Variability in Estimated Glomerular Filtration Rate by Area under the Curve Predicts Renal Outcomes in Chronic Kidney Disease

Greater variability in renal function is associated with mortality in patients with chronic kidney disease (CKD). However, few studies have demonstrated the predictive value of renal function variability in relation to renal outcomes. This study investigates the predictive ability of different methods of determining estimated glomerular filtration rate (eGFR) variability for progression to renal replacement therapy (RRT) in CKD patients. This was a prospective observational study, which enrolled 1,862 CKD patients. The renal end point was defined as commencement of RRT. The variability in eGFR was measured by the area under the eGFR curve (AUC)%. A significant improvement in model prediction was based on the −2 log likelihood ratio statistic. During a median 28.7-month follow-up, there were 564 (30.3%) patients receiving RRT. In an adjusted Cox model, a smaller initial eGFR AUC%_12M (P < 0.001), a smaller peak eGFR AUC%_12M (P < 0.001), and a larger negative eGFR slope_12M (P < 0.001) were associated with a higher risk of renal end point. Two calculated formulas: initial eGFR AUC%_12M and eGFR slope_12M were the best predictors. Our results demonstrate that the greater eGFR variability by AUC% is associated with the higher risk of progression to RRT.


Introduction
Chronic kidney disease (CKD) is an increasing worldwide public health problem associated with increased morbidity and mortality [1,2]. Greater variability in renal function is associated with high mortality in CKD [3]. The mechanisms of variability in renal function are multifactorial, including intrinsic renal disease (e.g., renal microvascular disease or impaired autoregulatory mechanisms) and extrinsic factors (e.g., volume status and presence and severity of concurrent illness). However, only limited studies have demonstrated the predictive value of renal function variability in relation to renal outcomes in CKD patients.
Plotting slopes of 1/creatinine or estimated glomerular filtration rate (eGFR) versus time has been used as a tool for outcome variable in clinical studies [4]. eGFR slope has been identified as an independent risk factor for cardiovascular morbidity and mortality and progression to end-stage renal disease [5][6][7][8][9][10]. It is necessary to obtain the summarized information in the multivariate data by repeated measurement. Dynamic fluctuations in eGFR slope change may contribute to additional prognostic information beyond cross-sectional data. The area under the curve (AUC) computed with the trapezoidal formula is widely used as an approach to measure the dynamic and accumulating change of clinical laboratory parameters [11][12][13][14], but little is known about its application in the repeated measurements of eGFR variability. The eGFR AUC% formula by repeated measurements of eGFR was used as a measure of variability in renal function in our study. We hypothesize that eGFR AUC%, a method of estimating renal function variability, is a useful predictor for renal outcomes in CKD patients. Hence, the purposes of the present study are (1) to determine whether eGFR AUC% is associated with renal outcomes in progression to renal replacement therapy (RRT) and (2) to compare the predictive ability of renal outcomes using different methods of estimating renal function variability and eGFR slope in CKD stage 3-5 patients.  [15]. Patients who had only two serum creatinine measurements during follow-up ( = 413) or whose follow-up period was less than 12 months ( = 1048) were excluded. Besides, patients with CKD stages 1 and 2 ( = 152) were excluded. The final study population consisted of 1,862 CKD patients. Figure 1 showed the flowchart of the derivation of the cohort. Baseline variables included demographic features (age and sex), medical history (diabetes mellitus [DM], hypertension, and cardiovascular disease), examination findings (body mass index [BMI] and blood pressure), and laboratory data (albumin, fasting glucose, triglyceride, total cholesterol, hemoglobin, total calcium, phosphate, calcium-phosphorous product [Ca × P product], uric acid, and urine protein-tocreatinine ratio). DM and hypertension were defined by clinical diagnosis. Cardiovascular disease was defined as clinical diagnosis of heart failure, acute or chronic ischemic heart disease, and cerebrovascular disease. The laboratory data 3 months before and after enrollment in the CKD care system were averaged and analyzed. In addition, information of medications using angiotensin converting enzyme inhibitors (ACEI) and angiotensin II receptor blockers (ARB) during the study period was obtained from medical records.  [16]. Serum creatinine was measured by the compensated Jaffé (kinetic alkaline picrate) method in a Roche/Integra 400 Analyzer (Roche Diagnostics, Mannheim, Germany) using a calibrator traceable to isotope-dilution mass spectrometry [17].

Definition of Renal End
Point. The renal end point was defined as commencement of RRT. In patients reaching renal end point, renal function data were censored at the start of RRT. The other patients were followed until May 2010. The commencement of dialysis was determined according to the regulations by the National Health Service for dialysis therapy based on laboratory data, nutrition status, and uremic symptoms and signs.

Assessment of Rate of Renal Function
Decline and Variability. The rate of renal function decline was assessed by the slope of eGFR, defined as the regression coefficient between eGFR and time in units of mL/min/1.73 m 2 /year. At least three eGFR measurements were required to estimate eGFR slope. Faster renal function progression was reflected in a larger negative value of the slope. In addition, eGFR AUC% was used to estimate the variability in eGFR. Table 1 showed the specifics of the calculations and illustrates two different methods of estimating eGFR AUC%. A smaller eGFR AUC% indicates greater eGFR variability. Initial eGFR AUC% 12M was defined as initial eGFR for baseline value and estimated variability in eGFR during 12 months. Peak eGFR AUC% 12M was defined as peak eGFR for baseline value and estimated variability in eGFR during 12 months. Figure 2 represents two cases to illustrate the eGFR variability. Month (d) Peak eGFR AUC% 12M Figure 2: Two representative cases to illustrate the eGFR variability. In Case 1, the initial eGFR AUC% 12M (a) and peak eGFR AUC% 12M (b) were 85.0% and 68.9%, respectively. The eGFR slope of Case 1 was −1.06 mL/min/1.73 m 2 per month. In Case 2, the initial eGFR AUC% 12M (c) and peak eGFR AUC% 12M (d) were 64.8% and 51.3%, respectively. The eGFR slope of Case 2 was −1.51 mL/min/1.73 m 2 per month. Case 2 had greater eGFR variability than that of Case 1.
standard deviation, or median (25th-75th percentile) for triglyceride, urine protein-to-creatinine ratio, and days of follow-up. The number of all missing baseline data was less than 2% except the data of ACEI and/or ARB use. This adjustment of ACEI and/or ARB use was accomplished using multiple imputation approach. The differences between groups were checked by chi-square test for categorical variables, by independent t-test for continuous variables with approximately normal distribution, or by Mann-Whitney U test for continuous variables with skewed distribution. Cox proportional hazards analyses were used to investigate the relationships between different methods of estimating eGFR variability and eGFR slope with renal end point. The adjusted covariates included age, sex, a history of diabetes, hypertension, and cardiovascular disease, systolic and diastolic blood pressure, BMI, albumin, fasting glucose, triglyceride, total cholesterol, hemoglobin, eGFR, total calcium, phosphorous, CaXP product, uric acid, urine protein-to-creatinine ratio, ACEI and/or ARB use, and acute kidney injury episode. A significant improvement in model prediction was based on the −2 log likelihood ratio statistic, which followed a difference in likelihood ratio. The value was based on the incremental value compared with the basic model. Differences were considered significant if the value was less than 0.05.

Results
A total of 1,862 nondialyzed CKD patients were included. The mean age was 63.6 ± 13.4 years and there were 1,084 males and 778 females. The underlying etiologies of CKD in our patients included 646 with diabetic kidney disease (34.7%), 647 with nondiabetic glomerular diseases (34.7%), 200 cases of hypertension (10.7%), and 369 caused by other diseases (19.8%). The comparison of baseline characteristics between patients with and without renal end point was shown in Table 2. Compared with patients without renal end point, patients with renal end point were found to have a younger age, more female subjects, higher prevalence of DM and hypertension, higher systolic blood pressure, lower BMI, higher prevalence of advanced CKD stage, lower albumin, lower hemoglobin, lower baseline eGFR, lower calcium, higher phosphate, higher CaXP product, higher uric acid, higher urine protein-to-creatinine ratio, and higher prevalence of ACEI and/or ARB use. Besides, patients with renal end point had lower initial eGFR AUC% 12M, lower 4 The Scientific World Journal  Table 3.

Incremental Values in eGFR Variability and eGFR Slope for
Renal End Points. The incremental values in eGFR variability and eGFR slope, when they were added to the basic model to predict renal end point, were shown in Table 5. The addition of initial eGFR AUC% 12M ( < 0.001), peak eGFR AUC% 12M ( < 0.001), and eGFR slope 12M ( < 0.001) to the basic model significantly improved the predictive value of renal end points. The maximum change in the −2 log likelihood ratio was observed for initial eGFR AUC% 12M, followed by eGFR slope 12M, and peak eGFR AUC% 12M.
Besides, with regard to using standard error (SE) of the regression as a method of valuating eGFR variability, we found that the SE of the regression was not significantly associated with renal outcomes in multivariate analysis (HR, 0.926; 95% CI, 0.699 to 1.226; = 0.590).

Discussion
In the present study, we evaluated the association between eGFR variability and progression to RRT and compared the predictive ability of renal outcomes using different methods to assess eGFR variability in CKD patients. We found that smaller eGFR AUC% or larger negative eGFR slope during 12 month follow-up periods was associated with a higher risk of progression to RRT and provided additional predictive value. Two calculated formulas initial eGFR AUC% 12M and eGFR slope 12M were the best predictors.
The first important finding of our study is the identification of greater eGFR variability as a risk factor for adverse renal outcomes in CKD. Previous studies have identified eGFR slope as an independent risk factor for cardiovascular morbidity and mortality [5][6][7][8][9][10]. Al-Aly et al. [3] evaluated the impact of variability in renal function on mortality in a large sample size of CKD patients ( = 51,304). They used the coefficient of variation of the regression line coefficient fitted to all outpatient measures of eGFR to define variability in renal function and found that patients in the highest tertile of eGFR variability had an increased risk of death, independent of both baseline level of eGFR and eGFR slope. In addition, such variability provided further prognostic information beyond baseline eGFR and eGFR slope [3]. The reason for higher mortality in patients with greater variability in eGFR might be related to factors both intrinsic and extrinsic to the kidney and unmeasured exposures that may threaten renal homeostasis (e.g., changes in intravascular volume status, congestive heart failure, liver or pulmonary disease, or medications). In addition, hospital-acquired and frequent episodes of acute kidney injury, diuretics use, and cardiorenal syndrome all contribute to greater eGFR variability. Our present study revealed that greater eGFR variability, expressed as smaller eGFR AUC%, was associated with progression to RRT and provided additional predictive value. This implies that variability in renal function is a potentially more important consideration in deciding which patients will progress to RRT.
The second important finding of our study is that one calculated formula initial eGFR AUC% 12M was a better predictor of adverse renal outcomes than eGFR slope. Unlike other summary measures (e.g., baseline, average, or maximum eGFR value), both pieces of information can be captured through the use of AUC [18]. The use of AUC simplifies the statistical analyses by transforming the multivariate data into a univariate variable, especially when many repeated measurement numbers exist and there is a need to summarize the information [14]. In addition, when the time interval between repeated measurements is not identical, the use of AUC provides an alternative approach to adjust for these differences [19]. O'Hare et al. investigated the variability in eGFR in the 2 years before initiation of dialysis in 5,606 Veterans Affairs patients and found that patients with greater eGFR variability were less likely to have received predialysis care and had a higher risk of death in the first year after dialysis [20]. An understanding of kidney function variability preceding dialysis can help guide clinical decision making (e.g., nephrology referral, vascular assessment, and transplant referral), goals of care, and anticipated service needs [20][21][22]. The formula for eGFR AUC% may also be used in long term follow-up in patients with CKD.
As for other approaches, such as using SE of the regression as a method of evaluating eGFR variability, our result is in agreement with Perkins RM's finding that eGFR variability by the SE did not predict ESRD outcome among CKD stage 3 patients [23]. Despite this finding, it needs further 6 The Scientific World Journal  Values expressed as hazard ratio (HR) and 95% confidence interval (CI).
Covariates in the multivariate model included age, sex, a history of diabetes, hypertension, and cardiovascular disease, systolic and diastolic blood pressure, body mass index, albumin, fasting glucose, triglyceride, total cholesterol, hemoglobin, eGFR, total calcium, phosphorous, CaXP product, uric acid, urine protein-to-creatinine ratio, ACEI and/or ARB use, and acute kidney injury episode. value was based on the incremental value compared with the previous model which was adjusted for age, sex, a history of diabetes, hypertension, and cardiovascular disease, systolic and diastolic blood pressure, body mass index, albumin, fasting glucose, triglyceride, total cholesterol, hemoglobin, eGFR, total calcium, phosphorous, CaXP product, uric acid, urine proteinto-creatinine ratio, ACEI and/or ARB use, and acute kidney injury episode.
investigations to compare the significance of employing the different approaches for estimating eGFR variability.
In conclusion, our results demonstrate that the greater eGFR variability by AUC% is associated with the higher risk of progression to RRT. The formula of estimating eGFR variability by AUC method may provide a significant predictive value and guide the treatment strategies and choices in CKD patients.

8
The Scientific World Journal