Change in Measured GFR Versus eGFR and CKD Outcomes

Abstract

Measured GFR (mGFR) has long been considered the gold standard measure of kidney function, but recent studies have shown that mGFR is not consistently superior to eGFR in explaining CKD-related comorbidities. The associations between longitudinal changes in mGFR versus eGFR and adverse outcomes have not been examined. We analyzed a subset of 942 participants with CKD in the Chronic Renal Insufficiency Cohort Study who had at least two mGFRs and two eGFRs determined concurrently by iothalamate and creatinine (eGFR_cr) or cystatin C, respectively. We compared the associations between longitudinal changes in each measure of kidney function over 2 years and risks of ESRD, nonfatal cardiovascular events, and all-cause mortality using univariate Cox proportional hazards models. The associations for all outcomes except all-cause mortality associated most strongly with longitudinal decline in eGFR_cr. Every 5-ml/min per 1.73 m² decline in eGFR_cr over 2 years associated with 1.54 (95% confidence interval, 1.44 to 1.66; P<0.001) times higher risk of ESRD and 1.23 (95% confidence interval, 1.12 to 1.34; P<0.001) times higher risk for cardiovascular events. All-cause mortality did not associate with longitudinal decline in mGFR or eGFR. When analyzed by tertiles of renal function decline, mGFR did not outperform eGFR_cr in the association with any outcome. In conclusion, compared with declines in eGFR, declines in mGFR over a 2-year period, analyzed either as a continuous variable or in tertiles, did not consistently show enhanced association with risk of ESRD, cardiovascular events, or death.

Many studies in nephrology have focused on one-time measures of estimated or measured kidney function and their association with the diagnosis, complications, and prognosis of CKD.^1–3 However, in clinical practice, repeated measures of kidney function within the same individual are typically used to determine treatment effects and disease prognosis. A few recent studies have shown that changes in renal function using longitudinal measures of eGFR by either cystatin C (eGFR_cys) or creatinine (eGFR_cr) are strongly associated with risk of various CKD–related outcomes, including kidney failure, cardiovascular morbidity, or death.^4–8 However, a common limitation cited in these studies is the lack of availability of measured GFR (mGFR) as a gold standard.^4,5

More recently, there has been increased interest in the association between changes in GFR over time and CKD-relevant outcomes, especially in the context of potentially redefining surrogate end points in clinical trials of CKD interventions.^9,10 Although change in eGFR over time seems to be associated with risk of CKD-related outcomes,^4–6,8,11 we are unaware of any studies that have compared changes in mGFR versus eGFR in their associations with CKD-related outcomes. A few studies have compared the concordance between longitudinal measures of mGFR and eGFR using mGFR as the gold standard measure of renal function,^12–14 but these studies did not examine the association of change in renal function with outcomes of interest.

In this study, we examined whether repeated measures of measured GFR by iothalamate clearance (iGFR) are more strongly associated with CKD-related outcomes, such as ESRD, nonfatal cardiovascular disease, or death, compared with repeated measures of eGFR or creatinine clearance (CrCl) in the Chronic Renal Insufficiency Cohort (CRIC) Study. On the basis of prior work, we hypothesized that changes in iGFR would not consistently outperform changes in eGFR for the prediction of CKD-related outcomes.^1,15–19

Results

Baseline characteristics of 942 participants in the CRIC Study are shown in Table 1. Comparisons of baseline characteristics of the subcohort used in this study (n=942) versus the subsample of participants in the CRIC Study who had at least one GFR measurement (n=1214)¹⁶ and the entire CRIC Study cohort (n=3939) are provided in Supplemental Table 1.

Table 1.

Baseline demographic and laboratory characteristics of participants in the CRIC Study

Characteristic (n=942)	Mean±SD, N (%), or Median (IQR)
Mean age, yr	56.0 (±11.8)
Women	407 (43.2%)
Race
Non-Hispanic white	428 (45.4%)
Non-Hispanic black	334 (35.5%)
Hispanic	114 (12.1%)
Other	66 (7.0%)
Body mass index, kg/m2	31.3 (±6.6)
Diabetes	411 (43.6%)
Systolic BP, mmHg	126.2 (±20.3)
Diastolic BP, mmHg	72.3(±12.4)
History of cardiovascular disease	233 (24.7%)
CHF	47 (5.0%)
Peripheral vascular disease	42 (4.5%)
Stroke	72 (7.6%)
Myocardial infarction	146 (15.5%)
Median 24-h urine albumin, g/d	0.04 (IQR, 0.0–0.4)
Hemoglobin A1C, %	6.4 (±1.5)
LDL, mg/dl	102.9 (±35.2)
HDL, mg/dl	47.1 (±15.9)
Baseline renal function, ml/min per 1.73 m2
iGFR	51.1 (±19.5)
eGFRcr	46.4 (±13.2)
eGFRcys	59.2 (±22.4)
CrCl	66.2 (±31.7)
Mean (±SD) change in renal function between year 2 and baseline, ml/min per 1.73 m2 per 2 yrs
iGFR	−3.5 (±13.3)
eGFRcr	−3.5 (±8.9)
eGFRcys	−4.4 (±12.1)
CrCla	−4.3 (±27.7)
Median (IQR) change in renal function between year 2 and baseline, ml/min per 1.73 m2 per 2 yrs
iGFR	−3.4 (−10.4–3.1)
eGFRcr	−3.0 (−8.7–2.0)
eGFRcys	−4.3 (−11.0–1.9)
CrCla	−4.3 (−16.9–8.6)

IQR, interquartile range.

^an=942 for all GFR measures except for CrCl (n=839).

The mean change in GFR was similar (−3.5 ml/min per 1.73 m² over 2 years) by iGFR and eGFR_cr (Table 1), but the mean change in GFR was slightly larger using eGFR_cys and eGFR by CrCl (−4.4 and −4.3 ml/min per 1.73 m² over 2 years, respectively). The widest distribution in change in kidney function was by CrCl (SD of change =27.7 ml/min per 1.73 m² per 2 years) followed by iGFR (SD of change =13.3 ml/min per 1.73 m² per 2 years), eGFR_cys (SD of change =12.1 ml/min per 1.73 m² per 2 years), and eGFR_cr, which had the narrowest distribution of change in GFR (SD=8.9 ml/min per 1.73 m² per 2 years) (Table 1). The medians and interquartile ranges for the changes in each GFR metric are also provided in Table 1. When change in renal function was categorized into tertiles of each metric, the means of the fastest tertiles of decline were 16.4 ml/min per 1.73 m² per 2 years by iGFR, 12.9 ml/min per 1.73 m² per 2 years by eGFR_cr, 16.3 ml/min per 1.73 m² per 2 years by eGFR_cys, and 30.4 ml/min per 1.73 m² per 2 years by CrCl.

Figure 1 shows the correlations between iGFR and each of the eGFR measures. Change in eGFR_cr had the strongest correlation with change in iGFR (r=0.44; P<0.001), whereas change in CrCl had the weakest correlation with change in iGFR (r=0.15; P<0.001). Overall, the Pearson correlation was strongest between change in eGFR_cr and change in eGFR_cys (r=0.51; P<0.001). Changes in CrCl showed the weakest correlations with all other measures of kidney function.

Figure 1.

Correlation between different measures of renal function decline. Distributions and Pearson correlation coefficients for decline in each measure of GFR in a subcohort of the CRIC Study. All comparisons of the different GFR measures had Pearson correlation coefficients that met a threshold for statistical significance of P<0.001. All GFR changes are provided in ml/min per 1.73 m² per 2 years (n=942 for all GFR measures except for CrCl [n=839]).

After the second determination of renal function, there were 150 patients with ESRD (3.78 events per 100 person-years; mean follow-up =6.3 years), 130 patients with nonfatal cardiovascular events (3.33 events per 100 person-years; mean follow-up =6.0 years), 84 patients with congestive heart failure (CHF; 2.10 events per 100 person-years; mean follow-up =6.2 years), and 72 deaths (1.67 events per 100 person-years; mean follow-up =6.6 years).

Results of unadjusted Cox regression models comparing kidney function decline over a 2-year period using each of four measures of kidney function as a continuous predictor of CKD-related outcomes are shown in Table 2. Overall, the strengths of association between each measure of renal function decline and each clinical outcome varied, but iGFR did not consistently outperform other measures of kidney function decline. In terms of the magnitude of the hazard ratio, the associations for all outcomes were strongest for longitudinal decline in eGFR_cr. In terms of discrimination, the C statistic for iGFR was never the highest compared with the C statistic for other metrics of renal function for any of the outcomes of interest.

Table 2.

Unadjusted hazard ratios for clinical outcomes (ESRD, cardiovascular event composite, CHF, and all-cause mortality) associated with each measure of kidney function

Change over Time in Metric of Kidney Function	Cox Model Using Every 5-ml/min per 1.73 m2 Decrease over 2 yrs				Cox Model Comparing the Tertile of Greatest Renal Function Decline with the Other Two Tertiles
	iGFR	eGFRcr	eGFRcys	CrCl	iGFR	eGFRcr	eGFRcys	CrCl
ESRD
Hazard ratio (95% CI)	1.28a (1.20 to 1.36)	1.54a (1.44 to 1.66)	1.31a (1.23 to 1.40)	1.03a (1.00 to 1.05)	3.09a (2.23 to 4.26)	4.80a (3.42 to 6.72)	2.25a (1.63 to 3.10)	1.51a (1.06 to 2.14)
C statistic	0.72 (0.68 to 0.75)	0.76 (0.71 to 0.80)	0.70 (0.66 to 0.74)	0.60 (0.55 to 0.65)	0.65 (0.61 to 0.69)	0.70 (0.67 to 0.74)	0.61 (0.57 to 0.65)	0.56 (0.52 to 0.61)
Difference between C statisticsb	Reference	0.04 (0.001 to 0.08)	−0.02 (−0.06 to 0.02)	−0.12c (−0.18 to −0.07)	Reference	0.05c (0.01 to 0.10)	−0.04 (−0.08 to 0.003)	−0.09c (−0.14 to −0.04)
Cardiovascular event composite
Hazard ratio (95% CI)	1.13a (1.05 to 1.21)	1.23a (1.12 to 1.34)	1.13a (1.04 to 1.22)	1.00 (0.97 to 1.03)	1.58a (1.11 to 2.23)	1.96a (1.39 to 2.76)	1.25 (0.87 to 1.78)	0.88 (0.59 to 1.31)
C statistic	0.59 (0.53 to 0.64)	0.60 (0.54 to 0.65)	0.57 (0.52 to 0.63)	0.50 (0.46 to 0.54)	0.56 (0.51 to 0.60)	0.59 (0.54 to 0.63)	0.54 (0.49 to 0.58)	0.51 (0.48 to 0.54)
Difference between C statisticsb	Reference	0.01 (−0.04 to 0.07)	−0.01 (−0.06 to 0.04)	−0.09c (−0.15 to −0.02)	Reference	0.03 (−0.01 to 0.08)	−0.02 (−0.07 to 0.03)	−0.05 (−0.10 to 0.01)
CHF
Hazard ratio (95% CI)	1.13a (1.03 to 1.24)	1.27a (1.14 to 1.41)	1.15a (1.05 to 1.27)	0.99 (0.95 to 1.03)	1.60a (1.04 to 2.46)	2.17a (1.42 to 3.34)	1.11 (0.70 to 1.73)	1.02 (0.63 to 1.65)
C statistic	0.59 (0.52 to 0.65)	0.60 (0.53 to 0.68)	0.58 (0.51 to 0.65)	0.50 (0.44 to 0.56)	0.56 (0.50 to 0.62)	0.60 (0.54 to 0.65)	0.52 (0.48 to 0.57)	0.50 (0.46 to 0.54)
Difference between C statisticsb	Reference	0.02 (−0.05 to 0.08)	−0.01 (−0.07 to 0.05)	−0.09 (−0.17 to 0.003)	Reference	0.04 (−0.02 to 0.09)	−0.04 (−0.10 to 0.03)	−0.06 (−0.12 to 0.01)
All-cause mortality
Hazard ratio (95% CI)	1.08 (0.98 to 1.19)	1.10 (0.97 to 1.24)	1.01 (0.91 to 1.12)	1.00 (0.96 to 1.04)	1.66a (1.05 to 2.65)	2.00a (1.26 to 3.18)	1.32 (0.82 to 2.12)	0.88 (0.52 to 1.50)
C statistic	0.56 (0.49 to 0.63)	0.55 (0.48 to 0.63)	0.53 (0.46 to 0.59)	0.52 (0.45 to 0.60)	0.56 (0.50 to 0.62)	0.59 (0.52 to 0.65)	0.53 (0.48 to 0.57)	0.51 (0.47 to 0.55)
Difference between C statisticsb	Reference	−0.01 (−0.09 to 0.07)	−0.03 (−0.12 to 0.05)	−0.04 (−0.14 to 0.07)	Reference	0.02 (−0.04 to 0.08)	−0.04 (−0.10 to 0.03)	−0.05 (−0.13 to 0.02)

95% CI, 95% confidence interval.

^aHazard ratio is statistically significantly associated (P<0.05) with outcome.

^bAll C statistics are compared with C statistic for iGFR.

^cC statistic is statistically significantly different (P<0.05) compared with reference C statistic (iGFR).

For the outcome of ESRD, every 5-ml/min per 1.73 m² decline in eGFR_cr per 2-year period was associated with 1.54 (95% confidence interval, 1.44 to 1.66; P<0.001) times higher risk of ESRD. There was a statistically significant association between all metrics of renal function decline and ESRD in both continuous and categorical models. Of note, change in renal function by eGFR_cr had a statistically significantly higher C statistic than iGFR for the outcome of ESRD using both continuous and categorical models, whereas change in renal function by CrCl (continuous and categorical models) had a statistically significantly lower C statistic than change in renal function by iGFR (Table 2).

In the continuous models, all measures of renal function decline, except for CrCl, were statistically significantly associated with nonfatal cardiovascular events and CHF (Table 2). Results were similar by categorical models, except that the association between eGFR_cys and nonfatal cardiovascular events and CHF did not reach statistical significance. The size of the hazard ratio was again greatest for eGFR_cr in both continuous and categorical models for nonfatal cardiovascular events and CHF. For the outcome of cardiovascular events, CrCl did have a significantly lower C statistic compared with iGFR. No statistically significant differences in the C statistics were noted between any of the measures of renal function and the outcome of CHF, although the magnitude of the C statistic for decline in renal function by eGFR_cr remained the highest for CHF (c=0.60 in both continuous and categorical models).

All four measures of decline in renal function were not statistically significantly associated with all-cause mortality in continuous models, and no differences in the C statistics were noted, although confidence intervals were relatively wide (Table 2). In categorical models, only changes in iGFR and eGFR_cr were statistically significantly associated with all-cause mortality. Change in eGFR_cr had the highest C statistic (c=0.59) by tertile of renal function decline, although this was not statistically significantly different from the C statistic for iGFR (c=0.56).

Sensitivity analyses were performed using only the subcohort that had at least two CrCl measures (n=839) in all previous analyses, and results were similar (not shown). We also repeated our analyses after adjusting for demographics and comorbidities and derived similar conclusions: change in iGFR was not more strongly associated with future adverse events compared with changes in estimated measures of kidney function (Supplemental Table 2). Finally, we repeated our Cox models using the 2009 creatinine–based Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation and the 2012 creatinine– and cystatin–based CKD-EPI equations (Supplemental Table 3). The CKD-EPI 2009 model had a higher C statistic compared with the iGFR model in its association with risk of ESRD in the categorical model.

Discussion

In this study, we showed that longitudinal changes in iGFR, eGFR_cr, or eGFR_cys associated with risk of outcomes considered sequelae of worsening kidney function, such as ESRD, nonfatal cardiovascular events, and death. However, longitudinal changes in iGFR did not consistently outperform (and are occasionally inferior to) longitudinal changes in eGFR in the strengths of their associations with CKD-relevant outcomes, especially ESRD.

Most of the literature on measuring renal function has centered around one-time measures of GFR and their association with various adverse outcomes.^1,16,18,19 There have been fewer longitudinal studies of the association between repeated measures of GFR and their association with CKD-related outcomes. Recently, one paper reported that a 30% decline in eGFR is associated with increased risk of ESRD and death, although as expected, this association was stronger for ESRD than mortality.¹¹ A recent large meta–analysis also showed that declines in eGFR by 30% or 40% are strongly and incrementally associated with doubling of serum creatinine and ESRD, which was the traditional clinical trial outcome of CKD progression.²⁰ In other studies, change in eGFR has been shown to strongly associate with cardiovascular disease and mortality risk.^5,21 Our findings are, thus, largely consistent with these prior studies.

Only a few studies have compared the concordance between longitudinal changes in iGFR and longitudinal changes in eGFR using iGFR as the gold standard measure of renal function. For example, a secondary analysis of data from the Modification of Diet in Renal Disease (MDRD) Study showed that change in eGFR consistently underestimated change in iGFR, although no variable could be found to predict a systematic difference between the slopes.¹³ A recent study in type 1 diabetes compared longitudinal measures of iGFR with longitudinal measures of eGFR_cr, eGFR_cys, and eGFR_cr+cys and concluded that change in eGFR_cr and eGFR_cys most closely approximated change in iGFR, although the differences between mGFR and eGFR metrics were small.¹²

In this study, we conducted a head to head comparison of the performance of change in iGFR versus eGFR using clinical sequelae of worsening kidney disease (such as ESRD or death) as a fair umpire.²² The key finding of this study is that longitudinal change in iGFR is not consistently more strongly associated with sequelae of worsening kidney disease than longitudinal changes in eGFR. These conclusions were robust, even when earlier generations of estimating equations (such as the MDRD equation as opposed to the newer CKD-EPI equation) were used.

A core finding of our analysis is the marked differences in the SDs for the change in kidney function measures over time within individuals (27.7 ml/min per 1.73 m² per 2 years for CrCl, 13.3 ml/min per 1.73 m² per 2 years for iGFR, 12.1 ml/min per 1.73 m² per 2 years for eGFR_cys, and 8.9 ml/min per 1.73 m² per 2 years for eGFR_cr). Because we are comparing the same patients with themselves using four different GFR metrics, the underlying true change in kidney function should be the same for each person, and therefore, the widest SD for the change in renal function (CrCl) likely reflects greater measurement error. Thus, the metric with the narrowest SD for change in renal function (eGFR_cr) has greater precision.

Prior studies have shown that repeated measures of iGFR are limited in their precision with relatively large coefficients of variation (CVs).^16,23,24 The mean intertest CVs for repeated measurements of iGFR within relatively short periods of time (days to weeks) have ranged from 8.2% to 11.9%.^25,26 Potential reasons for this finding may include the fact that direct measurement of GFR using exogenous filtration markers is subject to considerable intraperson and diurnal variability,^27,28 including measurement error caused by poorly timed collections and inadequate voiding, among other reasons.^15,25 Fewer studies have reported the intertest CV for eGFR, but in one study, the intertest CV was 10.7% for eGFR_cys and 6.4% for eGFR_cr.²⁹ In comparison, the reported intertest CV for repeated 24-hour CrCl measures ranges from 11.3% to 15.7% in a few reports.^29–31 The SDs of the changes in GFR in our study are consistent with the current literature on intertest CVs for the various mGFR and eGFR metrics. The higher precision in GFR measurement using eGFR_cr may have led to stronger associations with the clinical outcomes of interest seen here. Although eGFR_cr may be more strongly associated with ESRD, because creatinine may play a role in the decision to initiate dialysis, the hazard ratios for change in eGFR_cr were also the largest for several other outcomes of interest, including nonfatal cardiovascular events and all-cause mortality.

We believe that, in the context of the published literature, our results highlight a number of important issues. First, no prior study has shown that one-time determination of iGFR is more strongly associated with future adverse events than eGFR.^1,17–19,32 We had previously shown in the CRIC Study that one-time measures of iGFR do not have stronger cross–sectional associations with concurrent metabolic consequences of decreased renal function (such as hyperphosphatemia or hyperkalemia) than eGFR.¹⁶ We now show that longitudinal measures of iGFR are also not more strongly associated with future clinical sequelae of decreased renal function than repeated measures of eGFR. Considering the laborious nature of measuring GFR and the invasiveness of such techniques, iGFR should ideally provide superior prediction of clinically relevant CKD outcomes compared with estimated measures of renal function. However, iGFR did not outperform eGFR in this study or others.^1,16–19,32

Second, outside of some very limited circumstances (as in patients with extremes of body composition), directly measuring GFR may have low utility, at least with currently available methodology, because these measurements are no better than eGFR for predicting CKD complications or prognosis. Among patients with very unusual body compositions, 24-hour urine collections to quantify CrCl have been proposed as an alternative way to assess renal function, but our data also argue against this approach, consistent with recent guidelines.³³ Specifically, we observed that 24-hour urinary CrCl correlated poorly with all other metrics of kidney function, and change in CrCl had the weakest associations (and lowest C statistics) for all end points.

Third, although the current Kidney Disease Improving Global Health 2012 guidelines suggest the use of cystatin C as a confirmatory test for CKD when creatinine-based measures of renal function may be questionable,³³ we did not find change in eGFR_cys to be more strongly associated with our outcomes of interest compared with change in eGFR_cr or iGFR.

Fourth, we believe that our findings support the use of eGFR measures as a primary methodology for studying the process of kidney function decline longitudinally.^4,5,34,35

Several issues with regards to the study design should be highlighted. We chose to use unadjusted Cox models as our primary analysis given that our main predictor, change in GFR, was determined as repeated measures within the same individual, and therefore, demographics and comorbidities remain the same, even without statistical adjustment. Strengths of this study include its relatively large size, its diversity in terms of demographics and inclusion of participants with and without diabetes mellitus, and its focus on clinically meaningful hard end points. The CRIC Study (a multicenter, national cohort) devoted considerable effort into maintaining quality control in the measurement of iGFR and rigorous adjudication of cardiovascular events.

One limitation of our study is the availability of only two measurements of renal function over a 2-year period. However, prior studies have suggested that a 2-year period is sufficient for observation of meaningful changes in GFR, and some studies have shown that even changes in GFR over a 1-year period are associated subsequently with both increased mortality and ESRD risk.^7,11 For example, in one Canadian study, a decline in renal function by ≥25% over 1 year associated with a 1.89 times higher risk of all-cause mortality.⁸ Using more than two measurements over 2 years would potentially have improved the precision in quantifying renal function trajectory. However, we did not repeat our analysis in patients who had at least three concurrent measures of iGFR and eGFR in the CRIC Study given the potential for informative censoring of study participants who dropped out because of poor outcomes or were unwilling to continue in the study. We also did not have repeated measures of renal function within short intervals in the CRIC Study (e.g., days) and are, therefore, unable to provide intertest CVs from this specific cohort.

Our results may not be generalizable to all patients with CKD, including those with polycystic kidney disease or rapidly progressive GN, because these were CRIC Study exclusion criteria. Furthermore, both cystatin C and creatinine values required calibration during the conduct of the CRIC Study, which may have introduced additional measurement errors beyond those already inherent in the laboratory assays themselves.²³

We also acknowledge that use of eGFR to track longitudinal changes in renal function may be limited by potential biases from non-GFR determinants of serum creatinine (such as interim amputation or other substantial changes in body composition or physiology). However, it is notable that, despite this, mGFR was not superior.

Lastly, our results may not necessarily extend to other methods of measuring GFR using alternative exogenous filtration markers (such as iohexol or inulin) or other clearance methods (such as plasma disappearance).^36,37

In conclusion, we showed that, in this prospective cohort of patients with CKD, compared with longitudinal changes in eGFR_cr or eGFR_cys, changes in iGFR were not more strongly associated with CKD-related outcomes, such as ESRD.

Concise Methods

Study Population

Data used for this analysis were derived from the baseline and year 2 visits of the CRIC Study. The CRIC Study is a National Institutes of Health–sponsored multicenter observational cohort that enrolled patients from seven clinical centers (13 recruitment sites) located throughout the United States.³⁸ Participants with eGFR at a screening visit between 20 and 70 ml/min per 1.73 m² on the basis of the MDRD equation were recruited for study between June of 2003 and September of 2008. The inclusion and exclusion criteria have been previously published.^38,39

A subsample (n=1214) of approximately one-third of the original CRIC Study enrollees (n=3939) underwent direct measurements of iGFR by urinary clearance. Exclusion criteria for this testing included known iodine allergy, pregnancy, and impaired urinary voiding. In this longitudinal study, 942 participants had two mGFRs and two eGFRs and were included for analysis, of which 839 participants also had two 24-hour urinary CrCl collections for analysis. We compared the baseline characteristics of the subcohort used in this analysis (n=942) with those who underwent at least one direct GFR measurement (n=1214) and the entire CRIC Study cohort.

Measures of Renal Function

We specified a priori that we would examine longitudinal changes in kidney function per every 5-ml/min per 1.73 m² decline over 2 years—which we deemed a clinically relevant change—and by tertiles of decline by each method. Change in GFR was calculated as the difference between the baseline and year 2 measures. We also specified a priori that we would conduct analyses dichotomizing changes of each measure of renal function decline at the fastest tertile to detect effect sizes that may be otherwise obscured by biologic variability.

Four longitudinal measures of renal function between the baseline and year 2 visits were compared. First, we examined change in GFR measured directly by urinary clearance of iothalamate (iGFR). Measurement of iGFR was conducted using a protocol similar to that in prior studies by trained GFR technologists.^24,40–43 Efforts were made in the CRIC Study to decrease variability in ¹²⁵I-iothalamate clearance results by performing all measurements after consumption of only a low-protein (<10 g) meal, in the supine or sitting position, and at approximately the same time of day. Nonsteroidal anti–inflammatory agents taken on an as-needed basis, including aspirin and ibuprofen, were withheld for at least 48 hours before each GFR test. The patients were instructed to drink 1 L water in the evening before and 5 ml/kg water on the morning of the test. Then, after five drops of a saturated solution of potassium iodide (190 mg iodide) were given, a water load of 10 ml/kg was required over the next 90 minutes. Subsequently, ¹²⁵I-iothalamate (140 µg) was injected subcutaneously at least 30 minutes after saturated solution of potassium iodide administration.^16,23 After a 60- to 90-minute waiting period, timed collections of urine and serum were performed, with a minimum of 30-minute waiting periods in between all voids. Urine flow rate was maintained at an absolute minimum of >1 ml/min, with a desirable urine flow rate of >3 ml/min, and if voids were <250 ml per time period, longer waiting periods ≤120 minutes were accepted in between timed collections. The goal was to obtain four timed urine collection periods bracketed by blood draws to measure plasma iothalamate levels. Concurrent urine counts and urine volumes for each period were determined. GFR was calculated as the time-weighted average of urine count × urine volume/plasma iothalamate and corrected for body surface area as performed similarly in other prior studies.⁴³

In the CRIC Study, 88% of subcohort enrollees had four or more urine collection periods, 6% had three, and 5% had two or fewer. It was noted that iGFR measures from the first period of the test were systematically higher than the time-weighted average of all testing periods, which could be explained by a lack of equilibrium reached in a substantial number of iGFR tests.²³ Therefore, the CRIC Study Steering Committee decided to eliminate the first period of the iGFR measure in all primary CRIC Study analyses involving iGFR data.^16,23 The mean intratest CV (defined as the SD across the individual testing periods within a given iGFR measurement divided by the average iGFR at the same visit) for the remaining timed urinary collections for the measurement of iGFR was 13.8% after excluding the first period. No adverse events were reported, and patients who were only able to provide one urine sample were excluded from the study (n=10). Measurements of iGFR were performed once at baseline and once at year 2 visits using the same technique, and samples were processed at the same GFR central laboratory at the Cleveland Clinic.

Second, we examined change in eGFR_cr calculated from the four–variable MDRD equation as follows: 175× serum creatinine^−1.154 ×age^−0.203 ×(0.742 if a woman) ×(1.212 if black).^44,45 Serum creatinine measurements were all calibrated initially to the Cleveland Clinic Research Laboratory and subsequently, isotope dilution mass spectrometry traceable standards.²³

Third, we examined change in eGFR_cys from the equation 127.7× cystatin C^−1.17 ×age^−0.13 ×(0.91 if a woman) ×(1.06 if black).⁴⁶ Cystatin C was measured using a particle–enhanced nephelometric immunoassay measured by a Siemens BNII Machine at the CRIC Study Central Laboratory with an intertest CV of 4.9%. Internal standardization was implemented to correct for drift over time when using different calibrator lots and reagent lots manufactured by Siemens. Specifically, all cystatin C values were calibrated to the combination of calibrator lot 049851 and reagent lot 167840.²³

Fourth, kidney function decline was examined using change in creatinine excretion determined from a 24-hour urine collection. Urine creatinine was determined spectrophotometrically by the Jaffe method (Roche Diagnostics, Indianapolis, IN). Samples were rejected, and recollection was attempted if total urine volumes were <500 ml or collection times were <22 hours or >26 hours. All CrCls were calculated in this paper as urine count × urine volume/plasma iothalamate standardized per 1.73 m² body surface area using standardized serum creatinine measurements.

In sensitivity analyses, we estimated change in GFR derived from the CKD-EPI 2009 (creatinine based) and 2012 (creatinine and cystatin C based) equations for comparison.^47,48

Outcomes Evaluated

We examined three clinical outcomes: ESRD, a composite of nonfatal cardiovascular events, and all-cause mortality. We included stroke, myocardial infarction, CHF, and peripheral arterial disease as a composite measure of nonfatal cardiovascular events. We also evaluated CHF alone as a separate outcome given the stronger association between reduced kidney function and heart failure than other cardiovascular events.³⁵ Ascertainment of cardiovascular outcomes in the CRIC Study has been previously described.⁴⁹ Only outcomes after the second kidney function assessment were included for analysis, and follow-up times for outcomes were counted after the second kidney function assessment occurred. Analyses for cardiovascular events or death were not censored after onset of ESRD.

Statistical Analyses

We determined Pearson correlations between changes in iGFR and changes in each of the three measures of eGFR. Cox proportional hazards models were used to examine the association between 5-ml/min per 1.73 m² decrements in each measure of GFR over 2 years and the outcomes of ESRD, a composite of cardiovascular events, CHF, and death in unadjusted models. Univariate models were used throughout given that the primary comparisons between renal function metrics were made within the same patients over time.

Harrell C statistic with bootstrapped confidence interval was used to evaluate the fit of each model using four measures of GFR, and differences between C statistics of each eGFR and the CrCl model were compared with the C statistic of the iGFR model.

In sensitivity analyses, we repeated our Cox models with each measure of change in GFR using only the subset of patients with at least two CrCl measures (n=839). In addition, we also repeated our univariate Cox models with adjustment for age, sex, race/ethnicity, systolic BP (as a continuous variable), total cholesterol (as a continuous variable), presence or absence of diabetes, history of any cardiovascular disease, and renal function at baseline (year 0).

Disclosures

None.