Inflammatory Biomarkers and Mortality Risk Among HIV-Suppressed Men: A Multisite Prospective Cohort Study

Serum concentrations of multiple inflammatory biomarkers strongly predict long-term mortality risk in human immunodeficiency virus–infected men receiving antiretroviral therapy with confirmed viral suppression. Results indicate several underlying inflammatory factors, 2 of which independently predict mortality risk.

Abstract

Background. Human immunodeficiency virus (HIV)–induced inflammation and immune activation persist after initiation of combination antiretroviral therapy (cART) and HIV suppression and may contribute to mortality risks that exceed those in HIV-uninfected populations, though associations are unclear.

Methods. In the prospective Multicenter AIDS Cohort Study, comprising men who have sex with men from Baltimore, Chicago, Los Angeles, and Pittsburgh, concentrations of 24 biomarkers of inflammation and immune activation were measured in stored serum from HIV-positive men obtained after cART-induced HIV suppression between 1996 and 2009. The outcome was nonaccidental death, with follow-up until 2014. We used Cox proportional hazards models to test whether biomarker concentrations predict time from HIV suppression to death and adjusted for multiple tests. Exploratory factor analysis (EFA) was employed to identify groupings of biomarkers that predict mortality risk.

Results. Of 670 men followed up from HIV suppression, 54 died by the end of 2013. After adjustment for age, CD4⁺ cell count, hepatitis B or C virus infection, and smoking, concentrations in the highest quartile of 4 biomarkers were significantly associated with mortality risk after controlling the false discovery rate at 5%: interleukin (IL) 6 (hazard ratio, 3.54; 95% confidence interval, 2.06–6.10), soluble IL 2Rα (3.29, 1.85–5.85), soluble CD14 (2.67, 1.55–4.61), and chemokine (CXC motif) ligand 13 (CXCL13; 2.26; 1.29–3.95). EFA yielded 2 biomarker groupings that were independent predictors of mortality risk.

Conclusions. Despite having undetectable HIV RNA levels during cART, men with higher concentrations of several biomarkers (particularly IL 6, soluble IL 2Rα, soluble CD14, and CXCL13) had higher hazards of long-term mortality. Correlations observed among biomarker concentrations may represent underlying inflammatory processes that contribute to mortality risk.

Human immunodeficiency virus (HIV)–infected individuals receiving combination antiretroviral therapy (cART) still face higher morbidity and mortality risks than HIV-uninfected individuals [1]. Notably, HIV infection increases the risk of certain serious non–AIDS-related diseases and death from some non-AIDS causes [2–6].

In addition to immune deficiency, HIV infection causes chronic immune activation and dysregulation of inflammatory processes [7]. Concentrations of serological biomarkers of inflammation and immune activation are higher in individuals with untreated HIV infection than in HIV-uninfected individuals [8]. Suppressive cART diminishes inflammation, but not all biomarkers return to the lower concentrations observed in HIV-uninfected individuals [9, 10]. Residual chronic inflammation and immune activation may increase mortality risks among HIV-infected individuals receiving cART. If so, serological biomarkers may provide prognostic value and help define harmful processes that could be targets of therapy.

Inflammatory biomarkers, particularly interleukin (IL) 6 and C-reactive protein (CRP), predict mortality and serious morbidity risks in HIV-uninfected populations [11, 12]. Most studies examining associations between inflammation and mortality risk among HIV-infected individuals have examined biomarker concentrations before cART initiation [13–16] or irrespective of cART use [17, 18]. Two studies restricted to individuals who became virologically suppressed on cART showed associations between death and biomarker concentrations immediately before death. Hunt et al [19] found very strong relationships between death and inflammatory markers measured within 1 year of death, and Boulware et al [20] linked CRP and IL-6 levels to incident AIDS or death in a case-control study that selected for outcomes within 1 year after cART initiation [20]. A case-control study examining 6 inflammatory biomarkers among HIV-suppressed individuals found that concentrations of IL-6, soluble tumor necrosis factor (TNF) receptor 1, soluble TNF receptor 2 (sTNFR2), D-dimer, and soluble CD14 (sCD14) before and after HIV suppression were associated with higher odds of non–AIDS-related disease or death, and its authors called for long-term prospective cohort studies to further investigate these associations [21].

It is still poorly understood whether biomarkers of inflammation and immune activation may have longer-term prognostic value among the broader population of individuals on treatment when measured soon after HIV suppression. Most of the few studies to date have been limited by case-control study designs and/or few numbers of biomarkers measured. We tested whether 24 biomarkers measured after HIV suppression could predict long-term mortality risk in a large cohort of men with a wide spectrum of comorbidity risk and immunologic status, followed up for up to 18 years.

METHODS

Study Cohort

The Multicenter AIDS Cohort Study (MACS; http://aidscohortstudy.org/) is an ongoing prospective cohort study of HIV infection in men who have sex with men from 4 sites (Baltimore, Maryland/Washington, DC; Chicago, Illinois; Los Angeles, California; and Pittsburgh, Pennsylvania). Follow-up began in 1984, with later waves of recruitment. Details of the MACS have been described elsewhere [22]; briefly, participants are followed up at semiannual study visits with standardized interviews, physical examinations, and phlebotomy for concurrent laboratory testing and storage of plasma and serum (at −80°C) and viable peripheral blood mononuclear cells (at −135°C). This study was approved by the institutional review boards of participating institutions.

Biomarkers

Serum samples from MACS participants were chosen for a study of biomarkers of inflammation and immune activation at 1-year intervals for men with a known date of seroconversion, from visits immediately before and after cART initiation (and at 2-year intervals thereafter) for all cART users and from a sample of HIV-uninfected men [10]. Serum concentrations of 24 biomarkers were measured in these samples. The cytokines IL-1β, IL-2, IL-6, IL-10, IL-12p70, TNF-α, granulocyte-macrophage colony-stimulating factor, and interferon γ were measured with the Human Pro Inflammatory 9-Plex Ultra-Sensitive Kit (Meso Scale Discovery [MSD]); and chemokine (CC motif) ligand (CCL) 2, CCL4, CCL11, CCL13, CCL17, chemokine (CXC motif) ligand (CXCL) 10, and IL-8 were measured with the MSD Human Chemokine 7-Plex Ultra-Sensitive Kit, according to the manufacturer's protocols. The MSD platform is a solid-phase electrochemiluminescence-based assay; MSD plates were analyzed with the SECTOR Imager 2400 (MSD).

The markers sCD14, soluble CD27 (sCD27), soluble GP130 (sGP130), soluble IL 2Rα (sIL-2Rα), soluble IL 6R (sIL-6R), sTNFR2, B-cell activating factor (BAFF), and CXCL13 were measured with the Luminex xMAP platform (Luminex), according to the manufacturer's protocols (R&D Systems). A single lot of assay kits was used to eliminate lot-to-lot variability. Serum samples were diluted 1:50. The Luminex platform is a fluorescent bead-based assay; data were collected and analyzed using a BioPlex 200 apparatus and BioPlex Manager software (Bio-Rad). All samples from a given individual were tested on a single plate to minimize variability. Each plate contained samples from both HIV-infected and HIV-uninfected men. One additional marker, CRP, was measured by a reference laboratory (Quest Diagnostics) using a high-sensitivity immunonephelometric assay.

Outcome and Covariate Measurement and Definitions

All-cause death was the outcome of interest. Deaths and their dates were ascertained via death certificates and matching with National Death Index records. When injury or poisoning was reported as the primary cause of death (n = 7), men were right censored at the date of death, because we assumed that residual inflammation/immune activation is unlikely to cause accidental death.

Plasma concentrations of HIV RNA were measured with the Roche Amplicor assay sensitive to 50 copies/mL, and visits with undetectable concentrations were classified as HIV suppressed. CD4⁺ T-cell counts were measured with flow cytometry [23] and dichotomized as ≤200 or >200 cells/µL. Hepatitis infection was defined as chronic infection with either hepatitis B virus (defined by presence of hepatitis B surface antigen) or hepatitis C virus (defined by detectable hepatitis C RNA). Smoking at baseline (yes or no) was defined by self-report. Age at baseline was treated as continuous.

Statistical Analyses

Individual Biomarkers as Predictors of Mortality Risk

Men were followed up from baseline to death or to the administrative censoring date of 1 January 2014. Baseline was the date of HIV suppression, defined as the midpoint between the last cART-exposed visit with detectable HIV RNA and the first visit with HIV RNA <50 copies/mL. Biomarker concentrations were defined at the first cART-exposed visit with suppressed HIV at which biomarker measurements were available. Because no men could be observed precisely at the baseline, all observations were treated using late entry methods. We fit separate Cox proportional hazards models for mortality risk with the highest quartile of each biomarker concentration as the exposure of interest, adjusting for baseline values of age, CD4⁺ cell count, chronic hepatitis infection, and smoking. We selected these possible confounding covariates by first assessing covariate-biomarker relationships, and then adding 1 candidate at a time to Cox models for mortality risk chosen by the greatest divergence between likelihoods of the new model and the more parsimonious model until no likelihood ratio test statistic was significant at P < .05. Because we had 24 biomarkers of interest, we adjusted for multiple tests by controlling the false discovery rate at 5%, using the Benjamini-Hochberg procedure [24].

To provide context for biomarker concentrations, we compared baseline biomarker concentrations in the study population with those among HIV-uninfected MACS participants. To do so, we fit multivariable generalized gamma models (with location parameter allowed to vary but scale and shape parameters held constant) on the inverse of biomarker concentrations [10], treating values below the lower limit of detection as right censored and controlling for age, nonwhite race, smoking, hepatitis C infection, obesity, diabetes, and MACS site. For a multibiomarker survival model, we included markers that were significant in adjusted single-biomarker models after controlling for multiple tests. To address the correlation structure among biomarker concentrations, we used exploratory factor analysis (EFA), described below.

EFA Methods

We hypothesized that patterns of observed biomarker concentrations reflect at least one unobserved immune process that contributes to mortality risk. To identify these processes and test their relationship with mortality risk, we performed EFA. EFA is a data reduction technique assuming underlying factors that give rise to observed correlations among variables; for every factor identified by EFA (corresponding to a hypothesized inflammatory process), an individual was assigned a factor score that was assessed as an independent predictor of mortality risk.

For the purposes of EFA, we log-transformed biomarker concentrations; concentrations below the lower limit of detection were imputed from lognormal models. We used principal component analysis on all 24 markers with parallel analysis to select the number of extracted factors. We performed EFA using iterative principal factor estimation and varimax rotation, keeping the factors orthogonal to one another for the sake of interpretability. We generated factor scores for each individual, then fit Cox models assessing the association of factors with mortality risk. Details of the EFA can be found in the Supplementary Materials.

RESULTS

Characteristics of Study Population

Table 1 summarizes characteristics of the 670 men followed up from HIV suppression, who contributed 6742 person-years to the analysis. The median year of HIV suppression was 2001 (interquartile range [IQR], 1998–2004). Baseline biomarker measurements were taken at a median (IQR) of 0.7 (0.2–1.7) years after HIV suppression, and the median follow-up time was 9.9 (7.0–13.8) years. The median (IQR) age at baseline was 44.1 (39.0–49.5) years, and the median CD4⁺ cell count was 479/µL (332–651/µL). By the end of 2013, 54 men had died (primary causes of death: AIDS in 23, cardiovascular in 9, cancer in 5, gastrointestinal in 3, liver disease in 3, and other/unknown in 9). These men died at a median (IQR) of 5.6 (3.0–9.0) years after HIV suppression, at a median age of 57.0 (52.0–61.6) years.

Table 1.

Characteristics of Study Population at Baseline (n = 670)

Mortality Risk Factor	No. (%)
Nonwhite race	173 (26)
Smoker	201 (30)
Heavy alcohol user (>13 drinks/wk)	30 (4)
Injection drug user	14 (2)
HBV infection	42 (6)
HCV infection	58 (9)
HBV or HCV infection	94 (14)
Overweight (BMI >25)a	283 (43)
Obese (BMI >30)a	73 (11)
Prior ART exposure	148 (23)
Hypertensionb	55 (8)
Diabetesc	111 (17)
Anemiad	180 (27)
Depressive symptomse	173 (26)

Abbreviations: ART, antiretroviral therapy; BMI, body mass index; HBV, hepatitis B virus; HCV, hepatitis C virus.

^a BMI is calculated as the weight in kilograms divided by height in meters squared.

^b Hypertension was defined as systolic blood pressure ≥140 mm Hg or diastolic blood pressure ≥90 mm Hg.

^c Diabetes was defined as hemoglobin A1C ≥6.5% or fasting glucose ≥126 mg/dL.

^d Anemia was defined as hemoglobin level below the fifth percentile.

^e Depressive symptoms were defined as Center for Epidemiologic Studies depression score ≥16.

Supplementary Table 1 displays biomarker distributions at baseline for the study population and for HIV-uninfected MACS men. Supplementary Figure 1 displays covariate-adjusted comparisons; as seen elsewhere [10], concentrations of 13 biomarkers were significantly higher (after adjustment for multiple comparisons) among men with suppressed HIV RNA relative to HIV-uninfected men, and concentrations of 2 markers (granulocyte-macrophage colony-stimulating factor and CCL11) were significantly lower in HIV-suppressed men.

Results From Adjusted Proportional Hazards Models

Figure 1 displays mortality hazard ratios (HRs) for each biomarker from multivariable proportional hazards models adjusted for hepatitis B or C infection, age, low CD4⁺ cell count (≤200/µL), and smoking. Higher concentrations of IL-6 were most strongly associated with mortality risk (HR, 3.54; 95% confidence interval [CI], 2.06–6.10). Three other markers were also statistically significant after adjustment for multiple comparisons: sIL-2Rα (HR, 3.29; 95% CI, 1.85–5.85), sCD14 (2.67; 1.55–4.61), and CXCL13 (2.26; 1.29–3.95). Six additional biomarkers (GP130, sTNFR2, CD27, IL-6R, CRP, and TNF-α) exhibited mortality HRs whose 95% CIs did not include the null, but these were not significant after adjustment for multiple comparisons. Detailed model estimates are provided in Table 2.

Figure 1.

Mortality hazard ratios (HRs) for men with the highest quartile of biomarker concentrations relative to the lower 3 quartiles, displayed on a natural log scale. Models are adjusted for age, CD4⁺ cell count, hepatitis B or C infection, and smoking status. Error bars represent 95% confidence intervals. Red squares denote HRs that are statistically significant after adjustment for multiple tests, with the Benjamini-Hochberg procedure used to control the false discovery rate at 5% [24]; orange squares, HRs significant at P < .05. Abbreviations: BAFF, B-cell activating factor; CCL, chemokine (CC motif) ligand; CRP, C-reactive protein; CXCL, chemokine (CXC motif) ligand; GM-CSF, granulocyte-macrophage colony-stimulating factor; IFN, interferon; IL-1β, interleukin 1β; IL-2, interleukin 2; IL-6, interleukin 6; IL-8, interleukin 8; IL-10, interleukin 10; IL-12p70, interleukin 12p70; sCD14, soluble CD14; sCD27, soluble CD27; sGP130, soluble GP130; sIL-2Rα, soluble interleukin 2Rα; sIL-6R, soluble interleukin 6R; sTNFR2, soluble tumor necrosis factor receptor 2; TNF, tumor necrosis factor.

Table 2.

Results From Adjusted Cox Proportional Hazards Models^a

Biomarker	Highest Quartile of Biomarker Concentrations Relative to Lower 3 Quartiles
	HR	95% LL	95% UL	P Value
BAFF	1.45	0.81	2.58	.21
CCL11	1.10	0.60	2.02	.76
CCL13	1.04	0.53	2.07	.90
CCL17	0.85	0.45	1.60	.61
CCL2	1.43	0.81	2.51	.21
CCL4	0.66	0.33	1.34	.25
CRP	1.82b	1.03	3.20	.04
CXCL10	1.71	0.95	3.09	.07
CXCL13	2.26b,c	1.29	3.95	.005
GM-CSF	0.91	0.49	1.69	.76
IFN-γ	1.21	0.66	2.21	.54
IL-10	1.03	0.57	1.87	.92
IL-12p70	0.60	0.30	1.17	.13
IL-1β	1.06	0.56	2.01	.86
IL-2	0.75	0.39	1.41	.37
IL-6	3.54b,c	2.06	6.10	<.001
IL-8	1.26	0.71	2.24	.43
sCD14	2.67b,c	1.55	4.61	<.001
sCD27	1.99b	1.13	3.53	.02
sGP130	2.06b	1.17	3.63	.01
sIL-2Rα	3.29b,c	1.85	5.85	<.001
sIL-6R	1.92b	1.10	3.37	.02
sTNFR2	2.04b	1.13	3.68	.02
TNF-α	1.80b	1.03	3.17	.04

Abbreviations: BAFF, B-cell activating factor; CCL, chemokine (CC motif) ligand; CRP, C-reactive protein; CXCL, chemokine (CXC motif) ligand; GM-CSF, granulocyte-macrophage colony-stimulating factor; HR, hazard ratio; IFN, interferon; IL-1β, interleukin 1β; IL-2, interleukin 2; IL-6, interleukin 6; IL-8, interleukin 8; IL-10, interleukin 10; IL-12p70, interleukin 12p70; LL, lower confidence limit; sCD14, soluble CD14; sCD27, soluble CD27; sGP130, soluble GP130; sIL-2Rα, soluble interleukin 2Rα; sIL-6R, soluble interleukin 6R; sTNFR2, soluble tumor necrosis factor receptor 2; TNF, tumor necrosis factor; UL, upper confidence limit.

^a Models adjusted for age, CD4⁺ cell count, hepatitis B or C infection, and smoking.

^b HRs significant at P < .05.

^c Biomarker HRs significant after controlling the family-wise error rate at 5% with the Benjamini-Hochberg procedure [24].

High concentrations of IL-6 (HR, 2.50; 95% CI, 1.41–4.45), sIL-2Rα (2.21; 1.20–4.10), and sCD14 (1.85; 1.03–3.32) retained strong relationships with mortality risk in an adjusted model that included the 4 biomarkers listed above that were significantly associated with mortality risk after adjustment for multiple tests. Model estimates are provided in Supplementary Table 2. Because multiple collinearity could obscure associations with mortality risk in this multivariable model, we report results from EFA below.

Results From EFA

Parallel analysis indicated 5 factors underlying the observed biomarker correlations (Supplementary Figure 2). Table 3 summarizes the biomarkers that comprised each factor after rotation. Table 3 also displays estimated HRs for each factor score from a Cox model adjusted for the same covariates as described above. Individuals with higher scores for factor 1 (defined by soluble receptors, particularly sTNFR2 and sIL-2Rα), had higher hazards of mortality (HR, 1.76; 95% CI, 1.43–2.16). Factor 5, defined by IL-6 and CRP, was also associated with mortality risk (HR, 1.41; 95% CI, 1.01–1.95). All biomarkers in factors 1 and 5 were individually associated with mortality risk at P < .05 except BAFF. Factors 2 (proinflammatory cytokines), 3 (chemokines), and 4 (IL-10 and IL-12p70) were not significantly associated with mortality risk in this model. Details of the EFA and factor score regression are found in Supplementary Tables 3 and 4.

Table 3.

Results of Exploratory Factor Analysis

Factor	Biomarkers (Loading)a	HR for Mortality (95% CI)b
1. Soluble receptors	sTNFR2 (0.91), sIL-2Rα (0.77), sCD27 (0.72), sGP130 (0.71), BAFF (0.64), sCD14 (0.50), sIL-6R (0.48)	1.76 (1.43–2.16)
2. Proinflammatory cytokines	IL-8 (0.83), IL-6 (0.67), IL-2 (0.67), TNF-α (0.63), IL-1β (0.58), CCL4 (0.56), GM-CSF (0.56)	1.11 (.82–1.49)
3. Chemokines	CCL2 (0.75), CCL13 (0.74), CCL11 (0.63), CCL17 (0.43)	0.98 (.72–1.33)
4. IL-10 and IL-12p70	IL-10 (0.90), IL-12p70 (0.71)	0.91 (.68–1.23)
5. CRP and IL-6	CRP (0.65), IL-6 (0.41)	1.41 (1.01–1.95)

Abbreviations: BAFF, B-cell activating factor; CCL, chemokine (CC motif) ligand; CI, confidence interval; CRP, C-reactive protein; GM-CSF, granulocyte-macrophage colony-stimulating factor; HR, hazard ratio; IL-1β, interleukin 1β; IL-2, interleukin 2; IL-6, interleukin 6; IL-8, interleukin 8; IL-10, interleukin 10; IL-12p70, interleukin 12p70; sCD14, soluble CD14; sCD27, soluble CD27; sGP130, soluble GP130; sIL-2Rα, soluble interleukin 2Rα; sIL-6R, soluble interleukin 6R; sTNFR2, soluble tumor necrosis factor receptor 2; TNF, tumor necrosis factor.

^a Biomarkers with loadings >0.4 are displayed for each factor. Loading ranges from −1 to 1 and represents the regression coefficient linking factors to observed variables; loading in a multiple-factor context is akin to, but not equivalent to, the correlation between a factor and an observed variable.

^b Mortality HRs are per unit increase in factor score in a model with all factors included, adjusted for age, CD4⁺ cell count, hepatitis B or C infection, and smoking status.

Sensitivity Analyses

We fit models treating biomarkers as categorical (comparing each quartile with the lowest) and continuous (per IQR increase on the log scale); results of these models are displayed in Supplementary Table 5. In categorical models, all but one of the significant HRs were for the top quartile. As expected given the large unit size, HRs per IQR increase on the log scale were highly significant (after controlling for multiple tests) for all 9 significant biomarkers from Figure 1 with the addition of BAFF. In addition, treating CD4⁺ cell count as a continuous covariate led to virtually identical conclusions.

For illustrative purposes, we present adjusted estimated survival curves using inverse probability of treatment weights for IL-6, sIL-2Rα, sCD14, and CXCL13 in Supplementary Figure 3. We also formally tested for linear interactions with time in the log HRs associated with quartiles of biomarker concentrations. Only 4 markers exhibited evidence (P < .05) for time-varying HRs, and none of these was significant after controlling for multiple comparisons. In each case, the HRs attenuated slightly over time (data not shown).

Biomarker concentrations in men who died soon after baseline might represent processes (eg, preexisting conditions or immune reconstitution inflammatory syndrome) distinct from those in the wider study population. However, only 2 deaths occurred within 1 year of HIV suppression, and exclusion of these observations produced results essentially identical to those reported above.

DISCUSSION

Residual inflammation and immune activation have been hypothesized to shorten life expectancies among those taking suppressive HIV therapy. Results presented here show strikingly large and robust associations between long-term mortality risk and certain inflammatory markers measured after cART-induced HIV suppression. Men with IL-6 or sIL-2Rα concentrations in the highest quartile had more than triple the hazards of long-term mortality relative to their peers; men with the highest quartile of sCD14 and CXCL13 concentrations had more than double the mortality hazards. These results are consistent with, and expand up, those derived from populations with less immunodeficiency, with biomarkers measured longer before death, and with fewer measured biomarkers than in the present study [19–21]. Correlations among biomarker concentrations suggest that more than one inflammatory process independently contributes to the mortality risk. These results were obtained despite long follow-up time (deaths occurred at a median 5.6 years after HIV suppression) and despite controlling for possible key confounders. The host inflammatory profile in early suppressed HIV infection seems highly predictive of long-term survival.

The biomarker most strongly associated with mortality risk was IL-6. IL-6 predicts mortality risk among various HIV-uninfected populations (eg, patients with cancer [25] and healthy elderly individuals [12]). IL-6 concentrations increase with HIV infection [26], though the mechanism is unclear [27]. cART-induced viral suppression did not normalize IL-6 concentrations in these data (Supplementary Figure 1) or in previous studies [8, 10]. Interestingly, IL-6 was the only biomarker in more than one factor, suggesting that multiple underlying processes affect IL-6 concentrations. IL-6 is a potent proinflammatory and T-helper 17–stimulating cytokine [28] and the central inducer of the acute-phase response, which leads to production of CRP and other acute-phase reactants [29]. The association of IL-6 with mortality risk may reflect a late common pathway of various infectious and/or inflammatory causes of death. If so, intervening directly with IL-6 would not lower mortality risks; future trials may address this question. It is unknown whether the association between IL-6 concentration and mortality risk was stronger among HIV-infected men than among those who were uninfected, because infrequent death among HIV-uninfected MACS participants precluded assessment of associations between biomarkers and mortality risk in this group.

Biomarker concentrations are observable outcomes of complex biological processes that are only partially understood. EFA can be useful in identifying broad patterns and addressing multicollinearity. It is reassuring that factor loadings from our results correspond to meaningful biomarker groupings: soluble receptors (factor 1), proinflammatory cytokines (factor 2), chemokines (factor 3), IL-10 and IL-12p70 (factor 4), and the IL-6/CRP pathway (factor 5).

It is not straightforward to establish what concentrations may be considered “abnormal” because of the lack of established, clinically relevant cutoffs (and variability across assays used in different studies). This is a primary reason for using quartiles of biomarker concentrations for analysis, as we did here. One limitation of this approach is that it cannot provide estimates of threshold values for risk. In addition, the quartiles defined in the study population defined in this study may not be generalizable to other populations. We partially addressed this issue by comparing distributions among the HIV-infected study population to those observed in HIV-uninfected MACS participants. Finally, factor scores from EFA are not generalizable to populations with different biomarker concentrations. Despite these limitations, our results strongly suggest that inflammation and immune activation are associated with long-term mortality risk in the population we studied; future studies will test the reproducibility of these associations in other populations.

Although it is conceivable that unmeasured, independent disease processes, rather than HIV per se, could explain the observed marker concentrations, we consider this unlikely: marker concentrations were higher in untreated HIV-infected men than in uninfected men drawn from the same population [10]. In addition, the fact that most deaths occurred many years after viral suppression strongly suggests that the men who died were not experiencing end-stage disease processes at the time of suppression. Finally, the time between suppression and measurement was fairly short in most cases (median, 0.7 years), and prior work by Wada et al [10] shows that biomarker concentrations are essentially static beyond 1 year after HIV suppression.

This study had several important strengths. The MACS cohort is very well characterized, including measurement of several key clinical and behavioral variables and precise timing of sentinel events. The long-follow-up time also permitted examination of mortality risks many years after measurement of potentially prognostic markers. Multiplex assays allowed us to examine a large number of biomarkers. Finally, comparison of biomarker concentrations with those in demographically similar HIV-uninfected MACS participants provided a useful degree of context for the concentrations observed in our study.

Biomarkers of inflammation and immune activation, particularly IL-6, were strongly associated with mortality risk despite HIV suppression in a cohort of cART-treated men. Residual inflammation may result from viremia below detectable levels or may represent immunologic processes that continue as a consequence of damage early in the course of infection. Measurement of the biomarkers reported here may help identify persons at higher risk of death during cART. Identifying inflammatory processes that are associated with mortality risk in those with treated HIV infection may help in the development of more effective therapies.

Supplementary Data

Supplementary materials are available at http://cid.oxfordjournals.org. Consisting of data provided by the author to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the author, so questions or comments should be addressed to the author.