Impact of HIV Type 1 DNA Levels on Spontaneous Disease Progression: A Meta-Analysis

Abstract

Several studies have reported the prognostic strength of HIV-1 DNA with variable results however. The aims of the current study were to estimate more accurately the ability of HIV-1 DNA to predict progression of HIV-1 disease toward acquired immunodeficiency syndrome (AIDS) or death, and to compare the prognostic information obtained by HIV-1 DNA with that derived from plasma HIV-1 RNA. Eligible articles were identified through a comprehensive search of Medline, ISI Web of Science, Scopus, and Google Scholar. The analysis included univariate and bivariate random-effects models. The univariate meta-analysis of six studies involving 1074 participants showed that HIV-1 DNA was a strong predictive marker of AIDS [relative risk (RR): 3.01, 95% confidence interval (CI): 1.88–4.82] and of all-cause mortality (RR: 3.49, 95% CI: 2.06–5.89). The bivariate model using the crude estimates of primary studies indicated that HIV-1 DNA was a significantly better predictor than HIV-1 RNA of either AIDS alone (ratio of RRs=1.47, 95% CI: 1.05–2.07) or of combined (AIDS or death) progression outcomes (ratio of RRs=1.51, 95% CI: 1.11–2.05). HIV-1 DNA is a strong predictor of HIV-1 disease progression. Moreover, there is some evidence that HIV-1 DNA might have better predictive value than plasma HIV-1 RNA.

Introduction

A s new agents aiming to reduce human immunodeficiency virus (HIV) type-1 reservoirs are developing, clinically valuable markers that permit the evaluation of these drugs are needed. Two markers are routinely used in the management of HIV-1-infected patients: the CD4 T cell count and the levels of HIV-1 RNA in plasma (viral load). The percentage or the absolute number of circulating CD4 T cells is an established predictor of HIV-1 disease progression,^1,2 and is used both for the initial assessment of HIV+ individuals and for monitoring therapeutic efficacy.³ The amount of HIV-1 RNA in plasma provides prognostic information independent of CD4 T cell count and constitutes a significant indicator of response to antiretroviral treatment (ARV).^2,4

–7

The introduction of highly active antiretroviral therapy (HAART) has improved the prognosis of HIV-1 infection.⁸ HAART suppresses HIV-1 RNA levels in plasma below the threshold of quantification and restores CD4 T lymphocytes without, however, affecting latently infected cells; there is evidence of persistence of HIV-1 DNA in cellular reservoirs and anatomical sanctuary compartments,^9

–13 thus making the elimination of virus extremely difficult.

Assays for HIV-1 DNA quantification have been developed to explore the pathogenetic role of HIV-1 reservoirs and to gain further insight into HIV-1 DNA prognostic value with variable results however.¹⁴ Therefore, the aim of the current meta-analysis was to estimate more precisely the ability of HIV-1 DNA to predict progression of HIV-1 disease toward acquired immunodeficiency syndrome (AIDS) or death by synthesizing previous research evidence, and to compare the prognostic information obtained by HIV-1 DNA with that derived from the commonly used plasma HIV-1 RNA.

Materials and Methods

Eligibility criteria, data sources, and search strategy

The current meta-analysis was conducted according to the PRISMA statement. The methodology was prespecified and documented in a protocol. Prospective or retrospective cohort and clinical studies, without language restrictions, were included in the analysis if (1) they had examined the association between the levels of HIV-1 DNA in peripheral blood mononuclear cells (PBMCs) and outcomes of HIV-1 disease, such as the emergence of severe clinical events (mainly AIDS) or all-cause mortality in HIV-1 infected patients of any age who had not received HAART (though subjects might had received ARV), and (2) they had reported an estimate of relative risk (RR) in the form of risk ratio, incidence rate ratio, or hazard ratio (HR), or they contained sufficient data to perform the calculation.

Relevant studies were identified by searching Medline, ISI Web of Science, Scopus, and Google Scholar using a combination of the following terms: “HIV-1 DNA,” “disease progression,” “prognostic marker,” and “HIV disease progression” (last search: October 2010). After an initial screening of titles and abstracts, only relevant records remained. Subsequently, full text articles were critically appraised for eligibility and their reference lists were manually scanned to identify further studies for inclusion. Work appeared in conference proceedings or as a short abstract was considered with a view to fully reflecting the existing evidential base.¹⁵

Data extraction and quality assessment

The literature search, the data extraction, and the quality assessment were undertaken by two independent authors (C.T. and G.N.), who resolved discrepancies by consensus. The following information was retrieved from each study and recorded in a data sheet, if available: (1) first author's name, journal, year of publication, and geographic setting in which the study was conducted, (2) the number of participants and the length of follow-up, (3) the method used to quantify HIV-1 DNA in PBMCs, (4) the period of time elapsed between HIV-1 seroconversion and sample collection, (5) the provided crude and adjusted RRs corresponding to the predictive value of HIV-1 DNA and HIV-1 RNA for assessing the risk of progression to severe clinical events or death, and (6) the average characteristics of the study participants. Study quality was judged by the clear description of the adopted eligibility criteria, the completeness of follow-up, and the proper conduct of the statistical analysis. Exclusions strictly because a study was perceived to be susceptible to bias were avoided.¹⁶ However, influential studies were determined by checking the effect of removing an individual study at a time on the overall significance of the summary estimate or on the heterogeneity statistic.

Statistical analysis

Summary estimates of RR along with their 95% confidence intervals (CIs) were computed fitting the conventional univariate random-effects method.¹⁷ The between-study heterogeneity was evaluated using the Cochran's Q statistic¹⁸ and the inconsistency of effects was measured with the I-squared index.¹⁹ Possible publication bias was investigated utilizing the tests of Begg²⁰ and Egger.²¹ The Shapiro–Wilk test was applied to assess the normal distribution of the logarithms of RRs.

In case of multiple parameters, a univariate meta-analytic methodology is a rather simplistic approach of analysis. A more sophisticated method would be to consider simultaneously the two correlated outcomes (i.e., in our study, the logarithms of the RRs for HIV-1 DNA and HIV-1 RNA) in a bivariate process.²² In many types of multivariate meta-analysis, the within-studies correlation needs to be calculated from individual data,²³ since ignoring or approximating the correlation leads to biased estimates of the variance of the overall effect.^24,25 However, individual data are often unavailable. Therefore, a recently proposed bivariate approach was implemented, which requires as input only the two correlated outcomes and computes a single parameter for the overall correlation.²⁶ The bivariate model is necessary not only because it produces more reliable estimates, but also because it provides the means to compare formally the two jointly synthesized outcomes. After a bivariate model was fitted, by using the multivariate delta method, the logarithm of the ratio of the two RRs (for HIV-1 DNA and HIV-1 RNA) along with its 95% CI was calculated.

Analyses were conducted in the statistical package Stata 10 (Stata Corporation, College Station, Texas). With the exception of heterogeneity statistics (significance was declared if p<0.10), results were considered significant if the corresponding p was less than 0.05. All p values were two tailed.

Results

Description of primary studies

The literature search (Fig. 1) resulted in six studies that comprised 1074 subjects (Table 1) and examined the predictive ability of HIV-1 DNA in relation to clinical progression (mainly AIDS) or death in HIV-1-infected patients.^27

–32 Among them, three studies evaluated the role of HIV-1 DNA in determining the risk of progression to both outcomes,^27,28,30 two studies assessed only one outcome, either AIDS²⁹ or death,³² while two studies used a combined (severe clinical event or death) endpoint.^31,32 One research team²⁷ recruited two groups of patients (the first group provided samples within 6 months of primary infection, whereas specimens from the second group were obtained 6–24 months after primary infection) and performed separate analyses resulting, thus, in two estimates of RR in the meta-analysis. The PBMC isolation was conducted soon after infection in four studies,^27,29
–31 while in one study²⁸ the median time between HIV-1 antibody seroconversion and sample collection was 6.7 years. The duration of follow-up ranged between 24 months³¹ and 16 years.²⁸ All investigators employed a Cox regression model in the statistical analysis and the provided HRs corresponded to one unit increase in the (base 10) logarithm (log₁₀) of HIV-1 DNA copies per 10⁶ PBMCs^27,28,31 or in the log₁₀ copies/μg of DNA in PBMCs.³²

FIG. 1.

Identification process for eligible studies concerning the ability of HIV-1 DNA to predict HIV-1 disease progression. HIV, human immunodeficiency virus; RR, relative risk; AIDS, acquired immunodeficiency syndrome; HAART, highly active antiretroviral treatment.

Table 1.

Summary of Eligible Studies Evaluating the Ability of HIV-1 DNA to Predict HIV-1 Disease Progression

Source	N^a	Enrollment	Age (at infection or baseline)	Follow-up (median)	Sample collection after seroconversion (median)	CD4 (cells/mm ³ ) (median)	HIV-1 RNA (log ₁₀ copies/ml) (median)	HIV-1 DNA (log ₁₀ copies/10 ⁶ PMBCs ^b ) (median)
Rouzioux 2005	112	1988–1996	27 years (median)	75 months	3.80 months	594	4.22	2.85
Rouzioux 2005	271	1988–1996	28 years (median)	84 months	11 months	534	4.01	2.86
Kostrikis 2002	127	NA^c	30.40 years (mean)	16 years	6.70 years (median)	320	4.25	2.65
Katzenstein 2002	90	<1989	NA	Up to 2001	Early infection (exact time not reported)	NA	NA	2.70
Minga 2008	200	1997–2006	29 years (median)	30 months	9 months	471	4.60	3.00
Goujard 2006	163	1996–2004	33 years (median)	24 months	68 days	628	4.60	3.20
Tierney 2003	111	NA	36 years (median)	NA	NA	325	4.35	2.08 (log₁₀ copies/μg DNA)

N, sample size.

PBMCs, peripheral blood mononuclear cells.

NA, not available.

In one study, the patients were divided in two groups based on the HIV-1 DNA measurement and the authors assessed the effect of HIV-1 DNA levels above 3 log₁₀ copies per 10⁶ PBMCs on the risk of clinical progression.³⁰ Katzenstein and co-workers created three categories in the statistical analysis (≤100, 500, and ≥2500 copies per 10⁶ PBMCs) having patients with HIV-1 DNA levels below 100 copies per 10⁶ PBMCs as the reference group.²⁹ The HR corresponding to the 500-copies category was used in the main analysis while the effect of the other reported estimate was explored in sensitivity analyses. In all studies, the multivariable analyses of HIV-1 DNA association with HIV-1 disease outcomes systematically explored the confounding effect of CD4 T cell counts and HIV-1 RNA level. Further adjustment for other variables differed across studies. Study-level characteristics including the median age of patients, the period of enrollment, and the baseline values of CD4 T cell counts, HIV-1 RNA, and HIV-1 DNA levels are presented in Table 1.

Univariate meta-analysis

As depicted in Fig. 2, HIV-1 DNA represented a strong predictive marker of severe clinical events (RR: 3.01, 95% CI: 1.88–4.82) and of all-cause mortality (RR: 3.49, 95% CI: 2.06–5.89). HIV-1 DNA remained, although with diminished magnitude, a potent predictor of disease progression when adjusted estimates were employed (RR for severe clinical events: 2.31, 95% CI: 1.48–3.59; RR for all-cause mortality: 2.35, 95% CI: 1.64–3.36). The logarithms of RRs were normally distributed according to the Shapiro–Wilk test. As illustrated in Fig. 2, there was evidence for heterogeneity (I-squared values were between 50% and 70%) in the meta-analysis of crude estimates. On the other hand, when adjusted RRs were used, the estimated heterogeneity decreased considerably. In most cases, formal statistical tests argued in favor of the absence of publication bias.

FIG. 2.

The results of univariate random-effects meta-analysis concerning the ability of HIV-1 DNA to predict HIV-1 disease progression. RR, relative risk; CI, confidence interval; AIDS, acquired immunodeficiency syndrome; HIV, human immunodeficiency virus.

Comparison of HIV-1 DNA and HIV-1 RNA predictive ability using a bivariate meta-analysis

The results of the bivariate meta-analyses are provided in Table 2, and, consistent with the univariate approaches, strongly supported the prognostic ability of HIV-1 DNA. Moreover, the multivariate techniques permitted the direct comparison of the ability of HIV-1 DNA and HIV-1 RNA levels to predict disease progression. The summary RR for each unit increase in HIV-1 DNA level was higher than for each unit increase in HIV-1 RNA level (8–13% higher in prediction of all-cause mortality to almost 50% when the outcome of the analysis was defined by the occurrence of a severe clinical event), a finding common in all models albeit being significant in only some of them. The bivariate model using the crude estimates reported in the primary studies indicated that HIV-1 DNA was a significantly better predictor than plasma viral load. More specifically, in the synthesis of studies that used the occurrence of severe clinical events as the outcome of the analysis, the ratio of RRs was 1.47 with 95% CI: 1.05–2.07. In the meta-analysis of primary studies that used as an endpoint either the occurrence of a severe clinical event or a combined measure (severe clinical events or death), the ratio of RRs was 1.51 with 95% CI: 1.11–2.05. The results, however, were insignificant in the meta-analysis of the adjusted estimates provided in primary research.

Table 2.

Results of the Bivariate Meta-Analyses Using Crude and Adjusted Estimates Reported in Primary Studies That Examined the Association Between HIV-1 DNA Levels and HIV-1 Disease Progression

Outcome	Summary RR ^a for HIV-1 DNA (95% confidence interval)	Summary RR for HIV-1 RNA (95% confidence interval)	Ratio of RRs (HIV-1 DNA vs. HIV-1 RNA) (95% confidence interval)
Severe clinical events	3.00 (1.84–4.89) (cr)^b	2.04 (1.31–3.15) (cr)	1.47 (1.05–2.07) (cr)
	2.30 (1.38–3.84) (ad)^c	1.75 (1.16–2.64) (ad)	1.31 (0.74–2.32) (ad)
Severe clinical events/combined outcome^d	3.09 (2.17–4.40) (cr)	2.04 (1.49–2.80) (cr)	1.51 (1.11–2.05) (cr)
	2.38 (1.64–3.44) (ad)	1.53 (1.04–2.24) (ad)	1.55 (0.96–2.51) (ad)
All-cause mortality	3.48 (2.00–6.06) (cr)	3.08 (2.40–3.94) (cr)	1.13 (0.55–2.34) (cr)
	2.55 (1.42–4.56) (ad)	2.35 (1.57–3.52) (ad)	1.08 (0.50–2.37) (ad)
All-cause mortality/combined outcome	3.50 (2.29–5.35) (cr)	2.59 (1.89–3.56) (cr)	1.35 (0.74–2.46) (cr)
	2.44 (1.58–3.77) (ad)	1.86 (1.11–3.12) (ad)	1.31 (0.61–2.79) (ad)

RR, relative risk.

Crude estimates.

Adjusted estimates.

Combined outcome: severe clinical event (mostly indicative of acquired immunodeficiency syndrome) or death.

Discussion

Many clinical and laboratory markers have been used to estimate the prognosis in HIV-1 infection with CD4 T cell counts and plasma HIV-1 RNA most convincingly related to the risk of AIDS or death. The current meta-analysis demonstrates that HIV-1 DNA has considerable merit in the assessment of the risk for HIV-1 disease progression. In particular, higher HIV-1 DNA levels strongly predicted poorer outcomes with estimated RRs for AIDS development or death approximating 2.3–3.5. Moreover, there was some evidence, although not statistically overwhelming, that compared to the commonly used plasma viral load, HIV-1 DNA has a stronger independent association with the emergence of clinical events indicating severe immunosuppression.

Despite the dramatic reduction of HIV-1 RNA, often below the assay limit of quantification, induced by HAART, latently infected cells persist as an important reservoir of the virus and represent a major barrier to HIV-1 eradication.^33
–35 At the time of moving forward to test new drugs aiming to reduce reservoirs, the assessment of previous research points to an important conclusion: viral reservoirs remain even in patients receiving HAART.^12,13,36 There are many cellular and anatomical reservoirs for HIV-1 that may contribute to long-term persistence of the virus including resting memory CD4 T cells, cells of the monocyte-macrophage lineage, and hematopoietic progenitor cells.³³ As opposed to HIV-1 RNA that indicates recent viral replication, HIV-1 DNA levels reflect the total cellular stock of HIV-1 including latently infected cells. In other words, the HIV-1 DNA level represents the ability for viral production that may differ among patients.^27,28 The HIV-1 DNA level in infected cells could be considered as the driving force of HIV-1 disease, which proceeds at different speeds according to the number of CD4 T cells count and to the viral replicative capacity represented by HIV-1 RNA measurement.²⁷

Although not the absolute means of judging performance, the selected multivariate meta-analytic approach permitted the direct comparison of the predictive abilities of HIV-1 DNA and HIV-1 RNA levels. In other words, despite its limitations, this method comprised a straightforward statistical manner to quantify the differences between the prognostic values of HIV-1 DNA and HIV-1 RNA levels. In some of these analyses, the HIV-1 DNA level proved to be a better predictive factor than HIV-1 RNA.

Previous research has shown that pre-HAART levels of cellular HIV-1 DNA predict the long-term success of HAART and, especially, the patients who will rebound after an initial response.³⁷ Quite interestingly, in that study, HIV-1 RNA was not a significant predictor.³⁷ Another study has also shown that PBMC-associated HIV-1 DNA was an effective prognostic marker for long-term virologic failure in asymptomatic, chronically infected, HIV-1-seropositive patients and HIV-1 DNA was the strongest variable associated with the virologic outcome.³⁸ The improved predictability of HIV-1 DNA might be attributed to the different kinetics of HIV-1 DNA and HIV-1 RNA in the natural history of HIV-1 disease. The cellular HIV-1 reservoir is established early after infection.^11,39 HIV-1 DNA levels in PBMCs, a marker associated with the viral reservoir, are less widely dispersed than HIV-1 RNA levels in primary infections³⁹ and remain relatively stable during the natural course of HIV-1 disease.^28,40 Even after the initiation of suppressive ARV, PBMC HIV-1 DNA exhibits an initial biphasic decay lasting 3 years, mostly because of a disproportionate loss of nonintegrated HIV-1 genomes, and no significant reduction thereafter.^41,42 Moreover, after long-term suppressive antiretroviral treatment, PBMC HIV-1 DNA has the ability to describe the intensity of viral replication and of immunological status reached before starting treatment.⁴³ HIV-1 DNA was predictive of disease progression irrespective of the adopted method for reporting HIV-1 DNA in blood (copies per 10⁶ PBMCs, copies per 10⁶ CD4 T cell, or copies/ml of whole blood).⁴⁴ Finally, it should be noted that the quantitative HIV-1 DNA measurement is technically easy to perform on PBMC pellet, as well as on whole blood using HIV-DNA real-time PCR.⁴⁵

On the other hand, soon after infection, the plasma viral load exhibits high variability, which might undermine its predictive value before reaching the plateau phase.^31,39,46 Even after reaching the set point, rather than a long-term steady-state viremia, data suggest that in most individuals, a short plateau phase is followed by a progressive, long-lasting increase of HIV-1 RNA levels.⁴⁶ In addition, it should be noted that even though a single HIV-1 RNA measurement, in accord with the central role of viral replication in AIDS development, is a strong predictor of times to AIDS or death, plasma HIV-1 RNA levels have in fact been found to account for only a small proportion of the variability in the rate of CD4 cell decline in chronic, untreated HIV-1 infection.⁴⁷ Moreover, other factors, apart from HIV-1 DNA, such as programmed cell death 1 (PD-1), seem to perform better than plasma HIV-1 RNA in terms of prediction for CD4 loss rates.⁴⁸ The pathogenesis of HIV-1-related immune deficiency is probably complex and further research on stronger predictors or new mechanisms of HIV-1 disease progression would be valuable.

Interestingly, HIV-1 DNA was a stronger predictor of the emergence of severe clinical events, often indicative of AIDS, than of overall mortality. This observation can be supported, for instance, by the direct correlation of PBMC HIV-1 DNA to HIV-associated dementia (HAD). In a small cross-sectional study, the median level of HIV-1 DNA among HAD patients was almost 20 times higher than the corresponding level in patients with normal cognition.⁴⁹ The association between HAD and circulating HIV-1 DNA remained significant in patients with undetectable plasma viral load. Moreover, in a longitudinal study, the level of HIV-1 DNA was persistently higher over time in patients experiencing HAD, despite the administration of HAART, compared with normal individuals.⁵⁰ It seems that infected monocytes transmigrate the blood–brain barrier, an event that probably triggers gene expression of otherwise quiescent virus, resulting in viral seeding and inflammation of the central nervous system (CNS), and thus these cells actively contribute to CNS injury.⁵⁰ Apart from this potential, direct connection of HIV-1 DNA with AIDS, the attenuation of RRs observed in mortality analyses can simply be attributed to the dilution of the existed associations by the inclusion in the analyses of fatal events unrelated to HIV-1 infection.

The statistical pooling of studies has its limitations. First, the meta-analysis cannot overcome the drawbacks of primary research. Second, performing a literature-based meta-analysis is less accurate and complete than the individual-level approach. Nevertheless, the former is a practical and credible alternative. Third, the number of studies included in the current synthesis was small, which limits the ability to detect statistically significant effects. Moreover, although publication bias was absent as suggested by formal statistical procedures, the implementation of these tests in case of meta-analyses of few studies is open to question. Fourth, there was evidence of heterogeneity, reflecting dissimilarities between the eligible studies. For instance, regarding the approaches to laboratory methods on HIV-1 DNA quantification, some studies were based on commercially available assays while others used in-house primers. Within the context of a literature-based meta-analysis, it was difficult to explore the exact contribution of laboratory variations to the overall heterogeneity because the laboratory techniques were described in detail in some studies but only in a few sentences in others. It seems that the standardization, at least, of laboratory methodology and of outcomes and variables examined in statistical models is necessary in order to reduce extended variability in future research. Finally, due to a lack of data in primary research, we were incapable of comparing the prognostic ability of HIV-1 DNA with other potential markers of HIV-1 disease progression such as the expression of CD38 on CD8⁺ T cells or the immunopathogenic factor PD-1.⁴⁸

In conclusion, this meta-analysis was performed to determine the strength of HIV-1 DNA as a predictor of clinical progression and/or death. Synthesizing published studies, the analysis showed that the HIV-1 DNA level is a useful marker in prognosis and, interestingly, may sometimes be more reliably predictive of outcomes than the clinically standard HIV-1 RNA quantitation. This meta-analysis reinforces the value of HIV-1 DNA as a prognostic indicator, especially in an era of research focusing on reducing the HIV-1 reservoir or on identifying the optimal time to start HAART.^14,27 Our data call for larger, multicenter, prospective, or retrospective studies, or for an individual-level meta-analysis synthesizing more information that has not been analyzed or published yet to ascertain the place of HIV-1 DNA measurement in the natural course of HIV-1 infection and perhaps, more importantly, in terms of therapeutic response.

Footnotes

Author Disclosure Statement

No competing financial interests exist.

References

Fahey

, Taylor

, Detels

et al. The prognostic value of cellular and serologic markers in infection with human immunodeficiency virus type 1. N Engl J Med, 1990; 322:166–172.

Mellors

, Muñoz

, Giorgi

et al. Plasma viral load and CD4+ lymphocytes as prognostic markers of HIV-1 infection. Ann Intern Med, 1997; 126:946–954.

Panel on Antiretroviral Guidelines for Adults and Adolescents: Guidelines for the use of antiretroviral agents in HIV-1-infected adults and adolescents. Department of Health and Human Services, 2009; 1–161. http://www.aidsinfo.nih.gov/ContentFiles/AdultandAdolescentGL.pdf. 2010 December 17.

Katzenstein

, Pedersen

, Nielsen

, Lundgren

, Jakobsen

, Gerstoft

. Longitudinal serum HIV RNA quantification: Correlation to viral phenotype at seroconversion and clinical outcome. AIDS, 1996; 10:167–173.

Mellors

, Kingsley

, Rinaldo

et al. Quantitation of HIV-1 RNA in plasma predicts outcome after seroconversion. Ann Intern Med, 1995; 122:573–579.

Mellors

, Rinaldo

Jr , Gupta

, White

, Todd

, Kingsley

. Prognosis in HIV-1 infection predicted by the quantity of virus in plasma. Science, 1996; 272:1167–1170.

Hughes

, Johnson

, Hirsch

et al. Monitoring plasma HIV-1 RNA levels in addition to CD4+ lymphocyte count improves assessment of antiretroviral therapeutic response. ACTG 241 Protocol Virology Substudy Team. Ann Intern Med, 1997; 126:929–938.

Mocroft

, Vella

, Benfield

et al. Changing patterns of mortality across Europe in patients infected with HIV-1. EuroSIDA Study Group. Lancet, 1998; 352:1725–1730.

Wong

, Hezareh

, Günthard

et al. Recovery of replication-competent HIV despite prolonged suppression of plasma viremia. Science, 1997; 278:1291–1295.

10.

Zhang

, Dornadula

, Beumont

et al. Human immunodeficiency virus type 1 in the semen of men receiving highly active antiretroviral therapy. N Engl J Med, 1998; 339:1803–1809.

11.

Chun

, Engel

, Berrey

, Shea

, Corey

, Fauci

. Early establishment of a pool of latently infected, resting CD4(+) T cells during primary HIV-1 infection. Proc Natl Acad Sci USA, 1998; 95:8869–8873.

12.

Finzi

, Blankson

, Siliciano

et al. Latent infection of CD4+ T cells provides a mechanism for lifelong persistence of HIV-1, even in patients on effective combination therapy. Nat Med, 1999; 5:512–517.

13.

Chun

, Nickle

, Justement

et al. Persistence of HIV in gut-associated lymphoid tissue despite long-term antiretroviral therapy. J Infect Dis, 2008; 197:714–720.

14.

Beloukas

, Paraskevis

, Psichogiou

, Hatzakis

. The role of HIV-1 DNA as an additional marker of HIV-1 infection. Curr HIV Res, 2009; 7:255–265.

15.

Conn

, Valentine

, Cooper

, Rantz

. Grey literature in meta-analyses. Nurs Res, 2003; 52:256–261.

16.

Stroup

, Berlin

, Morton

et al. Meta-analysis of observational studies in epidemiology: A proposal for reporting. Meta-analysis Of Observational Studies in Epidemiology (MOOSE) group. JAMA, 2000; 283:2008–2012.

17.

DerSimonian

, Laird

. Meta-analysis in clinical trials. Controlled Clin Trials, 1986; 7:177–188.

18.

Petitti

. Meta-Analysis, Decision Analysis, and Cost-Effectiveness Analysis. Oxford University Press: New York, 2000.

19.

Higgins

, Thompson

, Deeks

, Altman

. Measuring inconsistency in meta-analyses. BMJ, 2003; 327:557–560.

20.

Begg

, Mazumdar

. Operating characteristics of a rank correlation test for publication bias. Biometrics, 1994; 50:1088–10101.

21.

Egger

, Davey Smith

, Schneider

, Minder

. Bias in meta-analysis detected by a simple, graphical test. BMJ, 1997; 315:629–634.

22.

van Houwelingen

, Arends

, Stijnen

. Advanced methods in meta-analysis: Multivariate approach and meta-regression. Stat Med, 2002; 21:589–624.

23.

Berkey

, Hoaglin

, Antczak-Bouckoms

, Mosteller

, Colditz

. Meta-analysis of multiple outcomes by regression with random effects. Stat Med, 1998; 17:2537–2550.

24.

Riley

, Abrams

, Lambert

, Sutton

, Thompson

. An evaluation of bivariate random-effects meta-analysis for the joint synthesis of two correlated outcomes. Stat Med, 2007; 26:78–97.

25.

Riley

, Abrams

, Sutton

, Lambert

, Thompson

. Bivariate random-effects meta-analysis and the estimation of between-study correlation. BMC Med Res Methodol, 2007; 7:3.

26.

Riley

, Thompson

, Abrams

. An alternative model for bivariate random-effects meta-analysis when the within-study correlations are unknown. Biostatistics, 2008; 9:172–186.

27.

Rouzioux

, Hubert

, Burgard

et al. for the SEROCO Cohort Study Group. Early levels of HIV-1 DNA in peripheral blood mononuclear cells are predictive of disease progression independently of HIV-1 RNA levels and CD4+ T cell counts. J Infect Dis, 2005; 192:46–55.

28.

Kostrikis

, Touloumi

, Karanicolas

et al. for the Multicenter Hemophilia Cohort Study Group. Quantitation of human immunodeficiency virus type 1 DNA forms with the second template switch in peripheral blood cells predicts disease progression independently of plasma RNA load. J Virol, 2002; 76:10099–10108.

29.

Katzenstein

, Oliveri

, Benfield

, Eugen-Olsen

, Nielsen

, Gerstoft

. for the Copenhagen AIDS Cohort Study Group. Cell-associated HIV DNA measured early during infection has prognostic value independent of serum HIV RNA measured concomitantly. Scand J Infect Dis, 2002; 34:529–533.

30.

Minga

, Anglaret

, d' Aquin Toni

et al. HIV-1 DNA in peripheral blood mononuclear cells is strongly associated with HIV-1 disease progression in recently infected West African adults. J Acquir Immune Defic Syndr, 2008; 48:350–354.

31.

Goujard

, Bonarek

, Meyer

et al. for the Agence Nationale de Recherche sur le Sida PRIMO Study Group. CD4 cell count and HIV DNA level are independent predictors of disease progression after primary HIV type 1 infection in untreated patients. Clin Infect Dis, 2006; 42:709–715.

32.

Tierney

, Lathey

, Christopherson

et al. Prognostic value of baseline human immunodeficiency virus type 1 DNA measurement for disease progression in patients receiving nucleoside therapy. J Infect Dis, 2003; 187:144–148.

33.

Alexaki

, Liu

, Wigdahl

. Cellular reservoirs of HIV-1 and their role in viral persistence. Curr HIV Res, 2008; 6:388–400.

34.

Bagasra

. A unified concept of HIV latency. Expert Opin Biol Ther, 2006; 6:1135–1149.

35.

Trono

, Van Lint

, Rouzioux

et al. HIV persistence and the prospect of long-term drug-free remissions for HIV-infected individuals. Science, 2010; 329:174–180.

36.

Chun

, Stuyver

, Mizell

et al. Presence of an inducible HIV-1 latent reservoir during highly active antiretroviral therapy. Proc Natl Acad Sci USA, 1997; 94:13193–13197.

37.

Hatzakis

, Touloumi

, Pantazis

et al. Cellular HIV-1 DNA load predicts HIV-RNA rebound and the outcome of highly active antiretroviral therapy. AIDS, 2004; 18:2261–2267.

38.

Rodríguez-Sáinz

, Ramos

, Valor

et al. Prognostic value of peripheral blood mononuclear cell-associated HIV-1 DNA for virological outcome in asymptomatic HIV-1 chronic infection. J Clin Virol, 2010; 48:168–172.

39.

Ghosn

, Deveau

, Chaix

et al. for the ANRS PRIMO Cohort. Despite being highly diverse, immunovirological status strongly correlates with clinical symptoms during primary HIV-1 infection: A cross-sectional study based on 674 patients enrolled in the ANRS CO 06 PRIMO cohort. J Antimicrob Chemother, 2010; 65:741–748.

40.

Cone

, Gowland

, Opravil

, Grob

, Ledergerber

. Levels of HIV-infected peripheral blood cells remain stable throughout the natural history of HIV-1 infection. Swiss HIV Cohort Study. AIDS, 1998; 12:2253–2260.

41.

Viard

, Burgard

, Hubert

et al. Impact of 5 years of maximally successful highly active antiretroviral therapy on CD4 cell count and HIV-1 DNA level. AIDS, 2004; 18:45–49.

42.

Koelsch

, Liu

, Haubrich

et al. Dynamics of total, linear nonintegrated, and integrated HIV-1 DNA in vivo and in vitro. J Infect Dis, 2008; 197:411–419.

43.

Burgard

, Boufassa

, Viard

et al. for the ANRS AC11 Working Group. Factors influencing peripheral blood mononuclear cell-associated HIV-1 DNA level after long-term suppressive antiretroviral therapy in 236 patients. AIDS, 2009; 23:2165–2171.

44.

Avettand-Fènoël

, Boufassa

, Galimand

, Meyer

, Rouzioux

. for the ANRS SEROCO Cohort Study Group. HIV-1 DNA for the measurement of the HIV reservoir is predictive of disease progression in seroconverters whatever the mode of result expression is. J Clin Virol, 2008; 42:399–404.

45.

Avettand-Fènoël

, Chaix

, Blanche

et al.

LTR real-time PCR for HIV-1 DNA quantitation in blood cells for early diagnosis in infants born to seropositive mothers treated in HAART area (ANRS CO 01)

J Med Virol, 2009; 81:217–223.

46.

Hubert

, Burgard

, Dussaix

et al. Natural history of serum HIV-1 RNA levels in 330 patients with a known date of infection. The SEROCO Study Group. AIDS, 2000; 14:123–131.

47.

Rodríguez

, Sethi

, Cheruvu

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.

48.

Holm

, Pettersen

, Kvale

. PD-1 predicts CD4 loss rate in chronic HIV-1 infection better than HIV RNA and CD38 but not in cryopreserved samples. Curr HIV Res, 2008; 6:49–58.

49.

Shiramizu

, Gartner

, Williams

et al. Circulating proviral HIV DNA and HIV-associated dementia. AIDS, 2005; 19:45–52.

50.

Valcour

, Shiramizu

, Sithinamsuwan

et al. HIV DNA and cognition in a Thai longitudinal HAART initiation cohort: The SEARCH 001 Cohort Study. Neurology, 2009; 72:992–998.