Change in cardiac biomarkers and risk of incident heart failure and atrial fibrillation in chronic kidney disease: the CRIC Study

Abstract

Rationale and Objective:

Circulating cardiac biomarkers may signal potential mechanistic pathways involved in heart failure (HF) and atrial fibrillation (AF). Single measures of circulating cardiac biomarkers are strongly associated with incident HF and AF in chronic kidney disease (CKD). We tested the associations of longitudinal changes in N-terminal pro-B-type natriuretic peptide (NT-proBNP), high sensitivity troponin T (hsTnT), galectin-3, growth differentiation factor-15 (GDF-15), and soluble ST-2 (sST-2) with incident HF and AF in patients with CKD.

Study design:

Observational, case-cohort study design.

Setting and Participants:

Adults with CKD enrolled in the CRIC study.

Exposures:

Biomarkers were measured at baseline and 2 years later among those without ESRD. We created three categories of absolute change in each biomarker: the lowest quartile, the middle two quartiles and the top quartile.

Outcomes:

The primary outcomes were incident HF and AF.

Analytic approach:

Cox proportional hazards regression models were used to test the associations of the change categories of each cardiac biomarker with each outcome (the middle two quartiles of change as the referent group), adjusting for potential confounders and baseline concentrations of each biomarker.

Results:

The incident HF analysis included 789 (which included 138 incident HF cases) and the incident AF analysis included 774 participants (which included 123 incident AF cases). In multivariable models, top quartile of NT-proBNP change (>232 pg/mL over 2 years) was associated with increased risk of incident HF (HR 1.79, 95% CI: 1.06, 3.04) and AF (HR 2.32, 95% CI: 1.37, 3.93) compared with the referent group. Participants in the top quartile of sST2 change (>3.37 ng/ml over 2 years) had significantly greater risk of incident HF (HR 1.89, 95% CI: 1.13, 3.16), whereas those in the bottom quartile (≤ −3.78 ng/ml over 2 years) had greater risk of incident AF (HR 2.43, 95% CI: 1.39, 4.22) compared with the two middle quartiles. There was no association of changes in hsTnT, galectin-3 or GDF-15 with incident HF or AF.

Limitations:

observational study

Conclusions:

In CKD, increases in NT-proBNP were significantly associated with greater risk of incident HF and AF; and increases in sST2 were associated with HF. Further studies should investigate whether these markers of subclinical cardiovascular disease can be modified to reduce the risk of cardiovascular disease in CKD.

PLAIN LANGUAGE SUMMARY

Patients with chronic kidney disease are at higher risk of developing heart failure and atrial fibrillation for reasons that are not completely understood. Cardiac biomarkers measured in the blood may provide insight into possible mechanisms that contribute to the higher risk of heart failure and atrial fibrillation. In this study we examined changes in cardiac biomarkers over time and their association with development of heart failure and atrial fibrillation. We found that increases over two years in two of these cardiac biomarkers (NT-proBNP and sST2) were significantly associated with higher risk of heart failure; and increases in sST2 were associated with higher risk of atrial fibrillation.

INTRODUCTION

Heart failure (HF) and atrial fibrillation (AF) are leading causes of cardiovascular disease among patients with chronic kidney disease (CKD) and are associated with greater risks of death and other poor clinical outcomes.^1–9 Previous work by our group and others has shown that higher concentrations of cardiac biomarkers-N-terminal pro-B-type natriuretic peptide (NT-proBNP), high sensitivity troponin T (hsTnT), galectin-3, growth differentiation factor-15 (GDF-15), and soluble ST-2 (sST-2)-- are associated with increased risk of incident HF and AF and could signal potential mechanistic pathways that contribute to HF in CKD patients.^10–40 More specifically, NT-proBNP is secreted from cardiac myocytes in response to myocardial stretch from pressure or volume overload^41–44 Concentrations of hsTnT rise in response to myocardial injury or myocardial remodeling.^{45, 46} Galectin-3 belongs to the β-galactoside-binding protein family and is both proinflammatory and profibrotic in cardiomyocytes.^{47, 48} GDF-15 is a member of the transforming growth factor-ß (TGF- ß) cytokine family^{49, 50} and plays a role in cardiomyocyte repair.⁵¹ ST-2 is a member of the interleukin-1 (IL-1) receptor family. It has two forms, soluble ST2 (sST-2) and transmembrane ST-2 (ST-2L). ST-2 is a marker of cardiac stress that is upregulated with myocyte stretch.

Prior studies have observed associations of one-time measures of these cardiac biomarkers with HF and AF incidence. In select non-CKD populations such as those with prevalent HF or cohorts of elders, studies have suggested that trajectories of cardiac biomarkers, increases or decreases, may also be associated with incident HF or AF; thus repeated biomarkers may potentially provide prognostic information beyond single baseline measures.^{15, 52–60} However, it is not known whether changes in cardiac biomarkers indicate dynamic risk for developing incident HF and AF in persons with CKD. If potential associations are identified, then they might identify opportunities for risk surveillance and for targeted primary prevention of HF and AF. Therefore, we performed a prospective study of a well-characterized CKD cohort to describe changes in five cardiac biomarkers over 2 years, and to test for independent associations of these changes in each cardiac biomarker with risk of incident HF and AF.

METHODS

Study population

We studied adults with CKD at entry in the Chronic Renal Insufficiency Cohort (CRIC) Study. A total of 3,939 participants were enrolled into the CRIC study between June 2003 and August 2008 at seven clinical centers across the United States (Ann Arbor/Detroit, MI; Baltimore, MD; Chicago, IL; Cleveland, OH; New Orleans, LA; Philadelphia, PA; and Oakland, CA). Details on study design and baseline characteristics of the participants were previously published.^{61, 62} All study participants provided written informed consent, and the study protocol was approved by institutional review boards at each of the participating sites. Inclusion and exclusion criteria have been previously described.⁶¹

The present analysis was an ancillary study to the main CRIC study in which we conducted a case-cohort study designed to evaluate the association of change in cardiac biomarkers measured at baseline and year 2 with development of incident HF and AF after year 2 among participants without ESRD. We randomly sampled and measured cardiac biomarkers in a subcohort of N=700 participants at baseline, after excluding N = 144 with prevalent HF or AF either at study entry (by self report) or development of HF or AF between baseline and year 2 (by self report, ECG, or occurrence of an adjudicated HF or AF hospitalization) and N = 34 who developed ESRD by Year 2 (the subcohort size was determined based on power for the analyses and available resources).In addition, we measured cardiac biomarkers in 135 additional participants who developed incident HF or AF after Year 2, but were free of ESRD (added cases). Therefore, for the incident HF analysis, we included 700 participants in the subcohort (of whom N=49 developed incident HF) and an additional N=89 HF cases, for a total study population for 789 (including 138 incident HF cases). For the incident AF analysis, we included 700 participants in the subcohort (of whom N=49 developed incident AF) and an additional N=74 AF cases, for a total study population of 774 (including 123 incident AF cases).

Cardiac biomarkers

GDF-15, sST2 and galectin-3 were measured from EDTA plasma stored at 70°C from samples at baseline and at year 2 in a random subcohort in batch at the University of Pennsylvania CRIC Central Laboratory. All assays were measured in duplicate. GDF-15 and sST2 were measured using ELISA (R&D Systems). For GDF-15, the quantitative range was 23.4-1500 pg/mL with a lower limit of detection of <2.0 pg/mL. At a concentration of 98.8 pg/mL, the CV was 7.2%; and at a concentration of 624 pg/mL was 4.5%. For sST2, the quantitative range was 0.63-40 ng/mL with a limit of detection of <0.1 ng/mL. At a concentration of 2.6 ng/mL, the CV was 11.2% and at 0.94 ng/mL was 8.5%. For galectin-3, the quantitative range was 0.31-10.0 ng/mL with a lower limit of detection of <0.016 ng/mL. At an average concentration of 0.78 ng/mL, the CV was 4.0% and at 5.36 ng/mL was 4.2%.

hsTnT and NT-proBNP were measured at baseline in 2008 from EDTA plasma stored at −70°C, both using chemiluminescent microparticle immunoassay (www.roche-diagnostics.us) on the Elecsys 2010 at the University of Maryland. hs-cTnT was measured using a high sensitivity assay with a range of values from 3 to 10,000 ng/L.⁶³ The limit of blank was 3ng/L and limit of detection was <5 ng/L. For hsTnT, the coefficient of variation (CV) was 3% at a concentration of 30 ng/L and 5.8% at 2,213ng/L. The value at the 99th percentile cutoff from a healthy reference population was 13 ng/L for hsTnT with a 10% CV.⁶³ The range of values for NT-proBNP, was from 114 to 5900 ng/L and the CV was 4.25% at a concentration of 132 ng/L and 5.3% at 4,640 ng/L.

In 2017, we added year 2 measures of NT-proBNP and hsTnT and remeasured a subset of baseline samples at the same testing laboratory to calibrate the measures. The new measurements in 2017 were performed on the Roche E601. We remeasured NT-proBNP in 100 random samples from baseline and all of the year 2 samples. We developed and applied a Deming regression ⁶⁴ to calibrate the 2008 baseline NT-proBNP measures with the 2017 NT-proBNP measures. The goodness of fit with this calibration was an R² of 0.991507 for NT-proBNP. The residual plots for the original and recalibrated measures were similar (Supplemental Figure 1).

Similarly, for hsTnT, we remeasured any baseline hsTnT measure with a value <5ng/L using the newer Roche E601 instrument, which had a limit of blank of 2.5 ng/L and limit of detection of <3 ng/L. At a concentration of 13.5 ng/L, the CV was 1.9% and at 4,831 ng/L the CV was 0.8% with the E601. We also measured a random subset of 100 samples at baseline and all samples at the year 2 visit. We developed and applied a Deming regression (similar methods as used for NT-proBNP) to calibrate the 2008 baseline hsTnT measures with the 2017 hsTnT measures. The goodness of fit with this calibration was an R² of 0.970514 for htTnT. The residual plots for the original and recalibrated measures were similar (Supplemental Figure 2).

Incident heart failure

The primary outcome was incident HF that occurred between year 2 of study entry through May 2014 (the last date that HF events were adjudicated). HF was identified by asking study participants biannually if they were hospitalized and by reviewing electronic health records from selected hospitals or health care delivery systems. The first 30 discharge codes were identified for all hospitalizations, and codes relevant to HF resulted in retrieval of medical records by study personnel for centralized adjudicated review. At least two study physicians reviewed all possible HF events and deaths using medical records and guidelines on clinical symptoms, radiographic evidence of pulmonary congestion, physical examination of the heart and lungs and, when available, central venous hemodynamic monitoring data, and echocardiographic imaging. HF was confirmed when both reviewers agreed upon a “probable” or “definite” occurrence of HF based on modified clinical Framingham criteria.⁶⁵ Deaths were identified from report from next of kin, retrieval of death certificates or obituaries, review of hospital records, and through the Social Security Death Master File and National Death Index.

Incident atrial fibrillation

Incident AF was defined as a hospitalization for AF and confirmed by dual physician adjudication (19). Every six months, participants were asked if they had visited an emergency department or had been hospitalized. Medical records from corresponding hospitals or healthcare systems were queried for qualifying encounters. Diagnostic codes for AF (ICD-9-CM 427.31 or 427.32) prompted retrieval of medical records and centralized review for the ascertainment of incident AF. Final adjudication of events was done after at least two study physicians reviewed all possible AF events by manual review of relevant medical records. Hospitalized ECGs (when available), were reviewed and were part of the adjudication process.

Covariates

At the baseline visit and each annual visit, participants provided information on their sociodemographic characteristics, medical history, medication usage, and lifestyle behaviors. Race/ethnicity was categorized as non-Hispanic white, non-Hispanic black, Hispanic, and other. History of cardiovascular disease was determined by self-report. Current tobacco use was determined by self-report. Diabetes mellitus was defined as a fasting glucose >126 mg/dL, a non-fasting glucose >200 mg/dL, or use of insulin or other antidiabetic medication. Anthropometric measurements and blood pressure (BP) were assessed using standard protocols.⁶⁶ Body mass index (BMI) was derived as weight in kg divided by height in meters squared. Serum creatinine was measured using an enzymatic method on an Ortho Vitros 950 at the CRIC Central Laboratory and standardized to isotope dilution mass spectrometry-traceable values^{67, 68} and the CKD-EPI equation was used to estimate GFR.⁶⁹ Additional assays measured included serum phosphorus, 24-hour urine total protein, glucose, LDL cholesterol, HDL cholesterol, and total PTH. The aforementioned assays were performed at the central laboratory with the exception of PTH (measured at Scantibodies Laboratory Inc.).

Statistical approach

There are no established cut-offs to define clinically meaningful changes in each of the cardiac biomarkers of interest. Therefore, to define change, we subtracted the concentration at Year 2 from the baseline concentration; therefore, a negative number represented a longitudinal decline in the biomarker level while a positive number represented an increase in the biomarker level. We then examined the distribution of absolute change from baseline to year 2 of each cardiac biomarker among the subcohort and created three categories of change: the lowest quartile of absolute change in each biomarker, the middle two quartiles of absolute change and the top quartile of absolute change in each biomarker. The middle two quartiles were designated as the referent group as these were likely the participants with the most stable levels of each biomarker over 2 years.

We examined characteristics of the study population in the subcohort overall and across these categories of absolute change in each biomarker. Among the subcohort, we used proportional odds regression to test the association of participant characteristics with odds of a higher category of absolute change for each cardiac biomarker.

To account for the case-cohort study design, we used Cox models with robust variance estimation to examine the association of change of cardiac biomarkers (modeled in the categories described above) with incident HF and AF, using the Prentice weighting method.^70–72 Participants were considered at risk from the date of the year 2 visit until the first occurrence of definite or probable HF or AF, or until they were censored due to death, dropout, progression to ESRD or loss of follow-up. We performed unadjusted and adjusted Cox models, adjusting for covariates at year 2 including: demographics (age, sex, race, site), diabetes, cardiovascular disease (CVD, defined as history of myocardial infarction, stroke and heart failure), smoking, measures of kidney function (eGFR and 24 hour urine protein), systolic blood pressure, LDL and HDL cholesterol, and baseline levels of the biomarker.

We performed several sensitivity analyses. In the first, we adjusted for the change in eGFR between baseline and year 2 to evaluate whether concurrent change in kidney function may have accounted for some of the observed associations between change in cardiac biomarkers and risk of HF and AF. In the second, we adjusted for use of ACEi/ARBs, diuretics and β-blockers at baseline and year 2 to also evaluate whether differential use of cardiovascular medications, which may affect cardiovascular structure/function and circulating levels of the biomarkers, was a possible confounder in the observed associations. In the third sensitivity analysis, to account for possible effects of interval cardiovascular events on the observed associations, we adjusted for interim, time-updated adjudicated hospitalizations for AF and myocardial infarction (MI) for the HF outcome; and for HF and MI hospitalizations for the AF outcome. In the fourth sensitivity analysis, we examined the association of relative changes in each cardiac biomarker over 2 years with subsequent HF and AF. Similar to the approach for the primary analysis, we examined the distribution of relative change from baseline to year 2 of each cardiac biomarker among the subcohort and created three categories of relative change: the lowest quartile of relative change in each biomarker, the middle two quartiles of relative change and the top quartile of relative change in each biomarker. The middle two quartiles of relative change were designated as the referent group as these were likely the participants with the most stable levels of each biomarker over 2 years.

In a secondary analysis, we also modeled the association of baseline levels of the biomarker with risk of incident HF and AF. In another secondary analysis, we examined the ability of each cardiac biomarker to predict HF and AF, and evaluated the discriminatory ability via the 10-fold cross-validated Harrell’s C-index and the integrated discrimination index (IDI) with accompanying 95% confidence intervals.^{73, 74} We compared a several models to predict AF or HF: baseline cardiac biomarker only; baseline cardiac biomarker + clinical model (age, sex, race/ethnicity, site, diabetes, CVD, smoking, log-transformed 24hr urine protein, eGFR, SBP, BMI, LDL, HDL); and baseline cardiac biomarker + clinical model + change in cardiac biomarker.

There were small (<3%) amounts of missingness in the variables used in these sensitivity analyses; we accounted for this using multiple imputation with chained equations, combining the resulting estimates with Rubin’s rules to account for the variability in the imputation procedure.^{75, 76}

A nominal p-value of < 0.05 was taken as evidence of statistical significance in all analyses. All analyses were conducted using the R 3.4.3 computing environment (R Foundation for Statistical Computing, Vienna, Austria).

RESULTS

Change in cardiac biomarkers over two years in study population

Among participants in the subcohort, mean age was 59 years, 44% were women and 36% were black. Mean eGFR was 47 ml/min/1.73 m2 and 43% had a history of diabetes (Table 1). Among the subcohort, the median absolute change in NT-proBNP was 50.0 [IQR −11.9, 190.7] pg/ml; in hsTnT was 2.8 [−0.2, 9.4] ng/mL, in GDF-15 was 110.5 [−96.0, 523.5] pg/mL, in galectin-3 2.8 [−0.5, 6.4] ng/mL and sST2 was −0.5 [−3.7, 3.3] ng/mL (Supplemental Figure 3 and Table 1)). Median relative change in NT-proBNP was 85.0 [IQR −10, −330]%; in hsTnT was 23[−2, −70]%, in GDF-15 was 11 [−9, 41]%, in galectin-3 22 [−4, 52] % and sST2 was −4 [−24, 27]%. Participants with the greatest absolute elevation in NT-proBNP had lower baseline eGFR, were more likely have diabetes and higher baseline systolic blood pressure; and less likely to have used diuretics at baseline (Supplemental Table 1). The participants with the greatest absolute increase in hsTnT were more likely to be men, more likely to be black, to have diabetes, to have atrial fibrillation, and to have higher systolic blood pressure and lower eGFR (Supplemental Table 2). Participants with the greatest absolute increase in GDF-15 over 2 years were more likely to be black, have diabetes, smoke, and have higher systolic blood pressure and lower baseline eGFR (Supplemental Table 3). Participants with the greatest change in galectin-3 over 2 years were more likely to have diabetes, smoke, higher systolic blood pressure and lower eGFR at baseline (Supplemental Table 4). Participants with the greatest change in sST2 over 2 years were more likely to be black, have higher systolic blood pressure and have a greater decline in eGFR over 2 years (Supplemental Table 5).

Table 1.

Demographics of subcohort (N = 675)

N	N= 675
Baseline NT-proBNP, median (IQR) (pg/mL)	96 (37-216)
Absolute change in NT-proBNP, median (IQR) (pg/mL)	50 (−12 – 191)
Relative change in NT-proBNP, median (IQR) (%)	85 (−10 – 330)
Baseline hsTNT, median (IQR) (ng/mL)	12.4 (7.9-19.8)
Absolute change in hsTNT, median (IQR) (ng/mL)	2.8 (−0.2 – 9.4)
Relative change in hsTNT, median (IQR) (%)	23 (−2 – 70)
Baseline GDF-15, median (IQR) (pg/mL)	1318 (906-1874)
Absolute change in GDF-15, median (IQR) (pg/mL)	110 (−96 – 524)
Relative change in GDF-15, median (IQR) (%)	11 (−9 – 41)
Baseline Galectin-3, median (IQR) (ng/mL)	13.5 (9.7-17.9)
Absolute change in Galectin-3, median (IQR) (ng/mL)	2.8 (−0.5 – 6.4)
Relative change in Galectin-3, median (IQR) (%)	22 (−4 – 52)
Baseline sST-2, median (IQR) (ng/mL)	14.7 (10.5-20.1)
Absolute change in sST-2, median (IQR) (ng/mL)	−0.5 (−3.7 – 3.3)
Relative change in sST-2, median (IQR) (%)	−4 (−24 – 27)
Age (years)	59.4 (11.2)
Women	310 (44)
Race/ethnicity
Non-Hispanic white	332 (47)
Non-Hispanic black	249 (36)
Hispanic	86 (12)
Other	33 (5)
Baseline eGFR (CKD-EPI), mL/min/1.73m2	46.8 (14.1)
Relative change (%) in eGFR (CKD-EPI)	−8.2 (20.9)
24-hour urine protein (g/d), median	0.1 (0.1-0.5)
Diabetes	302 (43)
History of CVD
Never	522 (75)
At Year 2 only	21 (3)
At baseline and Year 2	157 (22)
Current smoker	70 (10)
BMI (kg/m2)	31.5 (7.4)
SBP (mmHg)	125.1 (20.4)
DBP (mmHg)	69.9 (12.4)
LDL cholesterol (mg/dL)	100.7 (33.6)
HDL cholesterol (mg/dL)	48.1 (15.5)
ACEi/ARBs
Never	160 (23)
At Year 2 only	64 (9)
At baseline and Year 2	426 (61)
Diuretics
Never	262 (37)
At Year 2 only	62 (9)
At baseline and Year 2	297 (42)
Beta blockers
Never	358 (51)
At Year 2 only	62 (9)
At baseline and Year 2	239 (34)
PTH, median	46.5 (31.0-76.0)

In multivariable models, lower eGFR was associated with greater odds of two change absolute change for all the cardiac biomarkers of interest, with the exception of hsTnT and sST2 (Table 2). Higher systolic blood pressure was associated with greater odds of change in NT-proBNP. Black race and diabetes were associated with greater odds of change in GDF-15. (Table 2).

Table 2.

Baseline characteristics associated with category of 2-year change in cardiac biomarker with the quartile representing the smallest increase of the biomarker serving as the referent (categories are Q1, Q2/Q3, and Q4 of 2-year biomarker change)

	Odds ratio (95% CI)
	NT-proBNP	hsTNT	GDF-15	Galectin-3	sST-2
Age (per 10 year increment)	0.94 (0.79, 1.12)	0.84 (0.72, 0.99)	0.98 (0.83, 1.15)	0.92 (0.78, 1.09)	1.04 (0.88, 1.23)
Female sex	1.64 (1.12, 2.39)	1.37 (0.95, 1.97)	0.97 (0.67, 1.40)	1.43 (1.00, 2.05)	0.71 (0.50, 1.01)
Race/ethnicity
Non-Hispanic Black	0.89 (0.61, 1.30)	1.12 (0.78, 1.62)	1.45 (1.01, 2.09)	1.08 (0.75, 1.54)	1.19 (0.83, 1.71)
Hispanic	1.95 (1.18, 3.24)	1.02 (0.62, 1.68)	1.26 (0.75, 2.10)	0.92 (0.56, 1.53)	1.27 (0.78, 2.07)
Other	0.78 (0.36, 1.66)	0.76 (0.36, 1.58)	0.72 (0.32, 1.61)	0.66 (0.31, 1.42)	0.73 (0.35, 1.56)
eGFR (per 15 mL/min/1.73m2 decrement)	1.21 (1.01, 1.46)	1.16 (0.98, 1.38)	1.25 (1.04, 1.49)	1.19 (1.01, 1.41)	0.93 (0.78, 1.10)
24 hour urine protein (per 1 g/day increment)	1.11 (0.95, 1.30)	0.98 (0.84, 1.14)	1.03 (0.88, 1.20)	1.18 (0.99, 1.39)	1.14 (0.98, 1.32)
Diabetes	0.99 (0.68, 1.45)	0.75 (0.51, 1.10)	1.44 (1.00, 2.09)	1.28 (0.90, 1.84)	1.05 (0.73, 1.50)
CVD	1.31 (0.91, 1.90)	0.72 (0.49, 1.05)	1.25 (0.86, 1.81)	1.30 (0.91, 1.85)	0.93 (0.65, 1.35)
Smoking	1.10 (0.66, 1.82)	1.24 (0.75, 2.07)	1.49 (0.88, 2.51)	1.27 (0.77, 2.09)	1.11 (0.68, 1.80)
BMI (per 5 kg/m2 increment)	0.99 (0.88, 1.12)	0.98 (0.88, 1.10)	1.00 (0.89, 1.13)	0.94 (0.84, 1.05)	1.07 (0.96, 1.20)
SBP (per 10 mmHg increment)	1.13 (1.02, 1.25)	1.06 (0.96, 1.17)	1.07 (0.97, 1.19)	0.96 (0.86, 1.06)	1.03 (0.94, 1.14)
DBP (per 5 mmHg increment)	0.88 (0.81, 0.97)	0.99 (0.91, 1.08)	0.95 (0.87, 1.03)	1.00 (0.92, 1.09)	0.99 (0.91, 1.08)
LDL (per 10 mg/dL increment)	1.03 (0.98, 1.09)	0.96 (0.91, 1.01)	0.99 (0.94, 1.04)	0.95 (0.90, 1.00)	0.94 (0.89, 0.99)
HDL (per 10 mg/dL increment)	0.90 (0.79, 1.03)	0.97 (0.86, 1.10)	0.98 (0.87, 1.10)	0.93 (0.83, 1.05)	1.06 (0.95, 1.19)
ACEi/ARB	0.68 (0.48, 0.97)	1.08 (0.76, 1.55)	0.84 (0.58, 1.20)	0.98 (0.69, 1.39)	0.99 (0.70, 1.40)
Diuretics	0.87 (0.62, 1.23)	1.10 (0.79, 1.53)	0.92 (0.65, 1.28)	1.01 (0.73, 1.40)	1.17 (0.84, 1.63)
Beta blockers	1.29 (0.92, 1.80)	1.11 (0.80, 1.54)	1.16 (0.83, 1.61)	1.15 (0.83, 1.59)	0.78 (0.56, 1.08)
Baseline NT-proBNP (per 1 SD increment)	1.09 (0.91, 1.31)	1.00 (0.83, 1.21)	0.89 (0.69, 1.15)	0.98 (0.80, 1.20)	1.20 (1.01, 1.43)
Baseline hsTNT (per 1 SD increment)	1.07 (0.90, 1.29)	1.79 (1.42, 2.25)	1.03 (0.87, 1.22)	1.16 (0.98, 1.37)	1.08 (0.92, 1.28)
Baseline GDF-15 (per 1 SD increment)	1.21 (0.98, 1.48)	1.01 (0.82, 1.24)	1.21 (1.01, 1.46)	0.96 (0.79, 1.17)	1.02 (0.84, 1.23)
Baseline Galectin-3 (per 1 SD increment)	0.81 (0.68, 0.96)	1.03 (0.88, 1.22)	1.02 (0.87, 1.21)	1.14 (0.98, 1.33)	1.01 (0.86, 1.20)
Baseline SST-2 (per 1 SD increment)	1.10 (0.94, 1.30)	1.21 (1.03, 1.43)	1.04 (0.88, 1.23)	0.97 (0.83, 1.15)	1.15 (1.00, 1.33)

Entries are odds ratios from a proportional odds model category of biomarker change, estimated among subcohort only, in a model that includes all listed covariates.

BOLD= p<0.05

Association of changes in NT-proBNP and hsTnT with incident HF and AF

The mean (SD) follow-up time from the second biomarker measurement among the cohort eligible for the HF analysis was 5.3 (2.4) years. In unadjusted models, the highest and lowest categories of change in NT-proBNP over 2 years were significant associated with increased risk of incident HF. However with adjustment for potential confounders in Model 1, only the highest quartile of change was significant associated with risk of incident HF. This association remained statistically significant when adjusted for baseline levels of NT-proBNP (HR 1.79, 95% CI 1.06-3.04, Table 3 and Supplemental Table 6). In sensitivity analyses, with adjustment of change in eGFR between baseline and year 2, use of cardiovascular medications at either baseline and Year 2, and development of interval AF or MI, the observed association remained statistically significant (Supplemental Table 7). When we evaluate relative change in NT-proBNP over 2 years, there was not a significant association with incident HF (Supplemental Table 9).

Table 3.

Associations of absolute change in cardiac biomarkers over 2 years with risk of incident heart failure

Cardiac Biomarker		Unadjusted		Model 1		Model 2
	N at risk (N events)	HR (95% CI)	p-value	HR (95% CI)	p-value	HR (95% CI)	p-value
Absolute change in NT-proBNP (pg/mL)
≤ −11.4	196 (41)	1.91 (1.22, 2.99)	0.005	1.68 (0.99, 2.85)	0.05	1.51 (0.88, 2.59)	0.13
−11.3 – 232	382 (47)	1.0 (Ref.)		1.0 (Ref.)		1.0 (Ref.)
> 232	185 (47)	2.61 (1.69, 4.05)	< 0.0001	1.94 (1.15, 3.27)	0.01	1.79 (1.06, 3.04)	0.03
Baseline NT-proBNP		N/A		NA		1.19 (1.03, 1.37)	0.02
Absolute change in hsTNT (ng/ml)
≤ −0.201	195 (35)	1.33 (0.85, 2.09)	0.21	1.12 (0.66, 1.92)	0.67	1.04 (0.60, 1.80)	0.88
−0.20 – 10.1	391 (57)	1.0 (Ref.)		1.0 (Ref.)		1.0 (Ref.)
> 10.1	191 (43)	1.85 (1.20, 2.85)	0.005	1.42 (0.86, 2.34)	0.17	1.28 (0.76, 2.15)	0.36
Baseline hsTnT		N/A		N/A		1.17 (0.99,1.38)	0.06
Absolute change in GDF-15 (pg/mL)
≤ −96.8	182 (33)	1.33 (0.83, 2.12)	0.23	1.15 (0.66, 2.00)	0.62	0.95 (0.52, 1.72)	0.87
−96.7 – 532	357 (49)	1.0 (Ref.)		1.0 (Ref.)		1.0 (Ref.)
> 532	184 (49)	2.56 (1.67, 3.94)	< 0.0001	1.27 (0.74, 2.17)	0.38	1.26 (0.74, 2.15)	0.40
Baseline GDF-15		N/A		N/A		1.28 (1.01, 1.61)	0.04
Absolute change in Galectin-3 (ng/ml)
≤ −0.405	194 (32)	1.03 (0.66, 1.63)	0.89	0.84 (0.49, 1.45)	0.54	0.66 (0.36, 1.24)	0.20
−0.404 – 6.95	405 (64)	1.0 (Ref.)		1.0 (Ref.)		1.0 (Ref.)
> 6.95	187 (42)	1.72 (1.13, 2.64)	0.01	0.90 (0.53, 1.53)	0.70	0.94 (0.55, 1.59)	0.81
Baseline galectin-3		N/A		N/A		1.25 (0.97-1.61)	0.08
Absolute change in SST-2 (ng/ml)
≤ −3.78	196 (40)	1.85 (1.18, 2.90)	0.007	1.93 (1.11, 3.35)	0.02	1.66 (0.91, 3.04)	0.10
−3.77 – 3.37	373 (49)	1.0 (Ref.)		1.0 (Ref.)		1.0 (Ref.)
> 3.37	203 (48)	2.35 (1.52, 3.62)	0.0001	1.87 (1.11, 3.13)	0.02	1.89 (1.13, 3.16)	0.02
Baseline sST2		N/A		N/A		1.14 (0.94, 1.39)	0.19

Model 1: Year 2 age, sex, race, site, diabetes, CVD, smoking, 24h urinary protein, eGFR, SBP, BMI, LDL, HDL

Model 2: Model 1 + baseline biomarker*

The mean (SD) of follow-up time from the second biomarker measurement among the cohort eligible for the AF analysis was 5.3 (2.4) years. The highest category of change in NT-proBNP was significant associated with greater risk of incident AF models adjusted for potential confounders, including kidney function and cardiovascular risk factors. This association remained statistically significant when adjusting for baseline levels of NT-proBNP (Model 1 HR 2.32, 95% CI: 1.37-3.93, Table 4 and Supplemental Table 6). This association remained robust with adjustment for change in eGFR, use of cardiovascular medications and development of interim HF or MI (Supplemental Table 7). There was no association between the lowest category of change in NT-proBNP with incident AF. When we evaluate relative change in NT-proBNP over 2 years, there was not a significant association with incident AF (Supplemental Table 10).

Table 4.

Associations of absolute change in cardiac biomarkers over two years with incident atrial fibrillation

Cardiac Biomarker		Unadjusted		Model 1		Model 2
	N at risk (N events)	HR (95% CI)	p-value	HR (95% CI)	p-value	HR (95% CI)	p-value
Absolute change in NT-proBNP (pg/mL)
≤ −11.4	191 (27)	1.34 (0.80, 2.23)	0.26	1.25 (0.71, 2.20)	0.44	1.19 (0.67, 2.09)	0.56
−11.3 – 232	373 (44)	1.0 (Ref.)		1.0 (Ref.)		1.0 (Ref.)
> 232	183 (47)	2.87 (1.83, 4.49)	< 0.0001	2.43 (1.44, 4.09)	0.0009	2.32 (1.37, 3.93)	0.002
Baseline NT-proBNP		N/A		N/A		1.14 (0.96, 1.36)	0.13
Absolute change in hsTNT (ng/ml)
≤ −0.201	186 (23)	0.79 (0.48, 1.30)	0.35	0.88 (0.51, 1.51)	0.63	0.79 (0.45, 1.38)	0.42
−0.20 – 10.1	392 (66)	1.0 (Ref.)		1.0 (Ref.)		1.0 (Ref.)
> 10.1	183 (29)	1.13 (0.71, 1.80)	0.61	1.09 (0.64, 1.86)	0.76	0.98 (0.56, 1.71)	0.94
Baseline hsTnT		N/A		N/A		1.21 (1.00, 1.47)	0.05
Absolute change in GDF-15 (pg/mL)
≤ −96.8	183 (36)	1.42 (0.89, 2.25)	0.14	1.15 (0.69, 1.93)	0.59	1.06 (0.61, 1.83)	0.84
−96.7 – 532	350 (50)	1.0 (Ref.)		1.0 (Ref.)		1.0 (Ref.)
> 532	174 (29)	1.48 (0.91, 2.41)	0.11	0.96 (0.55, 1.68)	0.89	0.95 (0.55, 1.66)	0.86
Baseline GDF-15		N/A		N/A		1.15 (0.89, 1.48)	0.29
Absolute change in Galectin-3 (ng/ml)
≤ −0.405	193 (24)	0.71 (0.43, 1.16)	0.17	0.74 (0.43, 1.29)	0.29	0.55 (0.29, 1.05)	0.07
−0.404 – 6.95	402 (68)	1.0 (Ref.)		1.0 (Ref.)		1.0 (Ref.)
> 6.95	176 (31)	1.23 (0.78, 1.95)	0.38	1.18 (0.69, 1.99)	0.55	1.22 (0.72, 2.07)	0.46
Baseline Galectin-3		N/A		N/A		1.30 (1.30, 1.69)	0.05
Absolute change in sST-2 (ng/ml)
≤ −3.78	201 (46)	2.22 (1.43, 3.43)	0.0004	2.55 (1.53, 4.26)	0.0004	2.43 (1.39, 4.22)	0.002
−3.77 – 3.37	372 (48)	1.0 (Ref.)		1.0 (Ref.)		1.0 (Ref.)
> 3.37	185 (29)	1.42 (0.87, 2.33)	0.16	1.14 (0.65, 1.99)	0.66	1.14 (0.65, 2.00)	0.65
Baseline sST2		N/A		N/A		1.05 (0.86, 1.29)	0.64

Model 1: Year 2 age, sex, race, site, diabetes, CVD, smoking, 24h urinary protein, eGFR, SBP, BMI, LDL, HDL

Model 2: Model 1 + baseline biomarker

In unadjusted models, there was a strong association of the highest quartile of change in hsTnT with risk of incident HF (HR 1.85, 95% CI: 1.20, 2.85); however this association was attenuated and no longer statistically significant with adjustment for potential confounders and mediators (Table 3 and Supplemental Table 7). There was no association of high or low categories of change in hsTnT with risk of incident AF (Table 4). Similarly, there was no association of relative changes in hsTnT with risk of incident HF or AF (Supplemental Tables 9 and10).

Association of change in GDF-15, galectin-3 and sST2 with incident HF and AF

In unadjusted models, the highest category of change in GDF-15 was associated with a two-fold greater risk of incident HF (HR 2.56, 95% CI: 1.67, 3.94, Table 3). However this association was attenuated and no longer statistically significant with adjustment for potential confounders, including baseline GDF-15. There was no association of change in GDF-15 with incident AF (Table 4). Similarly, there was no association of relative changes in GDF-15 with risk of incident HF or AF (Supplemental Tables 9 and 10).

In unadjusted models, the highest category of change in galectin-3 was associated with a greater risk of incident HF (HR 1.72, 95% CI: 1.13, 2.64, Table 3). However this association was completely attenuated and no longer statistically significant with adjustment for potential confounders. There was no association of change in galectin-3 with incident AF (Table 4). Similarly, there was no association of relative changes galectin-3 with risk of incident HF or AF (Supplemental Tables 9 and 10).

In unadjusted models, both the lowest and highest categories of change in sST2 were significantly associated with greater risk of incident HF. With adjustment for potential confounders and baseline sST2, the quartile of greatest change remained significantly associated with risk of incident HF (HR 1.89, 95% CI: 1.13, 3.16); and remained significant with adjustment for change in eGFR, use of cardiovascular medications and development of interim AF and MI (Table 3 and Supplemental Table 3). The association of higher relative change in sST2 with incident HF was also significant (Supplemental Table 9). There was also a significant association between the lowest category of absolute change in sST2 with higher risk of incident AF in adjusted models (Model 1 HR 2.43, 95% CI: 1.39, 4.22, Table 4); this association persistent when adjusting for change in eGFR, use of cardiovascular medications and development of interim AF and MI (Supplemental Table 7). Similar associations were observed for relative change in sST2 over two years as well (Supplemental Table 10).

Discrimination of cardiac biomarkers to predict incident HF and AF

Additional of the change in each cardiac biomarker did not improve the discrimination in predicting incident HF or AF beyond baseline level of the cardiac biomarker or clinical variables (Supplemental Tables 11 and 12).

DISCUSSION

In this study of a well-characterized CKD cohort free of HF and AF, we examined associations of longitudinal changes in NT-proBNP, hsTnT, GDF-15, galectin-3 and sST2 over 2 years with risk of incident HF and AF. Participants with the greatest level of change in NT-proBNP (>232 pg/mL increase over 2 years) had 79% higher risk of incident HF and 2-fold higher risk of incident AF. Furthermore, participants with the greatest change in sST2 (>3.37 ng/mL increase over 2 years) also had 89% greater risk of incident HF; and those with the greatest declines in sST2 (>3.78 ng/mL decline over 2 years) had greater risk of incident AF. There were no associations between changes in hsTnT, galectin-3, GDF-15 with incident HF or AF.

Our study found that there was a significant association of absolute increases in NT-proBNP over 2 years with risk of subsequent HF and AF in this CKD cohort, independent of important confounders such as cardiovascular disease and kidney function. Prior studies in CKD have evaluated one-time measures of NT-proBNP^{77, 78} but not longitudinal changes with risk of HF. However studies in other populations have reported that changes in NT-proBNP are significantly associated with HF and HF outcomes.^79–83 For example, a study of the Valsartan Heart Failure Therapy trial showed that temporal trends over 4 months in NT-proBNP superseded the prognostic value of a single measure.⁸³ In a study of 190 HF patients, changes from normal to elevated NT-proBNP were significantly associated with death and hospitalization.⁸⁴ Prior studies have also reported strong associations of one-time measures of NT-proBNP with AF in various populations (including CKD); ^{32, 85–87} however there is limited prior data on changes in NT-proBNP with risk of AF in CKD patients. It is plausible that increases in NT-proBNP reflect progression of underlying subclinical cardiovascular pathophysiology, such as volume overload and cardiac stretch. If so, NT-proBNP may be a good surrogate marker to reflect physiological abnormalities in patients with CKD that contribute to the risk of HF and AF; but may not be a strong biomarker to predict HF or AF beyond other clinical variables.

We did not see a similar association when evaluating relative increases in NT-proBNP suggesting that absolute increases in NT-proBNP capture risk of incident HF and AF better than relative change. The reasons to explain this finding are not clear from the present study. However, it is interesting that the median relative changes were quite large compared with the absolute changes; it is possible that large relative changes are actually capturing healthy participants who start with lower levels of NT-proBNP. Future interventional studies may help explain this finding.

We also did not find that absolute or relative decreases in NT-proBNP were associated with lower risk of HF and AF in participants with CKD. This differs from prior studies of patients with known HF, which have demonstrated that decreases in NT-proBNP are associated with reduced risk of HF. For example, an analysis of 172 HF patients found that reductions in NT-proBNP level were associated with reduced risk of death.⁸⁸ In a study of older adults, those with a 25% increase in NT-proBNP had a two-fold risk of HF and those who had a 25% decrease had a 40% decreased risk of HF.⁸⁹ It is possible that differences in the study population may explain the disparate findings.

We also noted a statistically significant association between absolute and relative increases in sST2 with risk of incident HF. However, sST2 measured at baseline or serially did not improve discrimination of HF in this CKD population. In two studies of patients hospitalized with HF, admission to discharge change in sST2 was significant associated with mortality.^{59, 60} In a study of 151 patients with chronic HF, serial sST2 measurement added prognostic information to the baseline concentrations and predicted change in left ventricular function.⁹⁰ An International Consensus Panel concluded that while there is strong data supporting single measurements of sST2, further data are needed to test the merits of serial testing of sST2.⁹¹ Our study provides support that serial sST2 measures may provide important mechanistic data in patients with CKD at risk for HF. It is surprising that we also noted an association between declining absolute and relative sST2 levels with increased risk of incident AF. The reasons for this observation are not clear but may be related to unmeasured confounders, such as treatment affects (e.g. confounding by indication). Further, this suggests that changes in sST2 are not causally linked to risk of AF.

After adjustment for potential confounders, we did not find statistically significant associations of absolute or relative changes in hsTnT, galectin-3 and GDF-15 with risk of incident HF or AF. This differs from previous studies of HF patients, older adults and dialysis patients which have found associations of changes in TnT with risk of events such as HF and death.^{79, 80, 82, 92} For example, in a study of older adults, decreases in TnT were associated with reduced risk of incident HF while increases in TnT over 2 or 3 years were associated with increased risk.⁸⁰ Also, in the same population of older adults, increases in hsTnT >10.97 were associated with a 90% increased risk of incident AF.⁵⁶ Furthermore, a few studies have evaluated longitudinal measures of GDF-15^{15, 58} and galectin-3^{37, 38, 93–95} with conflicting findings. For example, in a study of participants in the PREVEND study, the highest quartile of change in galectin-3 over 4 years was significantly associated with a composite outcome of new-onset heart failure, cardiovascular mortality, all-cause mortality and new-onset AF.⁵⁷ In contrast, prior studies reported no association of even baseline measures of galectin-3 with incident AF.^{38, 87} In HF patients, increases in GDF-15 were significantly associated with greater risk of mortality in one study.¹⁵ However, another study of HF patients did not report a significant association.⁹⁰ Thus, our data in a CKD population does not suggestion that serial measures of hsTnT, galectin-3 and GDF-15 add prognostic information.

The findings from this study have potentially important implications. Despite being at disproportionate risk for HF and AF, the mechanisms that contribute to these diseases in CKD remain incompletely understood. These circulating biomarker may help identify pathways that could be targets for therapies.. NT-proBNP has the advantage of being widely available in clinical practice. Furthermore, the use of NT-proBNP to guide use of HF therapies in patients with prevalent HF has been controversial with some trials suggesting a benefit in reducing recurrent HF hospitalizations^52–55 while others suggesting none.⁹⁶ The findings from prior trials appear to differ based on the study population of interest. Based on our findings, it is possible that serial measures of NT-proBNP and sST2 may successfully guide use of cardiovascular therapies to reduce the risk of incident HF and AF in select CKD patients. Further investigation of biomarker guided interventions and other surrogate markers are needed in CKD to reduce the risk of cardiovascular disease.

Our study had several strengths. We studied a large, multicenter, well-characterized CKD cohort with longitudinal follow-up. Our study population was diverse. We had serial measures of all five cardiac biomarkers measured two years apart. HF and AF were ascertained through a rigorous physician-adjudication process. We were able to adjust for a broad range of potential confounders. We recognize a few limitations as well. Cardiac biomarkers were only measured twice, 2 years apart; this may have been too short an interval for meaningful changes, although large differences were observed. There may have been day-to-day biological variability in the biomarker concentrations which may have made our assessment of change less precise. NT-proBNP and hsTnT at baseline and Year 2 were not measured in batch, however the excellent goodness of fit suggests that it is less likely that the timing of the measures would have influenced our findings. We adjusted for potential confounders and mediators, however the possibility of residual confounding remains. This was an observational study so we cannot determine causality. Finally, this was a study of research volunteers with CKD, so findings may not be generalizable to all populations.

In conclusion, in this longitudinal cohort of CKD patients without HF, we found that increases in NT-proBNP were significantly associated with greater risk of incident HF and AF; and increases in sST2 were associated with incident HF. Further studies are needed to test whether interventions to modify markers of subclinical cardiovascular disease can decrease the risk of HF and AF in CKD.

Supplementary Material

Supplement

Supplemental Table 1. Demographics based on absolute change over 2 years in NT-proBNP in subcohort (N = 675)

Supplemental Table 2. Demographics based on absolute change over 2 years in hsTNT in subcohort (N = 689)

Supplemental Table 3. Demographics based on absolute change over 2 years in GDF-15 in subcohort (N = 638)

Supplemental Table 4. Demographics based on absolute change over 2 years in Galectin-3 in subcohort (N = 697)

Supplemental Table 5. Demographics based on change in sST-2 over 2 years in subcohort (N=684)

Supplemental Table 6: Association of baseline cardiac biomarkers (per SD of the log-transformed concentration) with risk of incident HF and AF

Supplemental Table 7. Associations of absolute change in biomarkers over two years with risk of incident heart failure, adjusting for medication use and interim cardiovascular events

Supplemental Table 8. Associations of absolute change in biomarkers over two years with risk of incident atrial fibrillation, adjusting for medication use and interim cardiovascular events

Supplemental Table 9. Associations of relative change over two years in cardiac biomarkers with incident heart failure

Supplemental Table 10. Associations of relative change over two years in cardiac biomarkers with incident atrial fibrillation

Supplemental Table 11. Discrimination of cardiac biomarkers for predicting incident heart failure in CKD

Supplemental Table 12. Discrimination of cardiac biomarkers for predicting incident atrial fibrillation in CKD

Supplemental Figure 1. Residual plots for (a) original and (b) retested NT-proBNP

Supplemental Figure 2. Residual plots for (a) original and (b) retested hsTnT

Supplemental Figure 3. Histograms of absolute change in each biomarker in subcohort (N = 700)