Skip to main content
Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
Trop Med Int Health. Author manuscript; available in PMC 2014 May 1.
Published in final edited form as:
Trop Med Int Health. 2013 May; 18(5): 619–631.
Published online 2013 Feb 18. doi: 10.1111/tmi.12079
PMCID: PMC3625450
NIHMSID: NIHMS441379
PMID: 23419157

The interplay between CD4 cell count, viral load suppression and duration of ART on mortality in a resource-limited setting

Alana T Brennan,1,2 Mhairi Maskew,2 Ian Sanne,2,3,4 and Matthew P Fox1,2,5
Author information Copyright and License information PMC Disclaimer

Abstract

Objective

To examine the interaction between CD4 cell count, viral load suppression and duration of ART on mortality.

Methods

Cohort analysis of HIV-infected patients initiating ART between April 2004 and June 2011 at a large public-sector clinic in Johannesburg, South Africa. A log-linear model with Poisson distribution was used to estimate risk of death as a function of the interaction between current CD4 count, current viral load suppression and duration on ART in 12-month intervals. We calculated predicted mortality using estimated coefficients within combinations of predictors.

Results

Among 14,932 ART patients, 1,985 (13.3%) died. Current CD4 was the strongest predictor of death (<50 vs. ≥550 cells/mm3 - RR: 46.3; 95%CI: 26.8–80), while unsuppressed current viral load vs. suppressed (RR: 1.8; 95%CI: 1.5–2.1) and short duration of ART (0–11.9 vs. 66–71.9 months RR: 1.7; 95%CI: 1.2–2.3) also predicted death. Our interaction model showed that mortality was highest in the first 12-months on treatment across all CD4 and viral load strata. As current CD4 and duration on ART increased and viral load suppression occurred, mortality dropped. CD4 count was the strongest predictor of death. The relative effect of current CD4 count varied strongly by viral load and duration of ART (from 1.3 to 55). Lack of suppression increased the risk of mortality upwards of 6-fold depending on time on ART and current CD4.

Conclusions

Our findings show that while CD4 count is the strongest predictor of death, the effect is modified by viral load and the duration of ART. Assessment of risk should take into account all three factors.

Keywords: current CD4 count, current viral load, antiretroviral therapy, mortality, resource limited setting

INTRODUCTION

Antiretroviral therapy (ART) is highly effective at reducing morbidity and mortality through virologic suppression and immune function restoration in treatment-naive HIV-positive patients [1–4]. However, in low income countries, including those in sub-Saharan Africa, an estimated 6.5 to 8.9%, of HIV-positive patients receiving ART die within the first 12-months on treatment, with the majority of those deaths taking place in the first 3–6 months of treatment [5–7]. Most of these deaths are attributable to late presentation for care characterized by low CD4 cell counts at the start of ART (median CD4 100–150 cells/mm3 in programmes in sub-Saharan Africa [7–14]), advanced World Health Organization (WHO) clinical disease stage, low body mass index and anaemia [15–20].

Globally, as ART patients begin to age and remain on treatment longer, baseline predictors of poor outcomes, though valuable, may not provide as much information about long-term risk of HIV-related morbidity and mortality as measures updated over the course of treatment. A study examining the association between a patient’s response to treatment and risk of death found that the current, updated CD4 cell count is a more relevant and stronger predictor of mortality over time than the baseline CD4 cell count [21,22]. Other measures of sustained immunosuppression, such as cumulative person time with low CD4 cell counts (e.g. <100 cells/mm3), also are strong predictors of mortality [21], supporting the notion that baseline CD4 count alone may not be the most appropriate way to assess risk over time [14,15,23,24].

While CD4 cell count is the strongest determinant of mortality in HIV-positive patients who adhere to ART, few studies conducted in rich settings have directly explored the interactive relationship between CD4 count, viral load and time on treatment [25,26]. One study suggests that long-term changes in CD4 cell count after ART initiation depend on interactions between CD4 cell count at treatment initiation, viral load response, and time on treatment. Though this study did not assess the impact of these factors on mortality, it provides some insight into the interplay between these three crucial factors in determining overall patient risk for mortality during treatment. As none of these studies addressed the interaction between CD4 count and viremia, we set out to accurately measure the risk of mortality of patients on ART over time as a function of the interaction between current CD4 count, viral suppression and time on ART over the first six years of treatment using data from one of the largest HIV treatment clinics in South Africa (Themba Lethu Clinic in Johannesburg).

METHODS

Ethics Statement

The University of the Witwatersrand and Boston University provided ethical approval of the study. The study was conducted as an unlinked, prospective analysis of a data set that did not contain any individual identifiers.

Study site

Themba Lethu Clinic was opened at the Helen Joseph Hospital in April 2004 and has enrolled nearly 31,000 patients in care [7]. More than 22,000 of these patients have stated ART. Clinic staff provides HIV care according to South African National Department of Health guidelines [28,29]. All laboratory work is processed by the National Health Laboratory Service. First-line ART regimens before April 2010 consisted of stavudine or zidovudine with lamivudine and either efavirenz or nevirapine [28]. Tenofovir was substituted for stavudine after April 2010 [29]. After treatment has begun, patients are seen for follow-up visits and antiretroviral drug pickups monthly for the first 6–12 months on treatment, then every 2 months thereafter if stable. Patients have their first lab monitoring after 4 months to assess viral load suppression (NucliSENS EasyQ® HIV-1 assay, bioMérieux Clinical Diagnostics, France) and changes in CD4 count (PanLeucogated CD4+ Flowcount®, Beckman Coulter-Immunotech, France); they are monitored annually thereafter.

Study population

We performed a cohort analysis of data collected prospectively as part of routine care at the Themba Lethu clinic. We included all non-pregnant, ART-naïve, HIV-positive patients ≥18 years of age who started a standard public-sector first-line ART regimen at the clinic between April 2004 and June 2011. Patients included contributed both a CD4 count and a viral load measure during at least one 12-month time period. We excluded pregnant women as they are initiated on ART at higher average CD4 counts and have variable CD4 counts compared to the general population [30].

Study Variables

The primary outcome for this study was death. At Themba Lethu mortality is ascertained via family or hospital report, active tracing and linkage with the South African National Vital Registration Infrastructure Initiative, a system estimated to have 90% sensitivity for adults [31, 32]. Eligible patients contributed person-time from the date of ART initiation until the date of the earliest of: (1) death; (2) loss to follow-up (defined as not having attended the clinic for 4 months); (3) transfer; or (4) June 1, 2012. For analysis we divided person-time for each subject into 12-month periods starting from ART initiation. For each 12-month period a patient contributed one observation indicating whether death occurred, current viral load and current CD4 cell count. Subjects could contribute multiple observations to the analysis but not in the same time period. CD4 count and viral load were the only two variables in our analysis that were time dependent, and therefore “current” refers to the most recent value of that variable. All other covariates were fixed at ART initiation in this analysis.

Data Analyses

Patient baseline characteristics were summarized with descriptive statistics stratified by vital status. To estimate the risk of death as a function of current viral load status, current CD4 count and time on ART, we included multiple observations for each patient indicating their updated exposures and outcome status. Since we are interested in estimating absolute risks and not survival, we modeled one-year and overall risk of death as a log-linear function of these covariates using two different models, each using a Poisson distribution [36].

First we looked for predictors of overall mortality on ART by modeling death as a function of current CD4 count (categorized as: <50, 50–99, 100–149, 150–249, 250–349, 350–449, 450–549 and ≥550 cells/mm3), current viral load (<400 vs. ≥400 copies/mL) and duration on ART in yearly intervals referred to by the month the interval began with no interaction terms between predictors. Since this is a predictive model we included covariates at ART initiation a priori deemed important or which had a p-value <0.2 (age, gender, tuberculosis, body mass index, haemoglobin level and WHO stage).

To assess the interaction between the three predictors, we next fitted a model of the risk of death over one year as a function of the square root of current CD4 count, current viral load (<400 vs. ≥400 copies/mL) and the square root of duration on ART. In this model we also included all possible two- and three-way interactions terms between these three variables to allow the associations between the predictors and the outcome to vary within levels of each of the other covariates. Because we used a log-linear model assessing relative risks, our approach allowed for effect measure modification on the relative, not the absolute scale [37]. We used the estimated coefficients from this model to calculate predicted mortality within each combination of current viral load, CD4 count and time on treatment (Appendix 1). This model was also adjusted for previously mention covariates. To show the predicted values we estimated mortality for subjects who were female, 25–29.9 years of age, WHO stage I/II, body mass index ≥18.5 kg/m2, haemoglobin ≥10 g/dL and no tuberculosis at ART initiation.

Data for current CD4 count (22.9%) and current viral load (21.4%) were not available for all patients. We employed multiple imputation by chained equations method using the PROC MI command in SAS to deal with the missingness [33]. In order to use this method we are assuming that the data in our cohort are missing at random, since the missingness is most likely associated with the outcome (death) [34]. All prediction equations included log age at initiation of treatment, gender, square root of CD4 count (baseline and updated), square root of time period, log of viral load, haemoglobin at ART initiation (continuous), body mass index at ART initiation (continuous), WHO stage (I/II, III and IV) and tuberculosis at ART initiation. Indicator variables for death and loss to follow-up were also added to the equations but were not imputed. All models were fitted using 25 imputed datasets and estimated coefficients combined by averaging with the MIANALYZE procedure in SAS [35]. Appropriate standard errors were calculated using the within and between imputation standard errors of the estimates using Rubin’s rules [34]. The analysis of the interaction between current CD4 count, current viral load and time on ART on mortality was also performed on the original dataset prior to multiple imputation with complete cases only (Appendix 2).

RESULTS

Cohort Characteristics

A total of 14,932 patients were included in our analysis. Patients had a median follow-up time on ART of 28.9 months (Interquartile Range (IQR): 12.5–54.8). During follow-up, 1,985 (13.3%) patients died in a median of 5.8 months (IQR: 1.6–18.1). Of the remainder, 7,072 (47.4%) were alive and in care, 3,288 (22.0%) were lost to follow-up and 2,587 (17.3%) had transferred to another treatment facility. Patients who died were slightly older, presented for treatment with a substantially lower median CD4 cell count and at a more advanced stage of their disease (higher proportion with a WHO clinical stage III/IV condition) than those who did not (Table 1).

Table 1

Baseline characteristics of patients on antiretroviral therapy at the Themba Lethu Clinic in Johannesburg, South Africa by vital status (n = 14,932)

Vital Status
CharacteristicsDied
(n=1,985)
n (%)		Alive
(n=12,947)
n (%)		Total
(n=14,932)
n (%)
Gender
Male906 (45.6)4,864 (37.6)5,770 (38.6)
Female1079 (54.4)8,083 (62.43)9,162 (61.4)
Age (years)
18–24.987 (4.4)583 (4.5)670 (4.5)
25–29.9230 (11.6)1883 (14.5)2113 (14.2)
30–39.9839 (42.3)5884 (45.5)6723 (45.0)
40–49.9538 (27.1)3363 (26.0)3901 (26.1)
≥50291 (14.7)1234 (9.5)1525 (10.2)
Age at ART Initiationmedian (IQR)37.7 (32.3–45.2)36.6 (31.3–43.0)36.7 (31.4–43.2)
CD4 at ART Initiation
0–50 cells/mm3982 (49.5)3869 (29.9)4851 (23.5)
51–100 cells/mm3430 (21.7)2685 (20.7)3115 (20.9)
101–200 cells/mm3455 (22.9)4676 (36.1)5131 (34.4)
201–350 cells/mm3100 (5.0)1553 (12.0)1653 (11.1)
>350 cells/mm318 (0.9)164 (1.3)182 (1.2)
CD4 at ART Initiation (cells/mm3)median (IQR)51 (16–112)99 (39–168)92 (35–162)
WHO Stage at ART Initiation
I/II898 (45.2)7816 (60.4)8714 (58.4)
III882 (44.4)4322 (33.4)5204 (34.9)
IV205 (10.3)809 (6.3)1014 (6.8)
Tuberculosis at ART Initiation
Yes380 (19.1)1831 (14.1)2211 (14.8)
No1605 (80.9)11,116 (85.9)12,721 (85.9)
First-line ART Regimen
d4T/3TC/EFV1696 (85.4)9504 (73.4)11,200 (75.0)
d4T/3TC/NVP89 (4.5)800 (6.2)889 (6.0)
TDF/3TC/EFV142 (7.2)2093 (16.2)2235 (15.0)
TDF/3TC/NVP6 (0.3)134 (1.0)140 (0.9)
AZT/3TC/EFV50 (2.5)368 (2.8)418 (2.8)
AZT/3TC/NVP2 (0.1)48 (0.4)50 (0.3)
Time on ART (months)median (IQR)5.8 (1.6–18.1)33.1 (16.0–58.1)28.9 (12.5–54.8)
Haemoglobin at ART Initiation (g/dL)median (IQR)10.6 (9.1–12.2)11.6 (10.1–13.1)11.5 (9.9–13.0)
Body Mass Index at ART Initiationmedian (IQR)20.1 (17.5–24.2)21.8 (19.1–25.4)21.6 (18.9–25.3)

ART, antiretroviral therapy; WHO, World Health Organization; d4T, stavudine; 3TC, lamivudine; EFV, efavirenz; TDF, tenofovir; AZT, zidovudine

Patients who were lost to follow-up were on treatment for a median time of 11.8 months (IQR: 5.0–27.9). They were predominantly male (43.3% vs. 37.3%) and younger at ART initiation (35.6 vs. 37.0 years) then those not lost. However, they were similar to those not lost in terms of clinical factors at ART initiation: CD4 count (88 vs. 93 cells/mm3), body mass index (21.0 vs. 21.7 kg/m2), haemoglobin (11.3 vs. 11.5 g/dL) and ART regimen (79.0% vs. 73.9% on lamivudine-stavudine-efavirenz). Patients who transferred to another facility during follow-up were similar to those included in the analysis, except that a higher proportion of them were females (67% vs. 62%).

Mortality by Duration on ART, Viral Load Suppression and CD4 Count

Model 1 in Table 2 shows the results of our analysis of time updated measures of CD4 count, viral load and duration on ART with no interactions. The adjusted model showed that shorter duration on treatment, lower current CD4 count and an unsuppressed current viral load were all associated with an increased risk of death. Compared to a current CD4 cell count ≥550 cells/mm3, current CD4 count <50 cells/mm3 (RR: 41.6; 95%CI: 24.4–71.0 cells/mm3) and 50–99 cells/mm3 (RR: 27.3; 95%CI: 16.5–45.2) carried a substantially higher risk of death. Having an unsuppressed viral load vs. suppressed (RR: 1.8; 95%CI: 1.5–2.1) and shorter duration of ART (e.g. 0–11.9 vs. ≥48 months RR: 1.7; 95%CI: 1.2–2.3) were also predictive of mortality.

Table 2

Crude and adjusted predictors of mortality among patients on antiretroviral therapy at the Themba Lethu Clinic in Johannesburg, South Africa (n =14,932)

N (%) mortalityCrude
RR (95% CI)		Adjusted Model1
RR (95% CI)
Current Viral Load (copies/mL)
      <4001195 (11.0)ReferenceReference
      ≥400790 (19.6)3.6 (3.3–3.9)1.8 (1.5–2.1)
Current CD4 Count (cells/mm3)
      ≥55036 (1.2)ReferenceReference
      450–54961 (3.4)2.6 (1.7–3.9)2.2 (1.2–4.0)
      350–44992 (4.0)2.7 (1.8–4.0)2.9 (1.7–4.9)
      250–349240 (9.3)6.0 (4.1–8.6)5.0 (3.0–8.2)
      150–249456 (18.4)12.7 (8.8–18.1)8.6 (5.3–13.9)
      100–149332 (31.6)27.8 (19.3–39.8)16.8 (10.1–27.8)
      50–99360 (39.2)48.8 (34.1–69.9)27.3 (16.5–45.2)
      <50408 (54.7)96.5 (67.0–138.9)41.6 (24.4–71.0)
Time (months)
      0–11.91328 (36.9)5.9 (4.4–7.9)1.7 (1.2–2.3)
      12–23.9290 (10.0)1.7 (1.3–2.3)0.9 (0.7–1.3)
      24–35.9143 (6.4)1.1 (0.8–1.6)0.8 (0.6–1.1)
      36–47.991 (5.3)1.0 (0.7–1.4)0.8 (0.6–1.1)
      ≥48116 (5.0)ReferenceReference
Age (years) at ART initiation
      18–24.957 (13.0)1.2 (1.0–1.6)1.2 (0.9–1.5)
      25–29.9230 (10.9)ReferenceReference
      30–39.9839 (12.5)1.1 (0.9–1.3)1.1 (1.0–1.3)
      40–49.9538 (13.8)1.3 (1.1–1.5)1.2 (1.0–1.4)
      ≥50291 (19.1)1.9 (1.6–2.3)1.7 (1.4–2.0)
Gender
      Female1079 (11.8)ReferenceReference
      Male906 (15.7)1.4 (1.3–1.6)1.0 (0.9–1.1)
Tuberculosis at ART initiation
      No1605 (12.6)ReferenceReference
      Yes380 (17.2)1.3 (1.2–1.5)1.0 (0.9–1.2)
WHO Stage at ART initiation
      I/II898 (10.3)ReferenceReference
      III882 (17.0)1.6 (1.4–1.8)1.2 (1.1–1.4)
      IV205 (20.2)2.0 (1.7–2.4)1.3 (1.1–1.5
Body mass index at ART initiation
      ≥18.5 kg/m21313 (11.2)ReferenceReference
      <18.5 kg/m2672 (20.7)2.1 (1.9–2.3)1.3 (1.2–1.5)
Hemoglobin at ART initiation
      ≥10.0 g/dL1212 (10.9)ReferenceReference
      <10.0 g/dL773 (20.3)2.1 (1.9–2.3)1.5 (1.4–1.7)
1Model is adjusted for current CD4 count, current viral load, time on ART, female gender, years of age (18–24.9, 30–39.9, 40–49.9 ≥50 vs. 25–29.9), WHO stage III/IV vs. I/II, body mass index <18.5 kg/m2 vs. ≥18.5 kg/m2, haemoglobin <10 g/dL vs. ≥10 g/dL and tuberculosis vs. no tuberculosis at ART initiation. Model was not adjusted for CD4 count at ART initiation because of the collinear relationship between current CD4 count and CD4 count at ART initiation as described in the methods

ART antiretroviral therapy; WHO, World Health Organization

The previous model ignores the interactions between duration on treatment, current CD4 count and viral load. Table 3 and Figure 1 show predicted mortality from a model including current CD4 count, current viral load and duration on ART as well as all two- and three-way interactions between them (Appendix 1-parameter estimates and p-values). Predicted one year mortality ranged from 0.3% to 24.8%. For those with a CD4 cell count of 350 cells (current recommended WHO threshold) mortality was below 5% for all time periods for unsuppressed patients and ≤1% for suppressed patients. For those with a CD4 cell count of 200 cells (previous WHO threshold) mortality was below 8% for all time periods for unsuppressed patients and below ≤3% for suppressed patients. One-year mortality was consistently higher among those with lower CD4 counts and among those virally unsuppressed at all time points. Mortality was highest in the first year on treatment across all current CD4 and viral load strata and consistently declined over time.

An external file that holds a picture, illustration, etc. Object name is nihms441379f1.jpg
Predicted one year mortality by current viral load, current CD4 cell count and time on ART among patients on antiretroviral therapy at the Themba Lethu Clinic in Johannesburg, South Africa

Model also adjusted for female gender, years of age (18–24.9, 30–39.9, 40–49.9 ≥50 vs. 25–29.9), WHO stage I/II and III vs. IV, body mass index ≥18.5 kg/m2 vs. <18.5 kg/m2, haemoglobin ≥10 g/dL vs. <10 g/dL and no tuberculosis vs. tuberculosis at ART initiation. Model parameters given in appendix 1.

Table 3

Predicted risk of one year mortality by current viral load status, current CD4 count and time on ART among patients on antiretroviral therapy at the Themba Lethu Clinic in Johannesburg, South Africa

Viral
load
(copies/ml)		Months
on ART		Current CD4 Count
% Mortality
25 cells/mm350 cells/mm3100 cells/mm3200 cells/mm3250 cells/mm3350 cells/mm3450 cells/mm3550 cells/mm3
024.8 (20.1–29.4)19.0 (15.8–22.3)13.1 (10.9–15.3)7.76 (6.22–9.30)6.28 (4.90–7.66)4.36 (3.20–5.51)3.17 (2.19–4.16)2.40 (1.55–3.24)
≥4001215.6 (11.3–17.8)11.3 (8.99–13.6)7.84 (6.39–9.28)4.69 (3.82–5.55)3.81 (3.08–4.54)2.66 (2.09–3.24)1.95 (1.48–2.43)1.48 (1.08–1.88)
2411.7 (8.57–14.9)9.09 (6.88–11.3)6.34 (4.96–7.71)3.81 (3.00–4.61)3.10 (2.42–3.78)2.17 (1.63–2.71)1.60 (1.15–2.05)1.21 (0.83–1.60)
369.92 (6.83–13.0)7.70 (5.55–9.84)5.38 (4.05–6.71)3.25 (2.46–4.03)2.65 (1.98–3.32)1.86 (1.32–2.40)1.37 (0.92–1.83)1.05 (0.65–1.44)
<400488.61 (5.61–11.6)6.69 (4.61–8.78)4.69 (3.40–5.98)2.84 (2.07–3.61)2.32 (1.66–2.98)1.63 (1.10–2.17)1.21 (0.75–1.67)0.92 (0.52–1.32)
607.61 (4.69–10.5)5.92 (3.90–7.94)4.16 (2.91–5.40)2.52 (1.77–3.28)2.06 (1.41–2.72)1.46 (0.92–2.00)1.08 (0.62–1.54)0.82 (0.42–1.23)
023.0 (18.1–28.0)14.7 (12.0–17.4)7.75 (6.52–8.99)3.15 (2.64–3.66)2.19 (1.80–2.58)1.17 (0.92–1.42)0.68 (0.51–0.85)0.42 (0.29–0.54)
1213.1 (9.49–16.8)8.74 (6.60–10.9)4.92 (3.92–5.92)2.18 (1.82–2.55)1.58 (1.32–1.83)0.89 (0.74–1.04)0.55 (0.44–0.65)0.35 (0.28–0.43)
2410.4 (6.81–14.0)7.06 (4.92–9.20)4.08 (3.07–5.09)1.88 (1.52–2.23)1.37 (1.13–1.62)0.80 (0.65–0.95)0.50 (0.40–0.61)0.33 (0.25–0.41)
368.72 (2.19–12.3)6.00 (3.88–8.11)3.53 (2.53–4.53)1.67 (1.32–2.03)1.24 (0.99–1.48)0.74 (0.59–0.88)0.47 (0.36–0.58)0.31 (0.23–0.40)
487.52 (7.08–11.0)5.23 (3.15–7.30)3.13 (2.13–4.12)1.52 (1.16–1.87)1.13 (0.89–1.38)0.68 (0.53–0.84)0.44 (0.32–0.56)0.30 (0.20–0.40)
606.59 (3.26–9.93)4.63 (2.61–6.66)2.81 (1.93–3.79)1.39 (1.04–1.75)1.05 (0.80–1.30)0.64 (0.48–0.80)0.42 (0.30–0.55)0.29 (0.18–0.39)

Model also adjusted for female gender, years of age (18–24.9, 30–39.9, 40–49.9 ≥50 vs. 25–29.9), WHO stage I/II and III vs. IV, body mass index ≥18.5 kg/m2 vs. <18.5 kg/m2, haemoglobin ≥10 g/dL vs. <10 g/dL and no tuberculosis vs. tuberculosis at ART initiation. Model parameters given in appendix 1.

Despite being the strongest predictor of death, the relative effect of CD4 count was strongly modified by viral suppression and time on treatment (Appendix 3-relative risks for mortality by current CD4 count, viral load and duration on ART). The effect of CD4 count was strongest among those virally suppressed and in their first year on treatment when mortality was at its highest. When comparing a CD4 cell count of 25 to 550 cells/mm3 we found a 10- to 55-fold increased risk of death in patients across all time periods and viral load strata. In the first 12-months on ART for patients with a detectable viral load, those with a CD4 count of 25 cells/mm3 had over 10 times the risk of one year mortality compared to those with a CD4 count 550 cells/mm3 (24.8% vs. 2.4%). For comparable patients (in the first 12-months on ART) but with a undetectable current viral load where overall mortality is slightly lower, those with a current CD4 of 25 cells/mm3 had roughly a 55-fold mortality risk of those with a CD4 cell count of 550 cells/mm3. With increasing duration on ART, however, the relative effect of a CD4 cell count on one-year mortality shows strong modification by viral suppression. Over time, among virally unsuppressed patients, the relative effect of CD4 count on one-year mortality stayed steady (e.g. the effect of a 25 vs. 550 cells/mm3 decrease from a RR of 10.4 at 0 months to a RR of 9.3 at 60 months) but fell over time among virally suppressed patients (e.g. from a RR of 55.4 at 0 months to a RR of 22.6 at 60 months).

The relative effect of viral suppression consistently increased with higher CD4 counts but declined over time for those with a current CD4 count >200 cells/mm3 (Appendix 3). The largest effect of viral suppression (vs. non suppression) is for patients with low one-year predicted mortality. For those with a CD4 count of 550 (the highest CD4 for which we estimated mortality risk) in the first year of ART the relative reduction in risk for suppressed patients was 6-fold (0.42% vs. 2.4% for suppressed and unsuppressed patients, respectively) with a smaller effect in lower CD4 count strata.

The effect of duration on ART also was modified by current viral load and CD4 count. For both suppressed and unsuppressed patients, time was an important predictor of one-year mortality (Appendix 3). The effect consistently declines over time but is always strongest amongst those with low CD4 cell counts (ranging from a RR of 3.3 to 2.9 for unsuppressed patients and 3.5 to 1.4 for suppressed patients comparing month 0 to month 60 among those with a CD4 count of 25 and 550 cells, respectively). The effect of time on mortality holds fairly steady after 36 months on ART for all CD4 count and viral load strata.

DISCUSSION

Despite the clear survival benefits of ART, mortality in the first 24-months of treatment among HIV-positive patients is substantially higher than in the general uninfected population, particularly among patients who present for treatment severely immunocompromised [38]. Therefore the main goals of ART are to get patients onto treatment early, achieve viral suppression in the shortest time possible and then sustain suppression allowing patients’ CD4 count to increase. These targets lead to a marked reduction in poor clinical outcomes and an increase in life expectancy [39–42,43]. All three of these factors are critical to survival, yet the interplay between these factors remains unclear as this requires large sample sizes and long follow-up. In this analysis, we assessed the role of current CD4 count on mortality while accounting for current viral suppression and duration of ART. We found that current CD4 count was the largest relative predictor of death on ART regardless of how long patients had been on treatment or whether or not they currently had a detectable viral load. Relative rises in mortality comparing patients with a CD4 count of 25 to 550 cells/mm3 ranged from to 10 to 55-fold increases depending on the duration on ART and viral load status. These results support previous findings that CD4 cell count is the main driver of a patient’s risk of mortality on ART [13,17–21,26].

We also modelled interactions between these three critical factors that drive the risk of mortality and demonstrate that the effect of CD4 count on death varies by current viral load status, consistent with recent findings [26]. Not only was predictive mortality higher among patients who did not achieve viral load suppression, but they were at upwards of a 6-fold increased risk of death in the first 12 months on ART compared to patients who did achieve viral load suppression, depending on the patients CD4 count. While still a strong driver of mortality among virally suppressed patients, the effect of CD4 count was reduced, as overall risk of death is greater in unsuppressed patients.

In our cohort, the majority of patients (85.8%) achieved viral load suppression in the first 12 months on ART. However, not all patients on ART are able to achieve virologic suppression, either as a result of poor adherence to treatment [46] or due to resistance [47]. Several studies have shown that, while not as strong a predictor as CD4 count, circulating virus remains an important prognostic indicator of HIV disease progression [38, 41,42] and that rapid and consistent viral suppression is essential to maintaining positive clinical outcomes [38]. Our results show that the effect of viral load, while more stable than the effect of CD4, is not constant over time and appears strongest early on in treatment among patients with higher CD4 counts, when the overall risk of death is lower and viral suppression plays a stronger role.

Our results are comparable with an analysis of predicted mortality in cohorts throughout sub-Saharan Africa [7]. The general similarity of our findings confirms the very high mortality among patients with low CD4 counts but further develops predictive models based on viral suppression. Our findings emphasize the interaction between time, viral load and CD4 count and demonstrate how relative effects are modified by each of the other predictors.

In rich environments, HIV treatment monitoring typically includes frequent viral load monitoring, viral resistance testing and regular CD4 count measurements. However, due to the prohibitive cost of HIV RNA monitoring, standard care in many settings in sub-Saharan Africa consists of clinical monitoring, coupled with routine CD4 measurements when possible [48]. Our results confirm that lack of viral suppression continues to predict mortality. Because clinical deterioration and CD4 decline can occur well after virologic failure, viral load measures allow for faster and more appropriate use of second-line ART [49].

Our findings should be considered alongside their limitations. First, as regards our model, the preferred approach to analyzing the interactive relationship between CD4 count, viral load and time on ART would have been on the additive scale [37], which we tried to do, but failed when our linear model broke down. Second, while the South African National Vital Registration is highly sensitive [33,32], there is a 6-month delay in updating the registry which could result in under-ascertainment of deaths. However, patients (n=193) who were lost to follow-up <6 months prior to the linkage (June 2012) were removed from the analysis as they may not have been included in the registry if they had died because of delays in reporting. Also, because this misclassification is likely unrelated to any of our three primary exposures, the expectation is that this would reduce the size of estimated comparisons. Third, loss in our analysis refers to those patients that do not have an observable outcome. More than 20% of patients in our cohort were considered lost to follow-up and 17% transferred out. Compared to those included in the analysis, patients lost from care were predominantly male and younger, while the majority of transferred patients were female. Although we are less concerned with those transferred since females are at lower risk of mortality than males [50], it is important to acknowledge that there is likely some selection bias, potentially making our results underestimates of mortality since male patients and those patients who leave care are more likely to stop treatment and increase their risk of death [50,51]. Fourth, multiple imputation helps make it possible to handle missing data routinely and improve the validity of research. However, the procedure requires the user to model the distribution of each variable with missing values, in terms of the observed data [33]. The validity of results from multiple imputation depends on such modeling being done carefully and is based on the assumption that our data is missing at random. Deviations from this could have led to unpredictable biases in our parameter estimates.

Conclusion

Long-term virologic suppression helps to ensure the recovery of CD4 cells to levels that reduce the risk of opportunistic infections and increase life expectancy. Our findings show that while low current CD4 count is the largest predictor of one-year mortality on treatment, this relative effect is modified by current viral suppression and time on ART and that all three are important to assess when evaluating patient risk. Future efforts to refine mortality predictions should assess the role of body mass index, anemia, tuberculosis and other opportunistic infections.

Acknowledgment

Disclaimer

We express our gratitude to the directors and staff of Themba Lethu Clinic and to Right to Care, the Non-Governmental Organization supporting the study site through a partnership with USAID. We also thank the Gauteng and National Department of Health for providing for the care of the patients at the Themba Lethu Clinic as part of the Comprehensive Care Management and Treatment plan. Most of all we thank the patients attending the clinic for their continued trust in the treatment provided at the clinic.

Funding was provided by USAID. Matthew P Fox was also supported by the National Institute of Allergy and Infectious Diseases. The opinions expressed herein are those of the authors and do not necessarily reflect the views of National Institute of Health, NIAID, USAID, the Themba Lethu Clinic or Right to Care. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Appendix 1. Model results for predicted risk of death by current CD4, current viral load and duration on ART and all possible two- and three-way interactions terms between these three variables

ParameterBetaStandard
Error		95% Confidence Limitsp-value
Intercept−0.3819760.158555−0.6955−0.068430.0173
Age 18–24.9 vs. 25–29.9 years0.1484670.1287−0.10380.400730.2487
Age 30–39.9 vs. 25–29.9 years0.1035350.076338−0.04610.253160.175
Age 40–49.9 vs. 2–29.9 years0.1829870.0817850.022670.34330.0253
Age ≥50 vs. 25–29.9 years0.4968790.0918190.31690.67686<.0001
Male vs. Female gender−0.0497240.049708−0.14720.047740.3172
Tuberculosis vs. Non-tuberculosis0.0281780.067554−0.10420.160590.6766
Hb <10.0 g/dL vs. ≥10.0 g/dL0.3960340.0519010.294250.49782<.0001
BMI <18.5 kg/m2 vs. ≥18.5 kg/m20.2786730.0532420.174240.38311<.0001
WHO stage III vs. WHO stage I/II0.2009550.057190.088850.313060.0004
WHO stage IV vs. WHO stage I/II0.2634870.0814580.103810.423170.0012
Viral load ≥400 vs. <400 copies/mL0.759090.167412−0.7118−0.049210.0247
Current CD4 count½−0.3805260.011185−0.2398−0.1953<.0001
Time period½0.82351620.048822−0.2907−0.097640.0001
2-way interaction terms
Current CD4 count½ * Viral load ≥4000.0906110.0147610.061450.11977<.0001
Time period½ * Viral load ≥4000.0372350.065396−0.0920.166420.5699
Current CD4 count½ * Time period½0.0062640.0030780.000180.012350.0437
3-way interaction terms
Current CD4 count½ * Time period½ * Viral load ≥400−0.0054480.00443−0.01420.003290.2203

BMI, body mass index; Hb, haemoglobin; WHO, World Health Organization

Appendix 2. Use of original dataset of complete cases to predict the risk of one year mortality by current viral load status, current CD4 count and time on ART among patients on antiretroviral therapy at the Themba Lethu Clinic in Johannesburg, South Africa

Viral
load
(copies/ml)		Months
on ART		Current CD4 Count
% Mortality
25 cells/mm350 cells/mm3100 cells/mm3200 cells/mm3250 cells/mm3350 cells/mm3450 cells/mm3550 cells/mm3
011.0 (7.14–16.9)6.70 (4.46–10.1)3.33 (2.16–5.15)1.24 (0.71–2.18)0.84 (0.44–0.16)0.42 (0.19–0.90)0.23 (0.09–0.56)0.14 (0.05–0.37)
≥400127.16 (4.67–11.0)4.56 (3.05–6.81)2.40 (1.61–3.59)0.97 (0.61–1.55)0.68 (0.41–1.12)0.36 (0.20–0.65)0.21 (0.11–0.41)0.13 (0.06–0.27)
246.00 (3.62–9.96)3.88 (2.44–6.18)2.10 (1.34–3.29)0.88 (0.52–1.48)0.62 (0.35–1.10)0.34 (0.17–0.67)0.20 (0.09–0.44)0.12 (0.05–0.31)
365.24 (2.92–9.41)3.44 (2.03–5.82)1.89 (1.14–3.13)0.81 (0.45–1.48)0.58 (0.30–1.13)0.32 (0.14–0.72)0.19 (0.07–0.50)0.12 (0.04–0.36)
<400484.68 (2.42–9.04)3.10 (1.72–5.58)1.73 (0.99–3.03)0.76 (0.39–1.49)0.55 (0.26–1.17)0.31 (0.12–0.78)0.19 (0.06–0.56)0.12 (0.03–0.42)
604.04 (1.89–8.65)2.71 (1.38–5.32)1.55 (0.82–2.93)0.70 (0.32–1.54)0.51 (0.21–1.24)0.29 (0.10–0.88)0.18 (0.05–0.67)0.12 (0.03–0.53)
09.78 (6.17–15.5)5.90 (3.89–8.95)2.89 (1.99–4.20)1.05 (0.73–1.51)0.70 (0.48–1.02)0.35 (0.23–0.53)0.19 (0.12–0.30)0.11 (0.06–0.19)
125.24 (3.07–8.94)3.48 (2.15–5.66)1.96 (1.28–2.99)0.86 (0.60–1.24)0.62 (0.44–0.88)0.35 (0.25–0.50)0.21 (0.15–0.31)0.14 (0.09–0.21)
244.05 (2.13–7.69)2.80 (1.58–4.97)1.66 (1.02–2.71)0.80 (0.54–1.18)0.59 (0.41–0.86)0.35 (0.25–0.51)0.23 (0.15–0.51)0.15 (0.10–0.23)
363.32 (1.58–6.96)2.37 (1.23–4.56)1.47 (0.85–2.53)0.75 (0.49–1.15)0.57 (0.38–0.85)0.36 (0.24–0.53)0.24 (0.15–0.36)0.16 (0.10–0.27)
482.81 (1.23–6.45)2.06 (0.99–4.27)1.32 (0.73–2.41)0.71 (0.45–1.12)0.55 (0.36–0.85)0.36 (0.23–0.55)0.24 (0.15–0.39)0.17 (0.10–0.30)
602.43 (0.97–6.04)1.82 (0.82–4.03)1.21 (0.63–2.31)0.68 (0.41–1.11)0.54 (0.34–0.85)0.36 (0.23–0.57)0.25 (0.15–0.43)0.18 (0.10–0.34)

Model also adjusted for female gender, years of age (18–24.9, 30–39.9, 40–49.9 ≥50 vs. 25–29.9), WHO stage I/II and III vs. IV, body mass index ≥18.5 kg/m2 vs. <18.5 kg/m2, haemoglobin ≥10 g/dL vs. <10 g/dL and no tuberculosis vs. tuberculosis at ART initiation. Model parameters given in appendix 1.

Appendix 3. Relative risks of mortality by current viral load status, current CD4 count and time on ART among patients on antiretroviral therapy at the Themba Lethu Clinic in Johannesburg, South Africa

Relative risk of mortality by current CD4 counts
Viral
load		Time on
treatment
(months)		25 cells/mm350 cells/mm3100 cells/mm3200 cells/mm3250 cells/mm3350 cells/mm3450 cells/mm3550 cells/mm3
010.408.005.523.262.641.831.33Reference
129.887.645.313.182.581.801.32Reference
≥400249.667.495.233.142.561.791.32Reference
369.517.385.163.112.541.781.31Reference
489.377.295.113.092.531.781.31Reference
609.267.215.073.072.511.771.31Reference
055.3835.2918.667.585.272.811.63Reference
1237.1124.7313.946.194.472.531.55Reference
<4002431.4321.3512.355.704.172.431.52Reference
3627.6819.0711.265.343.962.351.50Reference
4824.8617.3310.415.063.782.281.48Reference
6022.6215.949.724.823.642.231.46Reference
Relative risk of mortality by current viral load status
Time on
treatment
(months)		Viral
load		25 cells/mm350 cells/mm3100 cells/mm3200 cells/mm3250 cells/mm3350 cells/mm3450 cells/mm3550 cells/mm3
0<400ReferenceReferenceReferenceReferenceReferenceReferenceReferenceReference
≥4001.081.301.692.462.863.724.675.72
12<400ReferenceReferenceReferenceReferenceReferenceReferenceReferenceReference
≥4001.111.291.592.142.422.983.564.18
24<400ReferenceReferenceReferenceReferenceReferenceReferenceReferenceReference
≥4001.131.291.552.032.252.713.183.67
Relative risk of mortality by current CD4 counts
36<400ReferenceReferenceReferenceReferenceReferenceReferenceReferenceReference
≥4001.141.291.531.942.142.532.923.32
48<400ReferenceReferenceReferenceReferenceReferenceReferenceReferenceReference
≥4001.151.291.501.872.042.382.723.06
60<400ReferenceReferenceReferenceReferenceReferenceReferenceReferenceReference
≥4001.161.281.481.811.962.262.552.84
Relative risk of mortality by time on treatment
Viral
load		Time on
treatment
(months)		25 cells/mm350 cells/mm3100 cells/mm3200 cells/mm3250 cells/mm3350 cells/mm3450 cells/mm3550 cells/mm3
03.273.233.173.083.053.002.952.91
121.921.911.891.861.851.831.821.80
≥400241.551.541.531.511.511.501.491.48
361.311.301.301.291.291.281.281.27
481.131.131.131.131.131.121.121.12
60ReferenceReferenceReferenceReferenceReferenceReferenceReferenceReference
03.533.192.772.272.091.821.611.44
122.011.901.761.571.501.391.301.22
<400241.591.531.451.351.311.251.191.14
361.331.301.261.201.181.141.111.09
481.141.131.111.091.081.061.051.04
60ReferenceReferenceReferenceReferenceReferenceReferenceReferenceReference

References

1. Bartlett JA, DeMasi R, Quinn J, Moxham C, Rousseau F. Overview of the effectiveness of triple combination therapy in antiretroviral-naive HIV-1 infected adults. AIDS. 2001;15:1369–1377. [PubMed] [Google Scholar]
2. Mocroft A, Ledergerber B, Katlama C, Kirk O, Reiss P, et al. Decline in the AIDS and death rates in the EuroSIDA study: an observational study. Lancet. 2003;362:22–29. [PubMed] [Google Scholar]
3. Pérez-Hoyos S, del Amo J, Muga R, del Romero J, García de Olalla P, et al. Effectiveness of highly active antiretroviral therapy in Spanish cohorts of HIV seroconverters: differences by transmission category. AIDS. 2003;17:353–359. [PubMed] [Google Scholar]
4. Sterne JA, Hernán MA, Ledergerber B, Tilling K, Weber R, et al. Long-term effectiveness of potent antiretroviral therapy in preventing AIDS and death: a prospective cohort study. Lancet. 2005;366:378–384. [PubMed] [Google Scholar]
5. Braitstein P, Brinkhof MW, Dabis F, Schechter M, Boulle A, et al. Antiretroviral Therapy in Lower Income Countries (ART-LINC) Collaboration; ART Cohort Collaboration (ART-CC) groups. Mortality of HIV-1-infected patients in the first year of antiretroviral therapy: comparison between low-income and high-income countries. Lancet. 2006;367:817–824. [PubMed] [Google Scholar]
6. Keiser O, Orrell C, Egger M, Wood R, Brinkhof MW, et al. Public-health and individual approaches to antiretroviral therapy: township South Africa and Switzerland compared. PLoS Med. 2008;5:e148. [PMC free article] [PubMed] [Google Scholar]
7. May M, Boulle A, Phiri S, Messou E, Myer L, Wood R, et al. IeDEA Southern Africa and West Africa. Prognosis of patients with HIV-1 infection starting antiretroviral therapy in sub-Saharan Africa: a collaborative analysis of scale-up programmes. Lancet. 2010;376:449–457. [PMC free article] [PubMed] [Google Scholar]
8. Lawn SD, Harries AD, Anglaret X, Myer L, Wood R. Early mortality among adults accessing antiretroviral treatment programmes in sub-Saharan Africa. AIDS. 2008;22:1897–1908. [PMC free article] [PubMed] [Google Scholar]
9. Lawn SD, Myer L, Orrell C, Bekker LG, Wood R. Early mortality among adults accessing a community-based antiretroviral service in South Africa: implications for programme design. AIDS. 2005;19:2141–2148. [PubMed] [Google Scholar]
10. Stringer JS, Zulu I, Levy J, Stringer EM, Mwango A, et al. Rapid scale-up of antiretroviral therapy at primary care sites in Zambia: feasibility and early outcomes. JAMA. 2006;296:782–793. [PubMed] [Google Scholar]
11. Etard JF, Ndiaye I, Thierry-Mieg M, Gueye NF, Gueye PM, et al. Mortality and causes of death in adults receiving highly active antiretroviral therapy in Senegal: a 7-year cohort study. AIDS. 2006;20:1181–1189. [PubMed] [Google Scholar]
12. Ferradini L, Jeannin A, Pinoges L, Izopet J, Odhiambo D, et al. Scaling up of highly active antiretroviral therapy in a rural district of Malawi: an effectiveness assessment. Lancet. 2006;367:1335–1342. [PubMed] [Google Scholar]
13. Zachariah R, Fitzgerald M, Massaquoi M, Pasulani O, Arnould L, et al. Risk factors for high early mortality in patients on antiretroviral treatment in a rural district of Malawi. AIDS. 2006;20:2355–2360. [PubMed] [Google Scholar]
14. Moh R, Danel C, Messou E, Ouassa T, Gabillard D, et al. Incidence and determinants of mortality and morbidity following early antiretroviral therapy initiation in HIV-infected adults in West Africa. AIDS. 2007;21:2483–2491. [PubMed] [Google Scholar]
15. Lawn SD, Myer L, Harling G, Orrell C, Bekker LG, et al. Determinants of mortality and nondeath losses from an antiretroviral treatment service in South Africa: implications for program evaluation. Clin Infect Dis. 2006;43:770–776. [PubMed] [Google Scholar]
16. van Sighem AI, van de Wiel MA, Ghani AC, Jambroes M, Reiss P, et al. Mortality and progression to AIDS after starting highly active antiretroviral therapy. AIDS. 2003;17:2227–2236. [PubMed] [Google Scholar]
17. Egger M, May M, Chene G, Phillips AN, Ledergerber B, et al. Prognosis of HIV-1-infected patients starting highly active antiretroviral therapy: a collaborative analysis of prospective studies. Lancet. 2002;360:119–129. [PubMed] [Google Scholar]
18. Chene G, Sterne JA, May M, Costagliola D, Ledergerber B, Phillips AN, et al. Prognostic importance of initial response in HIV-1 infected patients starting potent antiretroviral therapy: analysis of prospective studies. Lancet. 2003;362(9385):679–686. Epub 2003/09/06. [PubMed] [Google Scholar]
19. Rosen S, Fox MP. Retention in HIV Care between Testing and Treatment in Sub-Saharan Africa: A Systematic Review. PLoS Med. 2011;8(7) [PMC free article] [PubMed] [Google Scholar]
20. Cornell M, Grimsrud A, Fairall L, Fox MP, van Cutsem G, et al. Temporal changes in programme outcomes among adult patients initiating antiretroviral therapy across South Africa, 2002–2007. AIDS. 2010 Sep 10;24(14):2263–2270. [PMC free article] [PubMed] [Google Scholar]
21. Lawn SD, Little F, Bekker LG, Kaplan R, Campbel E, et al. Changing mortality risk associated with CD4 cell response to antiretroviral therapy in South Africa. AIDS. 2009;23:335–342. [PMC free article] [PubMed] [Google Scholar]
22. Lanoy E, May M, Mocroft A, Phillip A, Justice A, Chene G, et al. Prognosis of patients treated with cART from 36 months after initiation, according to current and previous CD4 cell count and plasma HIV-1 RNA measurements. AIDS. 2009;23(16):2199–2208. [PMC free article] [PubMed] [Google Scholar]
23. Lawn SD, Myer L, Bekker LG, Wood R. Burden of tuberculosis in an antiretroviral treatment programme in sub-Saharan Africa: impact on treatment outcomes and implications for tuberculosis control. AIDS. 2006;20:1605–1612. [PubMed] [Google Scholar]
24. Viard JP, Mocroft A, Chiesi A, Kirk O, Røge B, et al. Influence of age on CD4 cell recovery in human immunodeficiency virus-infected patients receiving highly active antiretroviral therapy: evidence from the EuroSIDA study. J Infect Dis. 2001;183:1290–1294. [PubMed] [Google Scholar]
25. Egger S, Petoumenos K, Kamarulzaman A, Hoy J, Sungkanuparph S, Chuah J, et al. Long-Term Patterns in CD4 Response Are Determined by an Interaction Between Baseline CD4 Cell Count, Viral Load, and Time: The Asia Pacific HIV Observational Database (APHOD) J Acquir Immune Defic Syndr. 2009 Apr 15;50(5):513–520. [PMC free article] [PubMed] [Google Scholar]
26. The Opportunistic Infections Project Team of the Collaboration of Observational HIV Epidemiological Research in Europe (COHERE) in EuroCoord. CD4 Cell Count and the Risk of AIDS or Death in HIV-Infected Adults on Combination Antiretroviral Therapy with a Suppressed Viral Load: A Longitudinal Cohort Study from COHERE. PLoS Med. 2012;9(3) [PMC free article] [PubMed] [Google Scholar]
27. Fox MP, Maskew M, Macphail AP, Long L, Brennan AT, et al. Cohort Profile: The Themba Lethu Clinical Cohort, Johannesburg, South Africa. International Journal of Epidemiology. 2012 Mar 20; [PMC free article] [PubMed] [Google Scholar]
28. National Department of Health, Republic of South Africa. National Antiretroviral Treatment Guidelines. Jacana Publishers; 2004. [Google Scholar]
29. National Department of Health, Republic of South Africa. The South African Antiretroviral Treatment Guidelines. 2010 [ http://www.doh.gov.za/docs/factsheets/guidelines/art.pdf].
30. Smith CJ, Sabin CA, Youle MS, Kinloch-de Loes S, Lampe FC, et al. Factors influencing increases in CD4 cell counts of HIV-positive persons receiving long-term highly active antiretroviral therapy. J Infect Dis. 2004;190:1860–1868. [PubMed] [Google Scholar]
31. Statistics South Africa. Mortality and causes of death in South Africa, 1997–2003. Findings from death notification. 2005 P0309.3. [PubMed] [Google Scholar]
32. Timaeus I, Dorrington R, Bradshaw D, Nannan N. Mortality trends in South Africa 1985–2000: From apartheid to AIDS. Cape Town: South African Medical Research Council; 2002. [Google Scholar]
34. Rubin DB. Multiple imputation for Nonresponse in Surveys. New York: John Wiley & Sons; 1987. [Google Scholar]
36. Zou G. A Modified Poisson Regression Approach to Prospective Studies with Binary Data. Am. J. Epidemiol. 2004;159(7):702–706. [PubMed] [Google Scholar]
37. Rothman KJ, Greenland S, editors. Modern Epidemiology. Philadelphia: Lippincott Williams & Wilkins; 1998. [Google Scholar]
38. Brinkhof MW, Boulle A, Weigel R, Messou E, Mathers C, et al. for the International epidemiological Databases to Evaluate AIDS (IeDEA) Mortality of HIV-Infected Patients Starting Antiretroviral Therapy in Sub-Saharan Africa: Comparison with HIV-Unrelated Mortality. 2009 Apr 28;6(4) [PMC free article] [PubMed] [Google Scholar]
39. Gange SJ, Mellors JW, Lau B, Detels R, Phair JP, et al. Longitudinal patterns of HIV type 1 RNA among individuals with late disease progression. AIDS Res Hum Retroviruses. 2001;17:1223–1229. [PubMed] [Google Scholar]
40. Lima VD, Fink V, Yip B, Hogg RS, Harrigan PR, et al. Association between HIV-1 RNA level and CD4 cell count among untreated HIV-infected individuals. Am J Public Health. 2009;99:S193–S196. [PMC free article] [PubMed] [Google Scholar]
41. Deeks SG, Barbour JD, Martin JN, Swanson MS, Grant RM. Sustained CD4+ T cell response after virologic failure of protease inhibitor-based regimens in patients with HIV infection. J Infect Dis. 2000;181:946–953. [PubMed] [Google Scholar]
42. Miller V, Staszewski S, Nisius G, Lepri AC, Sabin C, et al. Risk of new AIDS diseases in people on triple therapy. Lancet. 1999 Feb 6;353(9151):463. [PubMed] [Google Scholar]
43. Ledergerber B, Egger M, Opravil M, Telenti M, Hirschel B, et al. Clinical progression and virologic failure on highly active antiretroviral therapy in HIV-1 patients: a prospective cohort study. Lancet. 1999;353:863–868. [PubMed] [Google Scholar]
44. Mocroft A, Katlama C, Johnson AM, et al. AIDS across Europe, 1994–98: the EuroSIDA study. Lancet. 2000;356:291–296. [PubMed] [Google Scholar]
45. Phillips AN, Lampe FC, Smith CJ, Geretti AM, Rodger A, et al. Ongoing changes in HIV RNA levels during untreated HIV infection: implications for CD4 cell count depletion. AIDS. 2010 Jun 19;24(10):1561–1567. [PubMed] [Google Scholar]
46. Brennan AT, Maskew M, Sanne I, Fox MP. The importance of clinic attendance in the first six months on antiretroviral treatment: a retrospective analysis at a large public sector HIV clinic in South Africa. Journal of the International AIDS Society. 2010;13:49. [PMC free article] [PubMed] [Google Scholar]
47. Sethi A, Celentano D, Gange S, Moore R, Gallant J. Association between adherence to antiretroviral therapy and human immunodeficiency virus drug resistance. Clin Infect Dis. 2003;37(8):1112–1118. [PubMed] [Google Scholar]
48. World Health Organization. Management of HIV Infection and Antiretroviral Therapy in Adults and Adolescents. A Clinical Manual. [Accessed 8 April 2008];2007 Available: http://www.searo.who.int/LinkFiles/Publications_Management_HIV_infection_antiretroviral_therapy_adults_adolescents.pdf.
49. ART-LINC of IeDEA Study Group. Keiser O, Tweya H, Boulle A, Braitstein P, Schecter M, et al. Switching to second-line antiretroviral therapy in resource-limited settings: comparison of programmes with and without viral load monitoring. AIDS. 2009 Sep;23(14):1867–1874. [PMC free article] [PubMed] [Google Scholar]
50. Fox MP, Brennan A, Maskew M, MacPhail P, Sanne I. Using vital registration data to update mortality among patients lost to follow-up from ART programmes: evidence from the Themba Lethu Clinic, South Africa. Trop Med Int Health. 2010 Apr;15(4):405–413. [PMC free article] [PubMed] [Google Scholar]
51. Druyts E, Dybul M, Kanters S, Nachega S, Birungi J, Ford N, Thorlund K, Negin J, Lester R, Yaya S, Mills EJ. Male sex and the risk of mortality among individuals enrolled in antiretroviral therapy programs in Africa: a systematic review and meta-analysis. AIDS. 2013;27:417–425. [PubMed] [Google Scholar]