Study Population
Individual patient data were pooled from 2 community-based, longitudinal studies, ARIC and CHS, available as de-identified data from the US National Institutes of Health. ARIC recruited 15,792 subjects, ages 45 to 64 years, between 1987 and 1989. CHS included 5,201 subjects, 65 years and older, randomly selected from Medicare eligibility files during 1989 and 1990. In both studies, follow-up occurred at 3-4 year intervals; data from the initial and the first follow-up visits are used in this analysis. An additional 687 African American participants were recruited in CHS from 1992-1993; they were not included here due to limited follow-up. Further details of these studies are described elsewhere [14, 15].
Creatinine Calibration
In ARIC, serum creatinine was assessed in 15,582 (99%) subjects at their initial visit, while in CHS it was assessed in 5,716 (97%) subjects. We indirectly calibrated mean individual first visit creatinine values from ARIC and CHS to mean NHANES III for a given age, race and sex, following a fixed offset of -0.23 mg/dL (20 μmol/L) to calibrate to Cleveland Clinic values, resulting in adjustments of -0.24 mg/dL (21 μmol/L) in first visit ARIC values and -0.11 mg/dL (10 μmol/L) in first visit CHS values [16].
Because informative censoring from death and dropout results in a non population-based sample, second visit measurements cannot be calibrated to NHANES values in the same manner. In ARIC, second visit serum creatinine values were adjusted by -0.24 mg/dL (21 μmol/L) according to published data [17]. In CHS, the first visit for the African American cohort and the second visit for the original cohort were concurrent. As creatinine calibration is performed to account for assay differences and there should not be a difference in calibration factor by race, we indirectly calibrated the African American cohort to African American participants in NHANES III as described above. This calibration model showed that serum creatinine values were 0.04 mg/dL (3.5 μmol/L) greater in the CHS African-American cohort than NHANES III; accordingly, we subtracted this value from second visit measurements in the CHS cohort. Estimated GFR was calculated with the 4-variable Modification of Diet in Renal Disease (MDRD) Study equation [18].
Using these two eGFR values, participants were then classified into 4 groups: 1) eGFR < 60 mL/min per 1.73 m2 (eGFR < 60 mL/min per 1.73 m2 at both visits); 2) eGFR≥ 60 mL/min per 1.73 m2 (≥60 mL/min per 1.73 m2 at both visits); 3) eGFR increase ( < 60 mL/min per 1.73 m2 at first visit and ≥60 mL/min per 1.73 m2 at second visit); and 4) eGFR decline (≥60 mL/min per 1.73 m2 at first visit and < 60 mL/min per 1.73 m2 at second visit).
Baseline Covariates
Other baseline variables included demographics (age, sex, race, education status), lifestyle characteristics (smoking, alcohol intake), glycemic and antihypertensive medication use, past medical history (diabetes, hypertension and cardiovascular disease), examination findings (systolic and diastolic blood pressure, waist-to-hip ratio (WHR), electrocardiogram results); and blood laboratory variables (total cholesterol, high density lipoprotein (HDL) cholesterol, albumin, glucose). Second visit data were used in multivariable models for all variables except albumin where 1st visit data were used.
Race was defined as white or African American. Education level was dichotomized by high school graduation status. Cigarette smoking was stratified as never, former or current, and alcohol use was dichotomized by current use. Diabetes was defined by self-reported history, use of oral hypoglycemic agents or insulin, or fasting glucose ≥126 mg/dL (7 mmol/L). Hypertension was defined by systolic blood pressure ≥140 mm Hg, diastolic ≥90 mm Hg or use of antihypertensive medications. WHR was calculated by dividing waist circumference by hip circumference. Left ventricular hypertrophy (LVH) was defined by electrocardiographic criteria [19]. History of cardiovascular disease was defined by prior recognized or silent myocardial infarction, angina based on the Rose questionnaire, stroke, transient ischemic attack, intermittent claudication, and/or prior coronary angioplasty or bypass procedures.
Study Outcomes
The primary outcome was a composite of cardiac events (myocardial infarction, coronary revascularization or fatal coronary disease), stroke, or all-cause mortality. Secondary outcomes included individual components of the primary composite outcome. ARIC only identifies time to the first cardiac and stroke event and does not provide data on subsequent events. Therefore, those participants who had a cardiac event (n = 178) or a stroke (n = 62) between their first and second visits were defined as having a history of CVD but were excluded from analyses examining future cardiac or stroke outcomes; these 235 individuals (5 had both a cardiac event and stroke) were only included in analyses examining mortality.
Study Sample
From a pooled sample of 21,680 individuals, we excluded the African American cohort from CHS enrolled at the time of the second visit (n = 687). Of the remaining 20,993 participants, we excluded 156 who were missing age, sex or race data, 184 missing first visit creatinine, and 27 with first visit eGFR < 15 mL/min per 1.73 m2 to avoid inclusion of dialysis patients, yielding 20,626 eligible participants. Of these, 2,560 (12.4%) were missing eGFR at their second visit, with reasons including death prior to the expected follow-up time (n = 675), no reported laboratory results (n = 460), no data after the second visit (n = 1), and no second visit (n = 1,424; 1,078 from ARIC and 346 from CHS), yielding a final study population of 18,066 individuals used in univariate analyses (Figure 1). There were 17,698 participants with no missing covariates used in multivariable analyses.
Statistical Analysis
Second visit characteristics were compared with analysis of variance for continuous variables and chi-square tests for categorical variables. Proportions of individuals falling into each kidney function classification were calculated and stability of classifications was defined by remaining above or below 60 mL/min per 1.73m2. As a sensitivity analysis, among participants missing a second GFR estimate the minimal and maximal variation in stability of classification of sustained eGFR < 60 was estimated by assuming that a) eGFR remained below 60 mL/min per 1.73m2 from the first to the second visit and b) eGFR changed groups from below to above 60 mL/min per 1.73m2.
Event rates were calculated and Kaplan-Meier survival analysis was used to estimate the nonparametric survival distribution among study participants by eGFR group beginning at the time of the second GFR estimate. Cox proportional hazards regression utilized the SAS procedure 'TPHREG' with a class statement for eGFR group to examine differences in study outcomes among the respective comparison groups while adjusting for covariates. All models a priori included the following: age, sex, race, education, study of origin; smoking and drinking status; diabetes, hypertension, and cardiovascular disease history; systolic blood pressure, WHR, and LVH; and non-HDL cholesterol and albumin. In additional analyses evaluating models that revealed no significant differences in hazards for study outcomes between individuals with eGFR decline and eGFR increase, these two groups were combined and analyses repeated with a 3-level exposure term that also included sustained eGFR < 60 and sustained eGFR ≥60 mL/min per 1.73m2. The proportional hazards assumption was checked by testing the significance of the correlation coefficient between survival time for the composite outcome and the scaled Schoenfeld residuals using a chi-square statistic with a two-sided p-value and was met for all covariates.
Because prior research has found less consistent relationships between individuals with eGFR between 50 and 59 and adverse outcomes [3], we performed sensitivity analyses assessing study outcomes in individuals with eGFR sustained ≥60 mL/min per 1.73m2, individuals with eGFR sustained between 50 and 59 mL/min per 1.73m2, and individuals with eGFR sustained below 50 mL/min per 1.73m2. We also tested the effect of including the initial eGFR in multivariable models. Lastly, we performed a second series of analyses that duplicated the primary analyses but utilized eGFR calculated with the CKD-EPI estimating equation rather than the 4-variable MDRD equation after indirect calibration of serum creatinine from a non-IDMS to an IDMS standard [20].
All analyses were performed with SAS version 9.1. The Institutional Review Board at Tufts Medical Center approved this research.