries of selected sociodemographic and clinical indicators and the HIV disease progression markers of interest measured at 32 weeks’ gestation. There were 10,150 person-months of follow-up (2424 and 7726 months of follow-up in cohorts 1 and 2, respectively). Ninety-four (19.6%) of the 480 women who were not known to die during the first year were censored because of loss to follow-up. Fifty (17.1%) of the 293 women who were in the cohort with 2 years of follow-up and were not known to die in the first 2 years were censored. Additional details about these cohorts, including specific causes of loss to follow-up, are reported elsewhere [13].

Table 1
Summary of the distribution of baseline markers and demographic characteristics.

Table 1 lists the number and percentage of women who met specific cutoffs for treatment. CD4 cell count and CD4% were related as follows: of the 51 (10.6%), 87 (18.1%), and 169 (35.1%) women who had CD4 cell counts of <200, <250 and <350 cells/µl, respectively, 43 (84.3%), 55 (63.2%), and 62 (36.7%) also had a CD4% measurement of <14%. A total of 410 women had CD4 cell counts and CD4% values available at both 32 weeks’ gestation and 1 month postpartum. The correlation between the measurements at the 2 time points was 0.67 and 0.84 for the CD4 cell count and the CD4%, respectively. There was a significant increase in the CD4 cell count between 32 weeks’ gestation and 1 month postpartum (87 cells/µL; 95% CI, 65–109 cells/µL) (P < .001), whereas the CD4% did not change appreciably during this period (0.15%; 95% CI, −0.34% to 0.64%) (P = .54). The CD4 cell count and the TLC were correlated (r = 0.63; P < .001), and women with a CD4 cell count of <200 cells/µL were more likely to have a TLC of <1200 cells/µL than were women with a CD4 cell count of ≥200 cells/µL (25% and 4.7%; P < .001).

A total of 20 deaths occurred during the 2-year postpartum period, one of which was known to be non–HIV related and, therefore, was censored at the time of death. The remaining deaths were either known or suspected to be HIV related. Figure 1 displays Kaplan-Meier curves for death and ART-free survival. The estimated cumulative mortality rate and combined cumulative mortality and ART initiation rates at 1 year after delivery were 2.1% (95% CI, 0.7%–3.4%) and 2.8% (95% CI, 1.2%–4.3%), respectively. At 2 years postpartum, these rates were 5.5% (95% CI, 3.0%–8.0%) and 10.6% (95% CI, 7.1%–14.0%). Figure 2 and Figure 3 show Kaplan-Meier mortality curves stratified according to the biomarker cutoffs. Corresponding cumulative rates are shown in table 2. Only one death occurred before 4 months. After 4 months, the event rates for groups stratified according to the CD4 cell count and the CD4% diverged. HIV-1 load, hemoglobin, and BMI curves did not diverge until after 6 months. The curves marking a hemoglobin of <8.5 and >11 g/dL were indistinguishable in the first year. Results for all of the aforementioned biomarkers were similar for ART-free survival.

Figure 1
Kaplan-Meier estimates of the cumulative proportion of women alive (black) and the cumulative proportion of women alive and antiretroviral therapy (ART) free (gray). The vertical dashes denote the times at which the women were censored.
Figure 2
Kaplan-Meier estimates of the cumulative mortality rate, as stratified by marker values. The vertical dashes denote the time points at which ≥1 woman’s event time was censored. BMI, body mass index; CD4, CD4 cell count; CD4%, CD4 cell ...
Figure 3
Kaplan-Meier estimates of the cumulative combined mortality rate and rate of initiation of antiretroviral therapy (ART), as stratified by marker values. The vertical dashes denote time points at which ≥1 woman’s event time was censored. ...
Table 2
Sensitivities, specificities, and predictive positive values (PPVs) for death within 1 or 2 years of delivery, with 95% confidence intervals (CIs) at various cutoffs of biomarkers measured at 32 weeks’ gestation.

All the markers, except hemoglobin, were significantly associated with death and/or ART-free survival within the first year (table 3). Women with CD4 cell counts <200 cells/µL had a risk of death that was 19.5 (95% CI, 4.9–78.2) times that of women with CD4 cell counts >200 cells/µL. When adjustment was made for ART initiation, the estimated risk increased to 29.3 (95% CI, 7.9–108.6). An HIV-1 load > 100,000 copies/mL was associated with a 6.7-fold (95% CI, 1.4-fold to 32.2-fold) increase in the risk of death. This estimate decreased to 3.8 (95% CI, 1.2–12.7) for ART-free survival. TLC was significantly associated with ART-free survival but not with time to death. Women whose TLC was <1200 cells/µL had a 2.8-fold (95% CI, 1.3-fold to 6.2-fold) increased risk of death or starting ARTs. The hazard ratios (HRs) noted in the second year were similar to those noted in the first year.

Table 3
Hazard ratio (HR) estimates from the Cox proportional hazards models with time-varying covariates and outcomes of death and antiretroviral therapy (ART)–free survival (death and ART initiation).

The CD4 cell count and the CD4% had similar ROC curves (figure 4), with high AUC values for predicting death within 1 year (0.94 and 0.92, respectively) and 2 years (0.91 and 0.93, respectively) after delivery. The AUC values for the TLC were close to 0.5, indicating that TLC predicts death approximately as well as tossing a coin. BMI and hemoglobin performed similarly, with AUC values between 0.61 and 0.69.

Figure 4
Receiver operating characteristic curves for several potential predictors of death within the first (left) and second (right) year postpartum, with the corresponding area under the ROC curve (AUC) values. BMI, body mass index; CD4, CD4 cell count; CD4%, ...

Table 2 presents the sensitivities, specificities, and PPVs of the biomarkers at the specific cutoffs. Although hemoglobin is listed, each of the cutoffs perfectly divided the observed events (they all occurred in women with a hemoglobin of 8.5–11 g/dL); therefore, the CIs for sensitivity could not be calculated. A CD4 cell count of <200 cells/µL had the highest cumulative mortality rate at 1 year, with an estimated 14.6% of women expected to die within 1 year. On the basis of the sensitivity of a CD4 cell count of <200 cells/µL, we would expect to identify 69.2% of the women who would die within the first year. In addition, specificity was high (90.7%), and, therefore, treatment would only have been initiated in 9.3% of the women who would not die in the next year. All the events in the first year occurred in women with a CD4 cell count of <350 cells/µL, resulting in a sensitivity of 100% without an available estimate of standard error and a specificity of 66.3% (95% CI, 62.0%–70.3%). A CD4% of <14% had the next highest PPV after a CD4 cell count of <200 cells/µL, which was equal to 13.0% (95% CI, 3.5%–21.6%). Use of this cutoff would have the potential to prevent 79.3% (95% CI, 50.4%–100.0%) of deaths in the first year, while treating only 12.0% (95% CI, 9.3%–15.0%) of women who would not die in the first year.

A Cox proportional hazards model that included CD4 cell count and HIV-1 load as predictors estimated HRs of 3.78 (95% CI, 1.82–7.86; P < .001) for a 100-unit decrease in the CD4 cell count and 1.07 (95% CI, 0.33–3.54) for a 10-fold increase in the HIV-1 load. A similar model with CD4 cell count and hemoglobin estimated HRs of 3.78 (95% CI, 1.72–6.84; P < .001) and 1.02 (95% CI, 0.64–1.62) for a 100-unit decrease in the CD4 cell count and a 1-unit increase in hemoglobin, respectively. The ROC curve based on the linear predictor from these models (with additional information from either the virus load or hemoglobin) resulted in ROC curves almost identical to those noted for the CD4 cell count and CD4% alone (not shown).

Setting a rule that initiates ART at a CD4 cell count of <200 cells/µL or an HIV-1 load >55,000 copies/mL improved the sensitivity to 80.1, compared with a CD4 cell count of >200 cells/µL alone, but it decreased the specificity to 50.6. Increasing the viral load threshold to 100,000 copies/mL increased both the sensitivity and specificity to 83.0 and 63.0, respectively. Adding the virus load to a rule with a CD4 cell count of <200 cells/µL may have some benefit; however, adding the virus load to a CD4 cell count of < 350 cells/µL resulted in no improvement in sensitivity or specificity.


HIV care programs in resource-limited settings often have limited diagnostic options for clinical decision-making. Although CD4 cell counts and viral loads are highly predictive of HIV-1 disease progression, these markers are not available in all settings. Thus, guidelines have arisen to propose initiation of therapy based on alternative clinical surrogate markers, including WHO staging, BMI, and TLC. To date, there are limited comparative data to define how well these markers predict mortality and which are optimal markers. Current programs use available diagnostic markers and comply with WHO guidelines to initiate therapy. Thus, they are not able to ascertain whether the markers were ideal to guide therapy. There are scant data regarding the relative predictive value of these biomarkers for death among HIV-1–infected individuals in general or in postpartum women specifically. This study provides important evidence regarding comparative merits of these markers in general, and, in particular, it applies to women who had HIV-1 diagnosed during pregnancy, who comprise a significant proportion of HIV-1–infected women overall.

In the present study, the 2.1% estimated risk of mortality at 1 year underscored the typically early disease status of HIV-1–infected women during pregnancy. We found that the CD4 cell count and the CD4% were most predictive of death and that the HIV-1 load did not provide predictive value beyond that provided by the CD4 cell count and was a less predictive marker than were the CD4 cell count and CD4% measurements. In addition, BMI, TLC, and hemoglobin were inferior predictors of mortality. Empirical decisions to have women initiate highly active ART (HAART) solely on the basis of these criteria would lead both to substantial numbers of women who should have received treatment not receiving it and to many women receiving treatment prematurely.

The CD4 cell count and CD4% measured at 32 weeks’ gestation were highly sensitive and specific predictors of maternal mortality within 1 year after delivery. Both biomarkers had a greater predictive value than did other markers, including HIV-1 load, hemoglobin, and TLC. We confirmed previous findings [2, 4] that the CD4% values measured at 32 weeks’ gestation and 1 month postpartum were more stable than were the CD4 cell counts measured at the same time points; however, this stability did not translate into a predictive benefit. Thus, the prenatal CD4% was not more useful that the CD4 cell count in predicting outcomes at 1 year, suggesting that either measurement may be useful in pregnant women.

The CDC defines AIDS on the basis of a CD4 cell count of <200 cells/µL, a CD4% of <14%, or the presence of an opportunistic infection. We found that using a CD4% of <14% had greater sensitivity for predicting death within 1 or 2 years than did a CD4 cell count of < 200 cells/µL; however, as determined from the ROC curves, a CD4 cell count of 210 cells/µL would have produced the same sensitivity and specificity as would a CD4% of <14%.

Mellors et al. [24] first recommended the combined use of the CD4 cell count and the HIV-1 RNA concentration in untreated HIV-1–infected adults for the prediction of subsequent disease progression. The relative merit of these 2 biomarkers has since been a topic of debate [2428]. We found that the HIV-1 load measured at 32 weeks’ gestation was associated with mortality in the first 2 years; however, we did not find that the HIV-1 load measured at 32 weeks’ gestation predicted mortality better than did the CD4 cell count or CD4%. In fact, the area under the ROC curve was much less for the HIV-1 load than for the CD4 cell count or CD4%. In addition, in the region of the ROC curve where we would like to see high sensitivity (specificity, >80%), the sensitivity remained <50%. Thus, the present study suggests that HIV-1 load assays are not incrementally beneficial for decisions regarding treatment initiation, particularly when better predictions can be obtained using less expensive measures.

Although previous studies have evaluated the role of hemoglobin in predicting a CD4 cell count <200 cells/µL, the predictive role of hemoglobin with regard to mortality has not been defined. In the present study, hemoglobin was associated with the risk of death in the second year after delivery, but not in the first. In the second year, the HR comparing women with a hemoglobin of <8.5 g/dL with women with a hemoglobin ≥8.5 g/dL was 4.65, which is consistent with the findings of O’Brien et al. [8]; however, this did not translate into a biomarker with adequate mortality prediction properties. Thus, including hemoglobin with the CD4 cell count did not improve prediction in our study. BMI has similarly been evaluated as a surrogate for the CD4 cell count [10, 29]. In the present study, BMI was associated with death and ART-free survival in the first year after delivery but not in the second year; however, it showed poor properties as a predictor of mortality, as shown by the ROC and Kaplan-Meier curves.

Because some women started receiving ART during follow-up, we also conducted analyses of ART-free survival as a sensitivity analysis for this potential source of informative censoring. We found results similar to those of analyses with mortality alone for all of the biomarkers.

The main strength of the present study was that each biomarker was assessed for its ability to predict a hard end point (death) instead of the ability of the biomarkers to predict other surrogate markers (i.e., how well TLC predicts the CD4 cell count or how the antenatal CD4 cell count correlates with the postnatal CD4 cell count). Our study illustrates the importance of distinguishing significant associations from predictive performance. As illustrated by Pepe et al. [30], not only is a significant association required for a biomarker to be a good screening or prognostic tool, but it must also have an association stronger than that usually encountered in epidemiologic studies. Thus, HRs significantly different than 1 do not necessitate that a marker has a strong predictive performance. In the present study, several biomarkers were significantly associated with mortality but failed to be strong predictive markers. A limitation of the present study is that WHO clinical stage information was not collected. In addition, the low event rate (i.e., death outcomes) in the study may have reduced the precision of sensitivity, specificity, and PPV estimates. However, the relatively low mortality rate among women in this study likely reflects early HIV disease status in this population. Because many women in resource-limited settings first have HIV-1 infection diagnosed during pregnancy, the value of such markers as CD4 cell count and CD4%, BMI, TLC, and hemoglobin during pregnancy to predict the subsequent course is directly relevant to women and caregivers. As such, the present study provides important comparisons of predictive value of these markers.

Currently, the WHO recommends using TLC and the WHO disease stage to inform decisions about initiating ART when the CD4 cell count is unavailable. TLC has been evaluated as a surrogate for the CD4 cell count in a number of settings with mixed results. In the present study, the TLC determined at 32 weeks’ gestation was not an adequate predictor of maternal mortality in the first 2 years postpartum, whereas the CD4 cell count and the CD4% measured at this time point were. The current study indicates that, although the CD4 cell count may fluctuate more than the CD4% during pregnancy, both are excellent predictors of maternal mortality within 2 years after delivery.

Mechanistic reasons may explain why the CD4 cell count is the best predictor. The CD4 cell count is more causally proximate to the outcome and, thus, may better predict the outcome than measurements like BMI and hemoglobin, which are more indirectly related, or virus load, which measures the amount of virus in the host rather than the immunologic influence of the virus. Alternatively, because the viral load and CD4 cell count are highly correlated, their predictive abilities may be similar, and it may be random chance that, in the present study, the CD4 cell count was the better predictor. Overall, this study suggests that programs providing care to HIV-1–infected women should aim to implement facilities to measure the CD4 cell count rather than rely upon other surrogate markers. Reassuringly, measurement of virus levels did not improve the predictive value, suggesting that a lack of viral assays does not compromise management decisions in terms of preventing maternal mortality due to HIV-1.


Financial support: US National Institutes of Health (NIH; research grants R01 HD 23412 and AI27757) and University of Washington Center for AIDS Research P.O., E.M.O., and C.F. were scholars in the University of Washington AIDS International Research and Training Program supported by the NIH Fogarty International Center (grant D43-TW00007).


Potential conflicts of interest: none reported.


1. World Health Organization/Joint United Nations Programme on HIV/AIDS/United Nations Children’s Fund. Towards universal access: scaling up priority HIV/AIDS interventions in the health sector: progress report. 2007. Apr [Accessed 10 October 2007]. Available at:
2. Ekouevi DK, Inwoley A, Tonwe-Gold B, et al. Variation of CD4 count and percentage during pregnancy and after delivery: implications for HAART initiation in resource-limited settings. AIDS Res Hum Retroviruses. 2007;23:1469–1473. [PubMed]
3. Mulcahy F, Wallace E, Woods S, et al. Counts in pregnancy do not accurately reflect the need for long-term HAART [abstract 704b]. Program and abstracts of the 13th Conference on Retroviruses and Opportunistic Infections (Denver); 2006. Available at:
4. Miotti PG, Liomba G, Dallabetta GA, Hoover DR, Chiphangwi JD, Saah AJ. T lymphocyte subsets during and after pregnancy: analysis in human immunodeficiency virus type 1–infected and –uninfected Malawian mothers. J Infect Dis. 1992;165:1116–1119. [PubMed]
5. Temmerman M, Nagelkerke N, Bwayo J, Chomba EN, Ndinya-Achola J, Piot P. HIV-1 and immunological changes during pregnancy: a comparison between HIV-1-seropositive and HIV-1-seronegative women in Nairobi, Kenya. AIDS. 1995;9:1057–1060. [PubMed]
6. Van der Paal L, Shafer LA, Mayanja BN, Whitworth JA, Grosskurth H. Effect of pregnancy on HIV disease progression and survival among women in rural Uganda. Trop Med Int Health. 2007;12:920–928. [PubMed]
7. Erikstrup C, Kallestrup P, Zinyama R, et al. Predictors of mortality in a cohort of HIV-1-infected adults in rural Africa. J Acquir Immune Defic Syndr. 2007;44:478–483. [PubMed]
8. O’Brien ME, Kupka R, Msamanga GI, Saathoff E, Hunter DJ, Fawzi WW. Anemia is an independent predictor of mortality and immunologic progression of disease among women with HIV in Tanzania. J Acquir Immune Defic Syndr. 2005;40:219–225. [PubMed]
9. Gupta A, Gupte N, Bhosale R, et al. Low sensitivity of total lymphocyte count as a surrogate marker to identify antepartum and postpartum Indian women who require antiretroviral therapy. J Acquir Immune Defic Syndr. 2007;46:338–342. [PubMed]
10. Chen RY, Westfall AO, Hardin JM, et al. Complete blood cell count as a surrogate CD4 cell marker for HIV monitoring in resource-limited settings. J Acquir Immune Defic Syndr. 2007;44:525–530. [PubMed]
11. Spacek LA, Griswold M, Quinn TC, Moore RD. Total lymphocyte count and hemoglobin combined in an algorithm to initiate the use of highly active antiretroviral therapy in resource-limited settings. AIDS. 2003;17:1311–1317. [PubMed]
12. Otieno P, Brown E, Mbori-Ngacha D, et al. HIV-1 disease progression in breast-feeding and formula-feeding mothers: a prospective 2-year comparison of T cell subsets, HIV-1 RNA levels, and mortality. J Infect Dis. 2007;195:220–229. [PMC free article] [PubMed]
13. Walson JL, Brown ER, Otieno PA, et al. Morbidity among HIV-1-infected mothers in Kenya: prevalence and correlates of illness during 2-year postpartum follow-up. J Acquir Immune Defic Syndr. 2007;46:208–215. [PMC free article] [PubMed]
14. Shaffer N, Chuachoowong R, Mock PA, et al. Short-course zidovudine for perinatal HIV-1 transmission in Bangkok, Thailand: a randomised controlled trial. Bangkok Collaborative Perinatal HIV Transmission Study Group. Lancet. 1999;353:773–780. [PubMed]
15. Kiarie JN, Richardson BA, Mbori-Ngacha D, Nduati RW, John-Stewart GC. Infant feeding practices of women in a perinatal HIV-1 prevention study in Nairobi, Kenya. J Acquir Immune Defic Syndr. 2004;35:75–81. [PMC free article] [PubMed]
16. Emery S, Bodrug S, Richardson BA, et al. Evaluation of performance of the Gen-Probe human immunodeficiency virus type 1 viral load assay using primary subtype A, C, and D isolates from Kenya. J Clin Microbiol. 2000;38:2688–2695. [PMC free article] [PubMed]
17. World Health Organization. Antiretroviral therapy for HIV infection in adults and adolescents: recommendations for a public health approach, 2006 revision. [Accessed 21 November 2007]. Available at: [PubMed]
18. Department of Health and Human Services Panel on Antiretroviral Guidelines for Adults and Adolescents. Guidelines for the use of antiretroviral agents in HIV-1-infected adults and adolescents. 2008. Nov 3 [Accessed 22 September 2008]. Available at:
19. Centers for Disease Control and Prevention. 1993 revised classification system for HIV infection and expanded surveillance case definition for AIDS among adolescents and adults. MMWR Recomm Rep. 1992;41:1–19. [PubMed]
20. Dybul M, Fauci AS, Bartlett JG, Kaplan JE, Pau AK. Guidelines for using antiretroviral agents among HIV-infected adults and adolescents: the panel on clinical practices for treatment of HIV. Ann Intern Med. 2002;137:381–433. [PubMed]
21. Panel on Clinical Practices for Treatment of HIV Infection. Guidelines for the use of antiretroviral agents in HIV-1-infected adults and adolescents. 2004. Oct 29 [Accessed 14 October 2008]. Available at:
22. Centers for Disease Control and Prevention. Recommendations to prevent and control iron deficiency in the United States. MMWR Recomm Rep. 1998;47:1–29. [PubMed]
23. Heagerty PJ, Lumley T, Pepe MS. Time-dependent ROC curves for censored survival data and a diagnostic marker. Biometrics. 2000;56:337–344. [PubMed]
24. Mellors JW, Munoz A, Giorgi JV, et al. Plasma viral load and CD4+ lymphocytes as prognostic markers of HIV-1 infection. Ann Intern Med. 1997;126:946–954. [PubMed]
25. Rodriguez B, Sethi AK, Cheruvu VK, et al. Predictive value of plasma HIV RNA level on rate of CD4 T-cell decline in untreated HIV infection. JAMA. 2006;296:1498–1506. [PubMed]
26. Mellors JW, Margolick JB, Phair JP, et al. Prognostic value of HIV-1 RNA, CD4 cell count, and CD4 cell count slope for progression to AIDS and death in untreated HIV-1 infection. JAMA. 2007;297:2349–2350. [PubMed]
27. Lepri AC, Katzenstein TL, Ullum H, et al. The relative prognostic value of plasma HIV RNA levels and CD4 lymphocyte counts in advanced HIV infection. AIDS. 1998;12:1639–1643. [PubMed]
28. Anastos K, Kalish LA, Hessol N, et al. The relative value of CD4 cell count and quantitative HIV-1 RNA in predicting survival in HIV-1-infected women: results of the women’s interagency HIV study. AIDS. 1999;13:1717–1726. [PubMed]
29. Zachariah R, Teck R, Ascurra O, Humblet P, Harries AD. Targeting CD4 testing to a clinical subgroup of patients could limit unnecessary CD4 measurements, premature antiretroviral treatment and costs in Thyolo District, Malawi. Trans R Soc Trop Med Hyg. 2006;100:24–31. [PubMed]
30. Pepe MS, Janes H, Longton G, Leisenring W, Newcomb P. Limitations of the odds ratio in gauging the performance of a diagnostic, prognostic, or screening marker. Am J Epidemiol. 2004;159:882–890. [PubMed]