Filtration Markers as Predictors of ESRD and Mortality: Individual Participant Data Meta-Analysis

Abstract

Background and objectives

Serum β-trace protein (BTP) and β-2 microglobulin (B2M) are associated with risk of ESRD and death in the general population and in populations at high risk for these outcomes (GP/HR) and those with CKD, but results differ among studies.

Design, setting, participants, & measurements

We performed an individual patient-level meta-analysis including three GP/HR studies (n=17,903 participants) and three CKD studies (n=5415). We compared associations, risk prediction, and improvement in reclassification of eGFR using BTP (eGFR_BTP) and B2M (eGFR_B2M) alone and the average (eGFR_avg) of eGFR_BTP, eGFR_B2M, creatinine (eGFR_cr), and cystatin C (eGFR_cys), to eGFR_cr, eGFR_cys, and their combination (eGFR_cr-cys) for ESRD (2075 events) and death (7275 events).

Results

Mean (SD) follow up times for ESRD and mortality for GP/HR and CKD studies were 13 (4), 6.2 (3.2), 14 (5), and 7.5 (3.9) years, respectively. Compared with eGFR_cr, eGFR_BTP and eGFR_B2M improved risk associations and modestly improved prediction for ESRD and death even after adjustment for established risk factors. eGFR_avg provided the most consistent improvement in associations and prediction across both outcomes and populations. Assessment of heterogeneity did not yield clinically relevant differences. For ESRD, addition of albuminuria substantially attenuated the improvement in risk prediction and risk classification with novel filtration markers. For mortality, addition of albuminuria did not affect the improvement in risk prediction with the use of novel markers, but lessened improvement in risk classification, especially for the CKD cohort.

Conclusions

These markers do not provide substantial additional prognostic information to eGFR_cr and albuminuria, but may be appropriate in circumstances where eGFR_cr is not accurate or albuminuria is not available. Educational efforts to increase measurement of albuminuria in clinical practice may be more cost-effective than measurement of BTP and B2M for improving prognostic information.

Introduction

CKD is associated with progression to ESRD and increased risk of death (1–4). eGFR from creatinine and albuminuria are two important measures for detection and prognosis of CKD. However, albuminuria is not widely measured and providing accurate prognostic information from a single blood draw would be efficient. Creatinine, the most commonly used filtration marker, is limited especially in patients with reduced muscle mass. Cystatin C provides more accurate prognoses than creatinine for a range of adverse outcomes (5,6), but has not been widely adopted, in part because of limitations in standardization of its assay (7). Like cystatin C, β-trace protein (BTP) and β-2-microglobulin (B2M) are novel low mol wt serum protein filtration markers that are less affected by muscle mass than creatinine. Serum concentrations of BTP or B2M are more strongly associated with risk for adverse outcomes than creatinine and, in some cases, cystatin C (8–13). However, results have differed among studies, possibly because of differences among populations and lack of a common method to compare the four filtration markers.

Our objectives were to determine whether recently developed GFR-estimating equations using BTP and B2M alone or in combination with creatinine and cystatin C provide improved assessment of risk for ESRD or death from any cause as compared with the use of creatinine-based eGFR across diverse cohorts.

Materials and Methods

Study Population

Our analysis included 23,318 individuals from six studies where BTP and B2M were measured as part of the CKD-Biomarkers Consortium. Two are from the general population (National Health and Nutrition Examination Survey [NHANES]; and Atherosclerosis Risk in Communities Study [ARIC],visit 4) and one is a population of people with diabetes but without CKD and at higher risk for ESRD and death (Pima) (herein referred to as GP/HR studies). Three were populations with CKD (Chronic Renal Insufficiency Cohort [CRIC]; Modification of Diet in Renal Disease [MDRD] Study; and African American Study of Kidney Disease [AASK]) (herein referred to as CKD studies). Participants from each study were included if they had nonmissing values for all four filtration markers (creatinine, cystatin C, BTP, and B2M) and demographic variables (Supplemental Table 1).

eGFR

See Supplemental Material for details related to assay of the filtration markers. We used Chronic Kidney Disease Epidemiology Collaboration equations to compute eGFR on the basis of creatinine (eGFR_cr), cystatin C (eGFR_cys), the combination of creatinine and cystatin C (eGFR_cr-cys), BTP (eGFR_BTP), and B2M (eGFR_B2M) (14). We averaged the four eGFRs to obtain an eGFR on the basis of all four markers (eGFR_avg).

Study Outcomes

The outcomes of interest were ESRD and all-cause mortality. ESRD events were defined as either initiation of maintenance dialysis or receiving a kidney transplant during follow-up. See Supplemental Table 2 for details.

Statistical Analyses

Analyses were conducted within each study and were then meta-analyzed across GP/HR and CKD studies using random-effects models. We examined heterogeneity among studies using the chi-squared test and I² statistic.

We first assessed the strength of associations of eGFR with the two outcomes. Within each study, we used Cox proportional hazards models adjusted for demographic and clinical confounders (systolic BP, diabetes, c-reactive protein, smoking, cardiovascular disease) and then additionally for eGFR_cr. We applied splines to examine the association of eGFR on the continuous scale with the outcomes. For the GP/HR studies, we used knots at 60 and 90 ml/min per 1.73 m². To provide stability to the data, we used a reference point slightly below the knot, at 80 ml/min per 1.73 m². For the CKD studies, we used a knot at 60 ml/min per 1.73 m² and a reference point at 50 ml/min per 1.73 m². Hazard ratios were expressed per 15 ml/min per 1.73 m². We used seemingly unrelated regression to compare the strength of hazard ratios for eGFR from other filtration markers to the hazard ratio for eGFR_cr within the GFR ranges of <60 and 60–90 ml/min per 1.73 m² (15). In sensitivity analyses, we also compared the novel markers to eGFR_cys and eGFR_cr-cys. In addition to the continuous analyses, we also examined hazard ratios for each eGFR category.

We next assessed discrimination of novel filtration markers for prediction of future events using c-statistics and net reclassification improvement (16). We used two base models for these comparisons, the multivariable model described above and eGFR_cr, and this model with the addition of albuminuria. A priori, we defined thresholds for risk using different thresholds on the basis of outcome and population given that ESRD is a less frequent outcome than mortality, and GP/HR cohorts are less likely to have either ESRD or mortality than CKD cohorts. For ESRD in the GP/HR and CKD studies, we used risk cutoffs at 1 year of 1% and 10%. For longer follow-up, GP/HR cohorts are likely to have fewer events than CKD studies, and we therefore used risk cutoffs of 10% and 65% at 10 years for the GP/HR cohorts and 5% and 40% at 5 years for the CKD studies. For mortality, we used a 1-year risk cutoff of 2% and 4% in both the CKD and GP/HR studies. In the GP/HR studies, the 10-year risk cutoffs were 18% and 34%, and for CKD studies the 5-year risk cutoffs were 10% and 18%.

We explored differences among subgroups where possible given study characteristics. Subgroup analyses were done according to sex, age (< and >65 years for GP/HR and > and ≤60 years for CKD), race (black versus nonblack, restricted to GP studies of ARIC and NHANES), diabetes status (restricted to GP studies of ARIC and NHANES and prevalent cardiovascular disease). We used different age cutoffs for the two populations because of different age distributions among the populations.

Results

The pooled GP/HR cohort included 17,903 participants and was slightly older, more likely to be women and white, and less likely to have diabetes and coronary heart disease than the pooled CKD cohort, which included 5415 participants. The mean values for the eGFR on the basis of creatinine, cystatin C, BTP, and B2M were 88, 79, 62, and 74 ml/min per 1.73 m², respectively, among the GP/HR cohorts and 45, 44, 45, and 43 ml/min per 1.73 m², respectively, among the cohorts with CKD (Table 1).

Table 1.

Clinical characteristics across studies

Clinical Characteristics	General Population		High Risk	GP/HR Total	CKD			CKD Total
	ARIC	NHANES	Pima		AASK	CRIC	MDRD
Total, N	10,583	7061	259	17,903	828	3792	795	5415
Age, yr	63 (6)	56 (20)	43 (11)	60 (14)	55 (11)	58 (11)	52 (12)	56 (11)
Women, %	57	52	69	55	38	45	40	43
Black, %	22	25	0	23	100	41	8	45
BMI, kg/m2	29 (6)	27 (5)	35 (8)	28 (6)	31 (7)	32 (8)	27 (4)	31 (7)
Current smoker, %	14	21	NR	17	28	13	10	15
Diabetes, %	17	14	100	17	0	52	11	38
Prevalent CHD, %	8	6	6	7	51	33	13	33
eGFRcr, ml/min per 1.73 m2	86 (16)	90 (26)	118 (20)	88 (21)	47 (16)	47 (18)	34 (14)	45 (18)
eGFRcys, ml/min per 1.73 m2	71 (17)	90 (28)	101 (24)	79 (24)	48 (18)	45 (21)	34 (13)	44 (20)
eGFRBTP, ml/min per 1.73 m2	66 (13)	64 (13)	82 (29)	62 (12)	49 (17)	46 (16)	33 (9)	45 (16)
eGFRB2M, ml/min per 1.73 m2	75 (15)	74 (21)	75 (16)	74 (18)	47 (14)	44 (16)	35 (12)	43 (15)
eGFRcr-cys, ml/min per 1.73 m2	78 (16)	91 (27)	113 (25)	84 (23)	46 (16)	45 (19)	33 (13)	44 (18)
eGFRavg, ml/min per 1.73 m2	73 (12)	79 (20)	94 (18)	76 (17)	47 (14)	46 (16)	34 (11)	44 (16)
ESRD
Number of events, n	252	114	76	442	246	801	586	1633
Follow-up time, yr	13 (3)	13 (5)	16 (6)	13 (4)	7.4 (3.4)	5.6 (2.2)	7.6 (5.5)	6.2 (3.2)
Incidence rate, per 1000-py	1.83	1.23	18.9	1.89	39.9	38.0	96.5	49.0
ACM
Number of events, n	2572	3307	113	5992	130	686	367	1183
Follow-up time, yr	13 (3)	15 (6)	16 (6)	14 (5)	7.4 (3.4)	6.2 (1.9)	14 (5)	7.5 (3.9)
Incidence rate, per 1000-py	18.8	31.0	28.0	24.2	21.1	29.3	33.0	29.1

Data are presented as mean (SD) unless otherwise indicated. Smoking status was defined as current, former, or never smoker on the basis of self-report. Prevalent cardiovascular disease at baseline was defined as a self-reported history of coronary artery disease or prior revascularization of blood vessels. Hypertension was defined as a systolic BP ≥140 mmHg, a diastolic BP ≥90 mmHg, or use of antihypertensive medication. All participants in Pima had diabetes. All participants in AASK did not have diabetes. For ARIC, diabetes was defined as a fasting blood glucose ≥126 mg/dl, a random blood glucose ≥200 mg/dl, or self-reported use of insulin or oral diabetes medication. For the MDRD Study, CRIC, and NHANES diabetes status was defined as a fasting blood glucose ≥126 mg/dl, a random blood glucose ≥200 mg/dl, HbA1c ≥6.5, or self-reported use of insulin or oral diabetes medication. GP/HR, general population/higher risk studies; ARIC, Atherosclerosis Risk in Communities study; NHANES, National Health and Nutrition Examination Study; Pima, longitudinal population-based cohort study in Pima Indians from the Gila River Indian Community; AASK, African-American Study of Kidney Disease and Hypertension; CRIC, Chronic Renal Insufficiency Cohort; MDRD, Modification of Diet in Renal Disease Study; BMI, body mass index; CHD, coronary heart disease; eGFR_cr, eGFR on the basis of creatinine; eGFR_cys, eGFR on the basis of cystatin C; eGFR_BTP, eGFR on the basis of β-trace protein; eGFR_B2M, eGFR on the basis of β-2-microglobulin; eGFR_cr-cys, eGFR on the basis of the combination of creatinine and cystatin C; eGFR_avg, eGFR on the basis of the average of eGFR_cr, eGFR_cys, eGFR_BTP and eGFR_B2M; py, patient year; ACM, all-cause mortality.

ESRD

Among the GP/HR cohorts, over a mean of 13 years, 442 participants (2.5%) developed ESRD. In the CKD cohorts, over a mean of 6.2 years, 1633 participants (30%) developed ESRD. Figure 1 compares the hazard ratios for eGFR and the other markers, and their average, across the range of GFR for the two cohorts, as well as the significant difference compared with eGFR_cr (symbols at bottom of the figure). For both the GP/HR and CKD cohorts, the hazard ratios for ESRD were greater at eGFR values below the reference point. At values above the reference point, there were minimally stronger associations for some of the markers in the GP/HR cohorts, but not in the CKD cohorts.

Figure 1.

Adjusted hazard ratios (HRs) of ESRD and all-cause mortality for general population and CKD cohorts. Adjusted HRs of ESRD (A and B) and all-cause mortality (C and D), for general population (A and C) and CKD cohorts (B and D) according to GFR when estimated using either creatinine (eGFR_cr), cystatin C (eGFR_cys), BTP (eGFR_BTP), B2M (eGFR_B2M), creatinine and cystatin (eGFR_cr-cys), or the average of eGFR from all four markers (eGFR_avg). Graphs show associations by plotting the adjusted HR versus the reference points, which are indicated by black diamonds (at 80 ml/min per 1.73 m² for the GP cohorts and at 50 ml/min per 1.73 m² for CKD) for piece-wise linear splines with a knot at 60 ml/min per 1.73 m² and at 90 ml/min per 1.73 m² (for GP cohorts only). The HRs were adjusted for age, sex, race, systolic BP, diabetes, c-reactive protein, smoking, cardiovascular disease, and eGFR from creatinine. On each figure, a solid dot indicates that the specific 1 ml/min per 1.73 m² has a statistically significant association with the outcome in adjusted analyses compared with the referent point. Lines at the bottom of each plot show the results for the seemingly unrelated regression which compares the results of models with the filtration marker versus model with eGFR_cr with a thick line indicating a stronger association at a significance level of P<0.05. Average 4, eGFR_avg; BTP, β-trace protein; B2M, β-2-microglobulin; Cr-Cys, combination of creatinine and cystatin C; GP, general population.

Table 2 shows the associations for eGFR categories for the novel filtration markers compared with eGFR_cr, eGFR_cys, and eGFR_cr-cys for eGFR categories <60 and between 60 and 90 ml/min per 1.73 m². For the GP/HR cohorts, eGFR_BTP, eGFR_B2M, and eGFR_avg had stronger associations for ESRD than eGFR_cr below 60 ml/min per 1.73 m², but not between 60 and 90 ml/min per 1.73 m². eGFR_BTP had a weaker association than eGFR_cr between 60 and 90 ml/min per 1.73 m². For the CKD cohorts, eGFR_BTP, eGFR_B2M, eGFR_cr-cys, and eGFR_avg had stronger associations than eGFR_cr below 60 ml/min per 1.73 m², whereas only eGFR_B2M and eGFR_avg had stronger associations than eGFR_cr between 60 and 90 ml/min per 1.73 m². Both eGFR_B2M and eGFR_avg had strong and linear associations throughout the range of eGFR. When eGFR_BTP and eGFR_B2M were compared with eGFR_cys or eGFR_cr-cys, the magnitude of improvement was similar to the comparison with eGFR_cr. Supplemental Table 3 shows the hazard ratios for eGFR categories. Overall, results were consistent within subgroups (Supplemental Table 4, A–E).

Table 2.

Comparison of associations (hazard ratios) of eGFR from novel filtration markers compared with eGFR from creatinine, cystatin, and creatinine–cystatin C across different eGFR ranges

Populations	Creatinine	Cystatin C	BTP	B2M	Cr-Cys	Average 4
ESRD GP
eGFR<60	3.36 (1.96 to 5.76)	3.17 (2.76 to 3.64)	5.27a,b,c (4.53 to 6.13)	5.26a,b,c (4.14 to 6.69)	3.25 (1.87 to 5.65)	3.98a,c (1.86 to 8.53)
eGFR 60–90	2.82 (1.72 to 4.62)	2.22 (1.65 to 3.00)	1.45a,b,c (0. 96 to 2.20)	1.68 (1.06 to 2.66)	2.88 (1.71 to 4.87)	2.99b,c (2.21 to 4.04)
ESRD CKD
eGFR<60	3.08 (2.36 to 4.03)	3.05 (2.33 to 3.99)	4.53a,b,c (3.29 to 6.23)	3.29a (2.34 to 4.61)	3.36a,b (2.56 to 4.43)	4.11a,b,c (2.87 to 5.87)
eGFR 60–90	1.32 (0.89 to 1.94)	1.34 (0.76 to 2.38)	1.11c (0.80 to 1.53)	5.14a,b,c (1.30 to 20.35)	1.44 (0.69 to 3.00)	3.14a,b,c (1.30 to 7.55)
ACM GP
eGFR<60	1.69 (1.38 to 2.06)	1.52 (1.36 to 1.69)	1.98b,c (1.63 to 2.41)	1.83b,c (1.53 to 2.19)	1.60 (1.39 to 1.83)	1.77b,c (1.52 to 2.05)
eGFR 60–90	1.13 (1.06 to 1.19)	1.20a (1.13 to 1.27)	0.93b,c (0.72 to 1.22)	1.28a,b,c (1.21 to 1.37)	1.21a (1.13 to 1.30)	1.26a,b,c (1.19 to 1.36)
ACM CKD
eGFR<60	1.31 (1.12 to 1.53)	1.67a (1.54 to 1.81)	1.69a,c (1.42 to 2.01)	1.79a,b,c (1.55 to 2.06)	1.48a,b (1.27 to 1.72)	1.67a,c (1.43 to 1.94)
eGFR 60–90	1.03 (0.75 to 1.41)	1.21 (0.94 to 1.56)	1.05 (0.84 to 1.31)	1.31 (0.49 to 3.50)	1.48 (0.82 to 2.66)	1.04 (0.28 to 3.86)

Hazard ratios and 95% confidence intervals are given as per 15 ml/min per 1.73 m² eGFR. BTP, β-trace protein; B2M, β-microglobulin; Cr-Cys, creatinine–cystatin C; Average 4, average of all four markers for GFR; GP, general population; ACM, all-cause mortality.

^aIndicates significantly different from eGFR from creatinine.

^bIndicates significantly different from eGFR from cystatin C.

^cIndicates significantly different from eGFR from creatinine–cystatin C.

As a measure of heterogeneity among studies, the I² statistics ranged from 0% to 87% in the GP/HR cohort and 0%–94% in the CKD cohorts. In the GP/HR cohorts, heterogeneity was because of stronger associations in ARIC at levels of GFR <60 ml/min per 1.73 m² than were observed in NHANES and Pima. In the CKD cohorts, heterogeneity was because of small differences among studies in precise estimates of associations with ESRD at GFR<60 ml/min per 1.73 m², without clinically relevant differences among the studies.

For the GP/HR cohorts, the addition of eGFR_B2M, and eGFR_cr-cys to this model significantly improved the c-statistic (Figure 2A, top panel). For the CKD cohorts, the addition of eGFR_BTP, eGFR_B2M, eGFR_cr-cys, and eGFR_avg significantly improved the c-statistic (Figure 2B, top panel). In the GP/HR cohorts, eGFR_cys, eGFR_B2M, eGFR_cr-cys, and eGFR_avg led to improved Net Reclassification Improvement (NRI) for risk categories beyond eGFR_cr alone, with eGFR_avg having the largest improvement (Figure 3A, top panel). In the CKD cohorts, only eGFR_B2M and eGFR_avg led to improved risk classification (Figure 3B, top panel). Addition of urine albumin/creatinine ratio (ACR) improved the overall c-statistic and lessened the improvement in c-statistic and NRI in models using novel filtration markers (Figure 2, A and B, bottom panel, Figure 3, A and B, bottom panel).

Figure 2.

Difference in c-statistics for models that include eGFR using novel filtration markers compared with models that include eGFR from creatinine or albuminuria. Top panel shows models that include eGFR from creatinine and bottom panel shows models that also include albuminuria. All models include age, sex, race, systolic BP, diabetes, c-reactive protein, smoking, cardiovascular disease, and eGFR from creatinine (top panel), and albuminuria (bottom panel). Top panel: c-statistics for these models for prediction of ESRD were 0.86 (95% confidence interval [95% CI], 0.79 to 0.94) and 0.82 (95% CI, 0.77 to 0.88) for general population (A) and CKD cohorts (B), respectively. c-statistics for all-cause mortality for prediction of all-cause mortality were 0.79 (95% CI, 0.73 to 0.86) and 0.73 (95% CI, 0.69 to 0.76) for general population (C) and CKD cohorts (D), respectively. Bottom panel: c-statistics for the models for prediction of ESRD were 0.90 and 0.86 for general population (A) and CKD cohorts (B), respectively. c-statistics for all-cause mortality for prediction of all-cause mortality were 0.80 and 0.74 for general population (C) and CKD cohorts (D), respectively. Average 4, average of all four markers for GFR; BTP, β-trace protein; B2M, β-2-microglobulin; Cr-Cys, combination of creatinine and cystatin C.

Figure 3.

Net reclassification index using eGFR from the novel filtration markers or their combination for ESRD and all-cause mortality. Net reclassification index for reclassification to higher or lower risk categories using eGFR from the novel filtration markers or their combination for ESRD (left) and all-cause mortality (right) for general population (top section of each panel) and CKD cohorts (bottom section of each panel). Top panel shows models without albuminuria and bottom panels shows models with albuminuria. Average 4, average of all four markers for GFR; BTP, β-trace protein; B2M, β-2-microglobulin; Cr-Cys, combination of creatinine and cystatin C.

All-Cause Mortality

Among the GP/HR cohorts, over a mean follow-up of 14 years, 5992 participants (33%) died. In the CKD cohorts, over a median of 7.5 years, 1183 participants (22%) died. For both the GP/HR and CKD cohorts, the risk of death was higher at eGFR values below the reference point for all markers and combinations (Figure 1, C and D). At values above the reference point, there appeared to be a higher risk for some markers in the GP/HR cohorts but not in the CKD cohorts (Figure 1, C and D).

For the GP/HR cohorts, eGFR_avg had stronger associations than eGFR_cr below 60 ml/min per 1.73 m², and eGFR_cys, eGFR_B2M, eGFR_cr-cys, and eGFR_avg had stronger associations than eGFR_cr between 60 and 90 ml/min per 1.73 m² (Table 2). For the CKD cohorts, eGFR_cys, eGFR_BTP, eGFR_B2M, eGFR_cr-cys, and eGFR_avg had stronger associations than eGFR_cr at GFR<60 ml/min per 1.73 m². When the new markers were compared with eGFR_cys or eGFR_cr-cys, the magnitude of improvement was similar to the comparison with eGFR_cr for the GP/HR cohorts, whereas for the CKD cohorts, there was less improvement. Supplemental Table 5 shows the hazard ratios for GFR categories. Overall, results were consistent within subgroups (Supplemental Table 4, A–E). I² statistics ranged from 0% to 80% in the GP/HR studies and 0%–82% in the CKD studies. In the GP/HR studies, heterogeneity was because of stronger but imprecise estimates of associations in the Pima study at levels of GFR<60 ml/min per 1.73 m². In the CKD studies, heterogeneity was observed only for B2M at a GFR of 60–90 ml/min per 1.73 m² because of highly imprecise estimates for AASK.

For the GP/HR cohorts, the addition of any filtration marker or their combination modestly but significantly improved the c-statistic (Figure 2C, top panel). For the CKD cohorts, the point estimates for the change in c-statistic were larger than observed for the GP/HR cohort for all markers and their combinations, but confidence intervals all crossed zero (Figure 2D, top panel). In multivariable models for 10-year risk prediction, in the GP/HR cohorts all of the markers led to improved NRI for risk categories beyond eGFR_cr alone (Figure 3C, top panel). Similarly, for 5-year risk prediction in the CKD cohorts, except for eGFR_cys all of the markers led to improved risk classification (Figure 2D, top panel). Addition of urine ACR improved the overall c-statistic for both cohorts, and lessened the improvement in c-statistic and NRI for models using novel filtration markers (Figure 2, A and B, bottom panel, Figure 3, A and B, bottom panel).

Discussion

In this individual patient meta-analysis, we provide a comprehensive description across GP/HR and CKD cohorts of the comparison of eGFR from two novel filtration markers, BTP and B2M, compared with creatinine in determining prognosis. Like eGFR_cys, we found that eGFR_BTP and eGFR_B2M had stronger risk associations and modestly improved prediction for ESRD and death beyond established risk factors and eGFR_cr. However, neither marker consistently showed improved associations or predictions compared with eGFR_cr, eGFR_cys, or eGFR_cr-cys across both outcomes and populations. eGFR on the basis of the average of the four filtration markers provided the most consistent improvement in associations and prediction across both outcomes and populations. The addition of albuminuria attenuated the improvement in risk prediction or risk classification with the novel filtration markers.

Creatinine is determined by muscle mass and dietary meat intake separately from the level of GFR. Because chronic illness often leads to low muscle mass and poor meat intake, confounding by these non-GFR determinants weakens the risk associations of eGFR_cr with adverse outcomes. It is well established that eGFR on the basis of cystatin C, which is less affected by muscle mass and meat intake, provides more accurate prognostic information than eGFR on the basis of creatinine. A large meta-analysis comparing eGFR from creatinine and cystatin C alone and in combination showed stronger associations for mortality, cardiovascular disease mortality, and ESRD in general population, high risk, and CKD cohorts, although the improvement for ESRD was limited to the high range of eGFR (6). However, cystatin C is more affected by fat mass and smoking (11,17–19); confounding by these non-GFR determinants of cystatin C is possibly the reason for the stronger associations between eGFR_cys and adverse outcomes compared with eGFR_cr (6,17,19).

Publications in individual studies have reported on the associations of BTP and B2M to ESRD and mortality (8–13). Conclusions were mixed as to the relative benefit of one novel filtration marker over another or compared with creatinine or cystatin C, likely because of some studies being underpowered to detect differences among the markers as well as variation in analytic methods among the studies (8–13). In this large individual patient meta-analysis and using consistent analytic methods, we were to show stronger risk associations between eGFR_BTP, eGFR_B2M, and their average, compared with eGFR_cr. We hypothesize that, like with cystatin, these stronger associations are because of confounding by factors associated with the non-GFR determinants of BTP and B2M rather than better estimation of measured GFR. In an analysis of the MDRD Study, AASK, and CRIC, we showed that BTP and B2M were also associated with higher urine protein, which would also likely strengthen associations with adverse outcomes (20). We also showed that GFR estimates based upon BTP and B2M provide less accurate estimates of measured GFR than eGFR_cr-cys in these three CKD studies (14). Consistent with the hypothesis that the stronger associations are because of non-GFR determinants of the markers, rather than better estimates of measured GFR, in prior reports from CRIC, Pima, and the MDRD Study, we showed that B2M or BTP were associated with ESRD and death from all causes beyond measured GFR (8,11,13).

The average of the eGFR from all four markers provides more consistent improvements on eGFR_cr across populations and outcomes with respect to both associations and prediction statistics. However, the improvement in prediction was modest. For clinical practice, these small changes would not affect decisions about care. On the basis of the data presented here, we would not recommend using these markers to replace eGFR_cr. Indeed, the addition of urine ACR to the model essentially eliminated the modest improvement for all markers, and this result reinforces the recommendation by Kidney Disease International Global Outcomes and recent studies that show that the combination of eGFR_cr and albuminuria provide accurate prognostic information for patients with CKD (21,22). For circumstances where urine is not available or there is uncertainty about the ACR results, the novel filtration markers could be used to provide additional information to that provided by eGFR_cr rather than ACR. In addition, although eGFR_cr provides important prognostic information, it is limited by the effects on its serum levels by muscle mass and diet that are separate from the effect of GFR. For patients with reduced muscle mass, it is possible that using the filtration markers other than creatinine to determine prognosis would provide more accurate results (23). This may be relevant for some research studies. For example, pharmaceutic companies evaluating interventions that affect the level of creatinine separately from GFR such as weight loss or dietary interventions may consider measuring these alternative markers.

Decisions to use these markers in practice must also consider availability, standards, and cost of the assay. Cystatin C and B2M are widely available and are relatively inexpensive. A key limitation to more widespread implementation and adoption of cystatin C is the substantial variability in the cystatin C assay despite the availability of an international reference material since 2010 (7). At present there is no reference material for B2M and there is no commercially available assay in the United States for BTP. Standardization of the measurement procedures by all in vitro diagnostic medical device manufacturers to reduce the current substantial interlaboratory variability in test results for these two analytes would need to occur before recommendation for implementation of these biomarkers for general use in research or clinical practice, but if done, could provide an advantage over cystatin C.

We had higher power than previous publications of individual studies to examine differences among the markers, especially for ESRD. We used complementary analytic methods to investigate the advantage of the novel markers for associations with outcomes as well as risk prediction. We were also able to better compare markers to each other and to the average of all four markers with the use of recently published GFR-estimating equations to express the filtration markers on the same scale, and we used standardized measurements of cystatin C and creatinine. However, there are several limitations. First, the BTP- and B2M-estimating equations were developed in CKD studies, and may underestimate the mean level of GFR in the three GP/HR studies. However, this bias likely does not affect the comparison among the risk associations and predictions. Second, despite the pooling of several studies, we had insufficient power to examine the results across most subgroups. Third, we made multiple comparisons among the markers and models. Fourth, for some ESRD models the hazards varied over time so the results should be interpreted as the average relative hazard.

In this comprehensive analysis of a large diverse population, we showed that eGFR_BTP, eGFR_B2M, and eGFR_avg were independent predictors of ESRD and death, with stronger associations than eGFR_cr, suggesting that novel markers contribute additional risk information beyond creatinine. The average of all four markers allows for the most consistent improved risk prediction beyond eGFR_cr across multiple outcomes and populations. However, improvements in prediction were modest, especially when albuminuria was included. Educational efforts focusing on increasing measurement of albuminuria in practice compared with current practice patterns would likely be more cost effective for obtaining accurate prognostic information for patients with and without CKD for ESRD and mortality.

Disclosures

L.A.I. reports funding to Tufts Medical Center for research and contracts with the National Institutes of Health, National Kidney Foundation, Pharmalink (Largo, FL), Gilead Sciences (Foster City, CA), Otsuka (Tokyo, Japan), and has a provisional patent (L.A.I., J.C., and A.S.L.) filed August 15, 2014 entitled “Precise estimation of glomerular filtration rate from multiple biomarkers” (no. PCT/US2015/044567). The technology is not licensed in whole or in part to any company. Tufts Medical Center, John Hopkins University, and Metabolon, Inc. (Durham, NC), have a collaboration agreement to develop a product to estimate GFR from a panel of markers.

A.S.L. reports funding to Tufts Medical Center for research and contracts with the National Institutes of Health, National Kidney Foundation, Amgen (Thousand Oaks, CA), Pharmalink, Gilead Sciences, and has a provisional patent (L.A.I., J.C., and A.S.L.) filed August 15, 2014 entitled “Precise estimation of glomerular filtration rate from multiple biomarkers” (no. PCT/US2015/044567). The technology is not licensed in whole or in part to any company. Tufts Medical Center, John Hopkins University, and Metabolon, Inc., have a collaboration agreement to develop a product to estimate GFR from a panel of markers.

J.C. reports funding to Johns Hopkins University for research and contracts with the National Institutes of Health, National Kidney Foundation, and has a provisional patent (L.A.I., J.C., and A.S.L.) filed August 15, 2014 entitled “Precise estimation of glomerular filtration rate from multiple biomarkers” (no. PCT/US2015/044567). The technology is not licensed in whole or in part to any company. Tufts Medical Center, John Hopkins University, and Metabolon, Inc., have a collaboration agreement to develop a product to estimate GFR from a panel of markers.

J.H.E. is a consultant to Gentian, Moss, Norway. Siemens Healthcare Diagnostics, Inc., Tarrytown, New York, has provided free or steeply discounted reagents for studies performed in J.H.E.’s research laboratory.