Urinary Biomarkers of Kidney Injury Are Associated with All-Cause Mortality in the Women’s Interagency HIV Study (WIHS)

Participants

We included 908 HIV infected women participating in the HIV Kidney Aging substudy of the Women’s Interagency HIV Study (WIHS). WIHS is a large observational study designed to understand risk factors for the progression of HIV in women. The WIHS study design and methods have been described previously. (17, 18) In brief, 3,766 women (2,791 HIV-infected and 975 HIV-uninfected) were enrolled in either 1994–1995 (n=2,623) or 2001–2002 (n=1,143). Women were recruited to be representative of the HIV-infected population at the time of enrollment in the following 6 U.S. communities (Bronx/Manhattan, Brooklyn, Chicago, Los Angeles, San Francisco, and Washington, DC). Participants are interviewed and examined every six months. Serum specimens were stored in a −80°C freezer until biomarker measurement.

For these analyses, we utilized data from the WIHS HIV Kidney Aging study. This substudy is a retrospective cohort designed to investigate the onset of kidney disease in the setting of HIV, utilizing stored urine and serum specimens. All HIV-infected women with available specimens were included in this substudy. Single, baseline measures were collected between October 1999 and March 2000. Although 908 HIV-infected and 289 HIV-uninfected women were included in the substudy, only HIV-infected women were included in these analyses (N=908). All 908 HIV infected women had urinary samples from baseline. WIHS was approved by the relevant institutional review boards at all study sites. The ancillary study of kidney injury was also approved by the University of California, San Francisco, San Francisco Veterans Affairs Medical Center, and Yale committees on human research.

Predictors

The urinary injury biomarkers ACR, IL-18, L-FABP, KIM-1, and NGAL were measured at the Cincinnati Children’s Hospital Medical Center Biomarker Laboratory. Urine microalbumin and creatinine were measured by immunoturbidimetry and colorimetric enzyme assay, respectively, using a Siemens Dimension Xpand plus HM clinical analyzer (Siemens, Munich, Germany). Urine IL-18 and L-FABP were measured using commercially available ELISA kits (Medical & Biological Laboratories Co., Nagoya, Japan and CMIC Co., Tokyo, Japan, respectively) per manufacturer’s instructions. The urine KIM-1 ELISA was constructed using commercially available reagents (R & D Systems, Inc., Minneapolis, MN). (19) Urine NGAL was assayed using a human-specific commercially available ELISA (AntibodyShop, Grusbakken, Denmark). (20) All urine specimens were in continuous storage without prior freeze-thaw. Laboratory personnel were blinded to clinical information about the participants. Coefficients of variation for the urine measures were: albumin, 5.9%; creatinine, 4.1%; IL-18, 5.2%; L-FABP, 8.9%; KIM-1, 5.2%; and NGAL, 5.4%.

Outcome

The primary outcome of this study was all-cause mortality. Data on vital status and date of death were obtained from medical records, providers, contacts, and the National Death Index. Deaths have been ascertained through May, 2011. Detailed methods have been previously published. (7, 21, 22)

Adjustment Variables

Covariates of interest included demographic characteristics, kidney disease risk factors, HIV-specific risk factors, and kidney disease measures. These were obtained as part of the WIHS semiannual exam. Candidate variables considered were age, race/ethnicity, menopause status, antihypertensive medication use, diabetes (defined as any of the following: fasting glucose ≥126mg/dL, self-reported diabetes, self-reported diabetes medication, or HbA1c ≥6.5), cigarette smoking (current, former, never); systolic and diastolic blood pressure, LDL and HDL cholesterol, triglycerides, body mass index, and waist circumference. GFR was estimated using the CKD Epi equation for cystatin C. (23) Cystatin C was measured by a particle-enhanced immunoturbidometric assay (Gentian, www.atlanticdiagnostics.com), which has been calibrated against the new World Standard Reference material ERM-DA471/IFCC). HIV-related factors included hepatitis C virus (HCV) infection (defined by detectable HCV RNA), current CD4 cell count, nadir CD4 cell count, history of AIDS diagnosis (including low CD4), heroin use, current HIV viral load, HAART use, NRTI use, NNRTI use at baseline, and current PI use. All covariates were measured at baseline, except CD4 count and HIV RNA which were time-updated. Multiple imputation with the Markov chain Monte Carlo method was used to impute missing data for other covariates, with 5 imputations to yield ~95% relative efficiency. The percentage of missing observations for each covariate ranged from less than 1% to 15%.

Statistical Analysis

We first present baseline characteristics of the 908 HIV-infected women included in this study. We estimated correlations of urinary biomarkers using Spearman correlation. Then, we evaluated the association of each injury marker with all-cause mortality using multivariable Cox proportional hazards regression. We first analyzed the urine injury biomarkers as continuous (log-transformed) predictors of mortality, but the assumption of linearity did not hold for all measures. We therefore present results with the biomarkers categorized into tertiles, as we have done in a prior study. (16) To determine whether each injury marker was independently associated with mortality, multivariable models were sequentially adjusted for: 1) demographics, and 2) traditional kidney disease risk factors, HIV-specific risk factors, eGFRcys and ACR. Factors forced in the full model included age, race/ethnicity, eGFRcys, ACR (for all the other biomarkers), hypertension and diabetes, current HIV viral load, current CD4 cell count, and HCV infection. Specifically, we included eGFRcys and ACR because these are established markers of kidney disease and they are strongly associated with mortality. We used a stepwise backward selection with a significance level of α=0.05 to remove candidate covariates that were not associated with the outcome. To allow each marker to compete as an independent predictor of mortality, we constructed a final model that concurrently adjusted for those markers that remained statistically significantly associated with mortality after full adjustment above.

In our initial phase of analyses, we observed that the association between urinary L-FABP and mortality appeared J-shaped. To understand joint associations of L-FABP and IL-18 as predictors of death, we compared the age-adjusted rates of mortality across L-FABP tertiles, stratified by IL-18 tertiles. Finally, to understand the utility of measuring several biomarkers concurrently, we evaluated the age-adjusted rates of mortality among persons who had 1, 2, 3 or more biomarkers in the “worst” tertile (i.e. indicating “worst” prognosis).

We conducted several sensitivity analyses. First, we repeated our analyses using the slope of eGFRcys as the adjustment variable. Decline in eGFRcys has been shown to be an independent predictor of adverse outcomes. Moreover, we believe that a useful urinary biomarker of subclinical kidney injury should be associated with adverse outcomes independently of change in kidney function over time. Second, we repeated our L-FABP analyses after exclusion of persons with Hepatitis C infection because L-FABP levels are known to vary in liver disease. We tested for an interaction by eGFRcys to examine whether associations were consistent across the eGFRcys spectrum. Since serum creatinine remains the clinical standard for eGFR estimation, we also conducted a sensitivity analysis accounting for eGFRcreat rather than eGFRcys, using the CKD Epi equation. (24) Finally, we used a complete case approach to model the association of urine biomarkers with mortality, excluding participants who were missing any of the covariates in the model. The largest fraction of missing data in our analysis was for Hepatitis C (11%), while 2–3% of subjects were missing CD4 or HIVRNA values. All other covariates were missing for less than 1% of subjects. All analyses were conducted using SAS version 9.2 (SAS Institute Inc., Cary, NC, USA).

Participant Characteristics

Among 908 HIV infected women, mean (SD) age was 41±8 years, and mean eGFRcys was 88±21 ml/min/1.73m² at baseline. Thirty-one percent of participants had undetectable viral load, 31% had concomitant hepatitis C infection, and 24% had albumin to creatinine ratio ≥30 mg/g. (Table 1)

Median (IQR) biomarker concentrations were: IL-18 122 (110–133) pg/mL, L-FABP 4.5 (1.8–9.5), KIM-1 483 (439–535) pg/mL, NGAL 36.4 (32.7–39.6) ng/mL, and ACR 10.0 (9.2–11.1) mg/g. Urinary tubular injury biomarkers were moderately intra-correlated, with most correlations ranging from 0.4 to 0.6. The correlation coefficients with ACR were: −0.018 for IL-18, 0.087 for L-FABP, 0.014 for KIM-1, and 0.119 for NGAL.

Overall, there were 201 deaths during 9,269 person-years of follow-up. Mean (SD) follow-up was 10.2 ± 2.4 years (median: 11.4, IQR: 10.3–11.5). In demographic-adjusted analyses, highest tertiles of IL-18, KIM-1, NGAL, and ACR were associated with increased ten-year mortality risk (Table 2), and the middle tertiles of IL-18 and ACR were also associated with increased risk.

In fully adjusted models, persons in the highest tertile of IL-18 had over 80% higher risk for death compared with the lowest tertile, and this association remained significant even after adjustment for ACR and eGFRcys. For comparison, persons in the highest tertile of ACR had an almost 50% higher mortality risk, compared to the lowest tertile. Persons in the highest tertile of KIM-1 were at higher mortality risk, but this association was substantially attenuated after adjustment and no longer statistically significant. Levels of NGAL showed little association with mortality after adjustment. (Table 2) We observed a J-shaped association between L-FABP and mortality risk. Persons in the middle tertile of L-FABP had a 27% lower risk for death in demographic-adjusted analysis and a 33% lower risk after multivariable adjustment, compared with the lowest tertile. (Table 2)

We next evaluated the joint associations of IL-18, L-FABP and ACR tertiles in the fully adjusted analyses. Persons in the highest tertile of IL-18 remained at a two-fold mortality risk compared to the lowest tertile. The middle tertile of L-FABP was associated with a 40% lower mortality risk, and the highest tertile appeared to have a moderately lower risk compared to the lowest tertile, but the latter finding was not statistically significant (p=0.12). The highest tertile of ACR remained associated with a 50% higher mortality risk. (Figure 1) When we also added KIM-1 to the joint model, IL-18 remained strongly associated with mortality: HR 1.95, 95%CI 1.29 to 2.93 for highest tertile compared to lowest. Findings were also similar for L-FABP, HR 0.60, 95%CI 0.41 to 0.89. In contrast, associations of KIM-1 with mortality were much weaker in the joint model: HR for highest tertile of KIM-1 compared to lowest was 1.23 (95%CI 0.81 to 1.87).

Due the unexpected direction of association between L-FABP and mortality, we further examined the joint associations of IL-18 and L-FABP with mortality. We found that similar association between L-FABP and mortality was observed within each tertile of IL-18, and that increasing IL-18 was associated with mortality, regardless of level of L-FABP (overall test for interaction from full model: p = 0.22). (Figure 2)

Finally, we evaluated additive associations of urine biomarkers with mortality risk by categorizing each participant as having an “abnormal” level if the biomarker was within the “worst” tertile (defined as the lowest tertile for L-FABP, highest tertile for all others). We found a progressive increase in mortality risk by number of elevated biomakers (Figure 3), whereby those with 3 or more abnormal markers had a three-fold higher death rate compared to those with no abnormal markers. (Figure 3)

Sensitivity Analyses

We found that adjustment for the slope rather than baseline eGFRcys did not change our findings. For example, persons in the highest tertile of IL-18 had a HR of 1.99 (95%CI 1.37 to 2.87) compared to the lowest. For L-FABP, HR for middle tertile was 0.68 (95%CI 0.47 to 0.99) compared with the lowest. Similar results were found in an additional sensitivity analysis in which we excluded persons with hepatitis C: HR=1.87 (95%CI 1.01 to 3.47) for tertile 3 vs. tertile 1 of IL-18; HR=0.55 (95%CI 0.30 to 0.99) for tertile 2 vs. tertile 1 of L-FABP, and HR=1.50 (95%CI 0.87 to 2.58) for tertile 3 vs. tertile 1 of ACR. We found no statistically significant interactions of eGFRcys with IL-18 (p=0.70), ACR (p=0.97), or L-FABP (p=0.57) on mortality. We also conducted a sensitivity analysis adjusting for eGFRcreat rather than eGFRcys. In fully adjusted models, concurrently adjusting for the other markers, the associations of each marker with death were not materially different. For example, compared to the lowest tertile of urinary IL-18, the HR for death for the highest tertile was 2.13 (1.44, 3.13). For ACR, HR was 1.56 (1.08, 2.26). The middle tertile of L-FABP had the lowest risk for death, with a HR of 0.65 (0.44, 0.95), compared with the lowest tertile. Finally, results using a complete case analysis showed that higher IL-18 (HR=1.66, p=0.02 for tertile 3 vs. tertile 1) and higher ACR (HR=1.75, p=0.007 for T3 vs. T1) remained independently associated with increased risk of mortality after adjustment, while L-FABP had a J-shaped association with mortality (HR=0.63, p=0.04 for T2 vs. T1).”