- Top of page
- Supporting Information
Reduced kidney function is common in HIV-infected individuals . In clinical practice, glomerular filtration rate (GFR) is usually estimated using serum creatinine. However, variability in creatinine production, which is principally determined by muscle mass and diet, may affect the accuracy of GFR estimates . Cystatin C is an alternate GFR marker that is produced by all nucleated cells in the body at a constant rate and appears to be less affected by variation in muscle mass than creatinine . However, non-GFR determinants of cystatin C have not been fully elucidated and correlations have been reported between cystatin C and thyroid disease and inflammation markers [4-6].
The Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) has developed GFR estimating equations based on standardized creatinine (GFRcr), cystatin C (GFRcys) and the combination of the two markers (GFRcr-cys), using data from over 5000 participants in 13 studies who had GFR measured by urinary or plasma clearance of an exogenous filtration marker . These three estimating equations have recently been compared with plasma clearance of iohexol in a sample of HIV-infected subjects . In both the general population study  and the HIV study , the GFRcr-cys equation was the most accurate of the three, while the accuracies of GFRcr and GFRcys were similar.
Cystatin C has consistently been found to be a stronger predictor of all-cause mortality and cardiovascular events (CVEs) than creatinine in the general population [9, 10]. However, less is known about the association between cystatin C and clinical events in HIV-infected persons. In the present study, we compared the associations of baseline GFRcr, GFRcys and GFRcr-cys with mortality, CVE or opportunistic disease (OD) in the Strategies for the Management of Antiretroviral Therapy (SMART) study.
- Top of page
- Supporting Information
In this secondary analysis of data from the SMART study, we found that baseline kidney function, as measured using any of the three CDK-EPI GFR estimating equations, was independently associated with all-cause mortality during follow-up. However, only GFRcys was statistically significantly associated with CVE and OD in fully adjusted models. GFRcr had the weakest associations with clinical outcomes of the three equations, while GFRcr-cys had associations with outcomes that were close to, but slightly smaller than, those of GFRcys. We also found that inclusion of GFRcys in a base model with GFRcr statistically significantly improved discrimination (IDI) for all three outcomes, whereas inclusion of GFRcr-cys in a base model with GFRcys did not improve discrimination for any outcome.
These results are consistent with studies of patient populations not infected with HIV, in which cystatin C-based measures have been found to be more strongly associated with mortality, CVE and incident heart failure than creatinine-based measures [9, 10]. Additionally, an analysis by Choi and colleagues of 922 HIV-infected subjects enrolled in the Fat Redistribution and Metabolic Change in HIV Infection study found that cystatin C-based GFR was significantly and independently associated with 5-year all-cause mortality, whereas creatinine-based GFR was not significantly associated with mortality in unadjusted or adjusted analyses .
The mechanisms by which cystatin C-based GFR estimates are more strongly associated with clinical events compared with creatinine-based estimates are not clear. It is unlikely that GFRcys is a stronger predictor of clinical events than GFRcr by virtue of being a more accurate index of GFR. In recent studies evaluating the performance of the CKD-EPI estimating equations in non-HIV-infected  and HIV-infected  individuals, there was no evidence that GFRcys was more accurate or precise than GFRcr. Additionally, GFRcr-cys, which is the most accurate and precise of the CKD-EPI equations [7, 8], was a slightly weaker predictor of clinical outcomes than GFRcys in our study, suggesting that improved surrogacy for clinical outcomes is not driven by greater GFR fidelity.
A second possibility is that cystatin C tracks inflammatory processes more closely than creatinine. We found that baseline GFRcys had stronger correlations than GFRcr with HIV RNA, CD4 cell count, hs-CRP, IL-6 and D-dimer, corroborating prior studies of these biomarkers in HIV-infected persons [17-19]. In populations without HIV infection, inflammation is an established risk factor for cardiovascular disease and mortality . Emerging data from populations with HIV infection highlight the strong associations between inflammatory markers, notably IL-6 and D-dimer, and clinical outcomes, including all-cause mortality  and cardiovascular disease [22, 23]. Consequently, inflammation may mediate the stronger association between GFRcys and clinical events, compared with GFRcr. However, we found that adjusting for hs-CPR, IL-6 and D-dimer led to only modest changes in the observed associations with clinical events. Similarly, we found that adjustment for HIV-related factors, including baseline HIV RNA, baseline CD4 count and nadir CD4 count, led to only small changes in the observed associations between GFR estimates and clinical events. Moreover, there was little evidence that adjustment for HIV-related factors or inflammation markers had larger effects on GFRcys−outcome associations than on GFRcr−outcome associations, as might be anticipated from the stronger baseline correlations between GFRcys and these factors. This suggests that cystatin C plasma concentrations have additional and as yet unidentified determinants that are predictive for clinical outcomes.
Our study has strengths, including a large and diverse sample of HIV-infected participants, prospectively defined outcomes that were adjudicated by an endpoint committee, use of standardized plasma creatinine and cystatin C measurements, use of the updated CKD-EPI GFR estimating equations that have validation data from HIV-infected subjects , and the ability to adjust for systematically collected and measured inflammation markers.
Our study also has limitations. First, we had to exclude 858 SMART participants from the analysis who lacked baseline measures of plasma creatinine or cystatin C, which reduced our power to assess associations with clinical events. Additionally, excluded subjects were at higher risk for clinical outcomes than included subjects. This would be expected to lead to an underestimation of clinical event rates in our analysis, although it is unclear why this would bias the comparison of the relative prognostic importance of the studied biomarkers. Secondly, we did not have measurements of proteinuria, which is a strong and GFR-independent risk factor for clinical events [16, 24]. Thirdly, we did not directly measure GFR with exogenous filtration markers. Fourthly, half of the participants in our study were managed with episodic antiretroviral therapy, which was found to be inferior to continuous treatment , and does not reflect the current standard of HIV care. However, we found no evidence that the observed associations differed by study arm, and a subgroup analysis of subjects assigned to continuous therapy produced qualitatively similar findings to the overall analysis.
Future research should examine the mechanisms by which cystatin C predicts clinical outcomes in studies that include directly measured GFR and biomarkers for alternative pathogenic pathways. Additionally, studies are indicated to determine if patient management and clinical outcomes can be improved with the use of cystatin C-based screening or treatment algorithms.
In summary, we found that GFR estimates that used plasma cystatin C had stronger associations with mortality, CVE and OD than GFR based on plasma creatinine in a secondary data analysis of HIV-infected participants in the SMART study. GFRcys had stronger correlations than GFRcr with a variety of factors that might mediate associations with clinical outcomes, including CD4 cell count, HIV RNA and inflammation markers. However, adjustment for these factors only modestly reduced the observed associations between GFRcys and clinical events, suggesting that these covariates are only partial mediators.