Case–Control Sample: Baseline Biomarkers and All-Cause Mortality
Most of the deaths (79 of 85) occurred in the US. Sites in the US began enrollment 2 to 3 y before most sites in other countries and accounted for 55% of the randomized participants. In univariate analyses, cases and controls differed with respect to age (p = 0.007), baseline CD4+ count (p = 0.03), co-infection with hepatitis B or C (p = 0.0008), smoking status (p = 0.0001), diabetes (p = 0.03), use of BP-lowering treatment (p = 0.02), and prior CVD (p = 0.04) (). In multivariate analyses that included the baseline covariates described in Methods for adjusted analyses, age (p = 0.02), smoking (p = 0.01), prior CVD (p = 0.04), co-infection with hepatitis B or C (p = 0.03), and baseline CD4+ cell count (p = 0.10) were associated (p = 0.10 or lower) with mortality.
Characteristics of Deaths and Matched Controls at Study Entry
With the exception of amyloid P, levels of the inflammatory markers were higher in deaths than matched controls (, “study entry”). Differences between deaths and controls for IL-6 and D-dimer were highly significant (p < 0.0001); hsCRP was also significantly higher among deaths than controls (p = 0.005).
Study Entry and Latest Levels of Six Biomarkers for Deaths and Matched Controls
display biomarker levels for each triad (death and two matched controls) ordered by the biomarker level for the death. For IL-6 (Figure S5
) and D-dimer (Figure S6
), and to a lesser extent hsCRP (Figure S2
), there is a shift in the distribution for deaths (red circles) compared to controls (blue circles). For other biomarkers, control levels are lower than deaths with high levels and higher than deaths with low levels (in part, a reflection of the regression to the mean phenomenon).
For both IL-6 and D-dimer, the discordance of cases and matched controls in the lower quartile (<25th percentile) and upper quartile (>75th percentile) was striking. For D-dimer, there were 23 controls with levels in the lowest quartile (<0.18 μg/ml) that were matched with deaths that had levels in the upper quartile. In contrast, there were only two controls with D-dimer levels in the upper quartile matched with deaths in the lowest quartile. For IL-6, there were 25 controls with levels in the lowest quartile (<1.6 pg/ml) matched with deaths that had levels in the upper quartile, and three controls with IL-6 in the upper quartile matched with deaths in the lowest quartile.
Strong risk gradients with mortality were evident for both IL-6 and D-dimer (). For IL-6, unadjusted ORs (for each of the three upper quartiles versus lowest) were 8.3 (95% CI, 3.3–20.8), 3.2 (95% CI, 1.3–7.9), and 1.3 (95% CI, 0.5–3.6). For D-dimer corresponding ORs were 12.4 (95% CI, 4.2–37.0), 4.0 (95% CI, 1.3–12.3), and 3.2 (95% CI, 1.1–9.0). In models that considered these biomarkers as continuous variables after log10
transformation, a difference corresponding to the IQR was associated with an OR of 3.4 (95% CI, 2.2–5.4) for IL-6 and 3.9 (95% CI, 2.3–6.6) for D-dimer. Covariate adjustment tended to strengthen these associations (), and sensitivity analyses yielded consistent findings (Table S2
). Similar analyses with the 1:4 matching yielded results with reduced, but still very large, ORs for IL-6 and D-dimer (Table S3
). For example, unadjusted ORs (upper quartile versus lowest) were 6.1 (95% CI, 2.7–13.6) and 6.6 (95% CI, 2.9–14.9) for IL-6 and D-dimer, respectively.
Risk of Death Associated with Biomarker Levels at Study Entry
Significant associations between study entry levels of hsCRP and amyloid P and mortality were also evident. Risk of death increased with increasing levels of hsCRP but not as strongly as for IL-6 and D-dimer. The unadjusted OR (upper versus lower quartile of hsCRP) was 2.0 (95% CI, 1.0–4.1). Higher levels of amyloid P were associated with a lower risk of death in analyses based on the continuous biomarker level, but no apparent trend by quartile was evident. This association was only of borderline significance after covariate adjustment and may be due in part to outliers. After excluding outliers (ten deaths and seven controls), the OR associated with a one IQR higher level of amyloid P was 1.1 (95% CI, 0.7–1.7) (Table S2
When IL-6 and D-dimer were considered together in the same model they each remained significantly associated with all-cause mortality. Unadjusted ORs for the highest versus lowest quartile were 4.7 (95% CI, 1.7–12.7; p = 0.002) for IL-6 and 6.1 (95% CI, 2.0–18.6; p = 0.001) for D-dimer. ORs for a difference corresponding to the IQR were 2.7 (95% CI, 1.6–4.4; p < 0.0001) for IL-6 and 2.6 (95% CI, 1.5–4.6; p = 0.001) for D-dimer.
In the analyses described above, the DC and VS groups were combined. In separate analyses for each group (), associations between biomarker levels at baseline and mortality were similar (, note last two columns for comparison to models that considered both treatment groups combined). For DC participants, median levels (expressed as deaths, controls) were 4.49, 1.78 μg/ml for hsCRP; 3.85, 2.24 pg/ml for IL-6; and 0.63, 0.22 μg/ml for D-dimer. For VS participants, these median levels were 3.60, 3.07 μg/ml for hsCRP; 3.78, 2.43 pg/ml for IL-6; and 0.37, 0.29 μg/ml for D-dimer.
Risk of Death Associated with Biomarker Levels at Study Entry for the Drug Conservation (DC) and Viral Suppression (VS) Treatment Groups
Likewise, associations were similar for those with prior cancer, CVD, renal or liver disease (31% of deaths and 16% of controls), and those with no history of these conditions (p = 0.33 for hsCRP, p = 0.88 for IL-6, and p = 0.74 for D-dimer interactions); for those co-infected with hepatitis or not (p = 0.37 for hsCRP, p = 0.41 for IL-6, and p = 0.76 for D-dimer interactions); and for those with a baseline CD4+ cell less than 600 cells/mm3 (approximate median) and 600 cells/mm3 or more(p = 0.85 for hsCRP, p = 0.21 for IL-6, and p = 0.41 for D-dimer interactions).
Associations with hsCRP, Il-6, and D-dimer were also similar for deaths in the first year (38 deaths) and after the first year (47 deaths). During the first year, the OR corresponding to the IQR for hsCRP, IL-6, and D-dimer were 1.5 (95% CI, 0.9–2.4; p = 0.16), 2.5 (95% CI, 1.4–4.3; p = 0.002), and 3.9 (95% CI, 1.9–8.0; p = 0.0003), respectively. For deaths occurring after the first year, ORs for hsCRP, IL-6, and D-dimer were 1.9 (95% CI, 1.1–3.2; p = 0.01), 5.0 (95% CI, 2.4–10.4; p < 0.0001), and 3.9 (95% CI, 1.8–8.4; p = 0.0005).
Deaths due to substance abuse (eight deaths) or to accidents, violence, or suicide (seven deaths) could attenuate the associations between the biomarkers and mortality. Thus, analyses were carried out excluding these 15 deaths. Unadjusted ORs corresponding to a difference equal to the IQR were 1.9 (95% CI, 1.2–2.8) for hsCRP, 3.7 (95% CI, 2.2–6.3) for IL-6, and 4.0 (95% CI, 2.1–7.4) for D-dimer. Each of these ORs is larger by a small amount compared to the corresponding estimates in . Twenty-one deaths were classified as CVD or were unwitnessed. Unadjusted ORs of CVD or unwitnessed death corresponding to a difference equal to the IQR were 2.3 (95% CI, 1.0–5.0; p = 0.04) for hsCRP, 3.2 (95% CI, 1.2–8.4; p = 0.02) for IL-6, and 3.2 (95% CI, 1.1–9.3; p = 0.04) for D-dimer.
Case–Control Sample: Change in Biomarkers and All-Cause Mortality
(“latest level”) compares latest levels of each biomarker for deaths and matched controls (Table S4
gives similar results for the 1:4 matching). In these univariate analyses, significant differences were observed for all of the biomarkers except F1.2. Deaths had higher latest levels than controls, except amyloid P for which latest levels were lower for deaths than controls. Findings were strongest for hsCRP, IL-6, and D-dimer. Average differences between deaths and controls were greater for latest levels than for study entry levels ( “study entry” and Table S4
gives adjusted ORs for latest levels of each biomarker. Adjusted ORs corresponding to a difference equal to the IQR were 2.4 (95% CI, 1.4–4.2) for hsCRP, 2.0 (95% CI, 1.2–3.1) for IL-6, and 2.2 (95% CI, 1.1–4.1) for D-dimer. These models also included study entry levels of each biomarker. Study entry levels of IL-6 and D-dimer remained significantly associated with all-cause mortality after consideration of latest levels (p = 0.0008 for IL-6 and p = 0.003 for D-dimer). After considering latest level of hsCRP, the study entry level was not significant (p = 0.07). In models that also included latest levels of HIV-RNA and CD4+ cell count, neither of which were significantly associated with all-cause mortality, these associations were diminished slightly but remained significant. For latest levels of hsCRP, IL-6, and D-dimer, the ORs were 2.2 (p = 0.009), 1.7 (p = 0.04), and 2.0 (p = 0.04).
Risk of Death Associated with Latest Level of Each Biomarker
Random Sample: Associations at Baseline
As previously reported, the majority of patients in SMART were using ART at entry [1
]. In the random sample, 74% were using ART; among those using ART, 71% had an HIV-RNA level 400 copies/ml or lower. Approximately 6% of patients had not previously used ART; the remainder of those not using ART had discontinued it prior to enrolling in SMART. In the random sample, treatment groups were well balanced ().
Characteristics of Random Sample of DC and VS Participants at Study Entry
Multiple regression analyses of each biomarker (after log10 transformation) on baseline covariates were performed. The covariates used in the regression analyses were the same as those used in the adjusted case–control analysis. An exception was history of CVD since no one in the random sample had a history of CVD (). D-dimer was significantly higher for those not on ART than for those on ART with an HIV-RNA level 400 copies/ml or lower (0.15 on log10 scale; p = 0.0007) and for those on ART with an HIV-RNA over 400 copies/ml (0.12; p = 0.02). Other biomarkers did not vary significantly according to use of ART and HIV-RNA level at study entry.
Smoking and co-infection with hepatitis B or C, which were both significantly related to mortality, were not significantly associated with either IL-6 (p = 0.19 for smoking and p = 0.16 for co-infection) or D-dimer (p = 0.64 for smoking and p = 0.39 for co-infection) at baseline. Smoking was not significantly associated with any of the biomarkers. hsCRP was lower by 0.217 μg/ml after log10 transformation (p = 0.0001) and amyloid P was lower by 0.068 (p = 0.0005) for those who were co-infected with hepatitis; co-infection was not associated with the other biomarkers.
Significant predictors of log10 IL-6 were age (0.069 higher with each 10 y in age; p < 0.0001) and BMI (0.009 higher with each kg/m2 greater BMI; p = 0.0004). For log10 D-dimer, in addition to ART and HIV-RNA level, levels were greater among older participants (0.062 higher with each 10 y; p = 0.002), for black participants (0.123; p = 0.0009), for participants with diabetes (0.161; p = 0.01), and for those with greater BMI (0.008 per unit higher; p = 0.02). D-dimer levels were lower for men (−0.151; p = 0.0003) and for those with higher CD4+ cell counts (−0.023 per 100 cells/mm3 higher; p = 0.002). To put these changes in perspective, the IQRs on the log10 scale for IL-6 and D-dimer were 0.39 pg/ml and 0.58 μg/ml, respectively. Differences as large as the IQR for each marker were associated with 3- to 4-fold greater risks of all-cause mortality.
Random Sample: Treatment Differences at 1 mo
shows average changes in log10 transformed biomarker levels 1 mo after randomization. IL-6 and D-dimer increased significantly (p = 0.0005 and p < 0.0001, respectively) from study entry to 1 mo in the DC group compared to the VS group (p < 0.0001). Considering the non-transformed levels, the median increase in D-dimer for DC patients was 0.05 μg/ml (a 16% increase); IL-6 increased by 0.60 pg/ml in the DC group (a 30% increase). For VS patients, the median increases were 0.0 μg/ml and 0.12 pg/ml (a 5% increase) for D-dimer and IL-6, respectively. Changes in hsCRP and amyloid A were in the same direction—greater increases for DC compared to VS patients—but did not differ significantly between treatment groups.
Biomarker, CD4+ Cell Count Change, and HIV-RNA Level Change 1 mo after Randomization
For both IL-6 and D-dimer, treatment differences were greater for patients who were on ART at entry and had HIV-RNA levels 400 copies/ml or below. For this subgroup (52% of patients), D-dimer increased in the DC group by 0.07 μg/ml (a 27% increase) and declined in the VS group by −0.02 μg/ml (p < 0.0001 for treatment difference). Similarly, for IL-6, the median increases for DC and VS patients were 0.98 (a 43% increase) and 0.08 pg/ml, respectively (p < 0.0001 for difference).
The changes in IL-6 and D-dimer for the subgroup of DC patients with HIV-RNA levels 400 copies/ml or below were further examined according to HIV-RNA levels at 1 mo. Following ART interruption, biomarker increases were greater for those with higher HIV-RNA levels at 1 mo ( and ).
Change in Log10 IL-6 (pg/ml) from Baseline to 1 mo According to HIV-RNA Level at 1 mo for Participants in the Drug Conservation (DC) Group (CD4+ Guided Intermittent ART) with an HIV-RNA level 400 Copies/ml or Less at Baseline
Change in Log10 D-dimer (μg/ml) from Baseline to 1 mo According to HIV-RNA Level at 1 mo for Participants in the Drug Conservation (DC) Group (CD4+ Guided Intermittent ART) with an HIV-RNA level 400 Copies/ml or less at Baseline
Results in were used to estimate the potential impact of treatment differences on mortality. For IL-6, the DC/VS difference after 1 mo on the log10 scale was 0.08 pg/ml. Based on the regression analysis cited in , a difference of this magnitude is associated with a 16% increased risk of death (95% CI, 10%–25%). Similarly, the 0.11 log10 higher level of D-dimer for DC compared to VS participants is associated with a 24% (95% CI, 13%–46%) increased risk of death.
Impact of Adjustment for Latest Levels of IL-6 and D-dimer on OR for DC Versus VS for All-Cause Mortality
Matched logistic models were used to assess the effect of adjusting for latest levels of IL-6 and D-dimer on the DC/VS OR for death. In the model with two controls per case, the unadjusted OR for the DC versus the VS group was 1.3 (95% CI, 0.8–2.2). Because this OR was considerably lower than the hazard ratio previously reported for all-cause mortality [1
], we explored reasons for it and created an expanded case–control data set. A chance imbalance in the number of DC and VS participants selected as controls is the reason the OR was lower. Among the 170 controls, 99 were in the DC group and 71 were in the VS group. The expected number was 85 in each group. With the expanded case–control study (four controls for each death), the unadjusted DC/VS OR for all-cause mortality was 1.8 (95% CI, 1.1–3.1). This estimate is identical to that previously reported [1
]. With adjustment for latest level of IL-6, the OR was 1.5 (95% CI, 0.8–2.8); with adjustment for latest level of D-dimer the OR was 1.4 (95% CI, 0.8–2.5).
We also considered the effect of adjusting for both study entry and latest levels of IL-6 and D-dimer and of adjusting for HIV- RNA and CD4+ cell count on the DC/VS OR (Table S5
). Similar to an earlier report, adjustment for CD4+ cell count had a greater effect on the OR (DC/VS) for mortality (OR = 1.2; 95% CI, 0.7–2.2) than adjustment for latest HIV-RNA levels (OR = 1.6; 95% CI, 0.9–2.9) [1
]. With adjustment for latest levels of IL-6, D-dimer, CD4+ cell count, and HIV-RNA, the OR (DC/VS) was 1.3 (95% CI, 0.6–2.6).