Worsening kidney function is the major mechanism of heart failure in hypertension: the ALLHAT study

Abstract

Objectives:

We aimed to quantify the extent to which the effect of antihypertensive drugs on incident heart failure (HF) is mediated by their effect on kidney function.

Background:

We hypothesized that the dynamic change in kidney function is the mechanism behind differences in the rate of incident HF in ALLHAT participants randomized to lisinopril and chlorthalidone, in comparison to those randomized to amlodipine and doxazosin.

Methods:

Causal mediation analysis of ALLHAT data (1994–2002) included participants with available baseline and 24–48 month estimated glomerular filtration rate (eGFR) (n=27,918; mean age 66±7.4; 32.4% black, 56.3% men). Change in eGFR was the mediator. Incident symptomatic HF was the primary outcome. Hospitalized/fatal HF was the secondary outcome. Linear regression (for mediator) and logistic regression (for outcome) analyses were adjusted for demographics, cardiovascular disease, and risk factors.

Results:

There were 1,769 incident HF events, including 1,359 hospitalized/fatal HF events. In fully adjusted causal mediation analysis, the relative change in eGFR mediated 18% of the effect of chlorthalidone, and 33% of lisinopril on incident symptomatic HF, and 25% of the effect of chlorthalidone, and 41% of lisinopril on hospitalized/fatal HF. In participants with diabetes, the relative change in eGFR mediated nearly 50% of the effect of lisinopril on incident symptomatic HF, whereas in diabetes-free participants – only 17%.

Conclusion:

On the risk difference scale, change in eGFR accounts for up to 50% of the mechanism by which antihypertensive medications affect HF.

Clinical Trial Registration—

URL:www.clinicaltrials.gov Unique identifier: NCT00000542.

Introduction

Kidney dysfunction is a well-known risk factor for heart failure (HF),(1) carrying the highest predictive value among the HF risk scores.(2) In patients with end-stage kidney disease (ESKD), and HF with reduced left ventricular ejection fraction (LVEF), kidney transplantation resulted in normalization of LVEF.(3) Although it is well-accepted that hypertension accelerates kidney injury, the role of hypertension in the initiation of kidney disease remains controversial.(4) While commonly used antihypertensive medications such as angiotensin-converting enzyme inhibitors (ACEI) and diuretics may cause an increase in serum creatinine, this effect may be viewed as a favorable prognostic indicator for HF(5) if associated with a decrease in signs and symptoms of congestion.(6) In patients with hypertension, it is unclear to what degree impaired kidney function is a marker of HF severity or a reflection of a mechanism contributing to HF progression.(7) The diagnosis, pathophysiology, prognosis, and treatment of the cardiorenal syndrome(1) are focused on established chronic kidney disease (CKD) and HF. In contrast, knowledge gaps persist in understanding of the role of impaired kidney function as a mechanism that predisposes to the development of incident HF in hypertensive patients.

To address these knowledge gaps, we analyzed data from the Antihypertensive and Lipid-Lowering Treatment to Prevent Heart Attack Trial (ALLHAT).(8) ALLHAT was a multicenter, randomized, double-blind trial designed to compare cardiovascular outcomes in high-risk hypertensive patients assigned to the ACEI lisinopril, the calcium channel blocker (CCB) amlodipine, and the ά-blocker doxazosin, in comparison to the thiazide-like diuretic chlorthalidone. We quantified the extent to which the effects of lisinopril and chlorthalidone, as compared to amlodipine and doxazosin on incident HF, are mediated by their effects on kidney function. We hypothesized that a dynamic change in kidney function is the mechanism behind previously observed differences in the rates of incident HF in ALLHAT participants randomized to lisinopril, amlodipine, and doxazosin, in comparison to those randomized to chlorthalidone.(9,10)

Methods

We used the ALLHAT dataset that is publicly available from the National Heart, Lung, and Blood Institute, via BioLINCC.(11) The Oregon Health & Science University Institutional Review Board reviewed the study and determined the deidentified nature of the publicly available dataset.

Study population

ALLHAT(8) was conducted from 1994 to 2002. Adults age 55 and above with hypertension and at least one additional cardiovascular disease (CVD) risk factor were enrolled. The risk factors included documented coronary heart disease, type 2 diabetes mellitus, left ventricular hypertrophy (LVH) on ECG or echocardiogram, smoking, high-density lipoprotein (HDL) < 35mg/dL, and ST-T ECG changes indicative of ischemia. Exclusion criteria included symptomatic HF or LVEF <35%, recent myocardial infarction, stroke, poorly controlled hypertension, and a serum creatinine level >2 mg/dL (>176.8 μmol/L).(12)

In the current study, we included ALLHAT participants with available data on estimated glomerular filtration rate (eGFR) changes during in-trial follow-up. We excluded participants with missing mediator or covariates. The final study population (Figure 1) included 27,918 participants: 10,487 were randomized to chlorthalidone, 5,388 to doxazosin, 6,166 to amlodipine, and 5,877 to lisinopril.

Figure 1.

Study flow diagram.

Mediator: eGFR change

As previously described(12), at baseline, serum creatinine was measured at a central laboratory using the VITROS chemistry system (Ortho-Clinical Diagnostics, Rochester, NY) with a coefficient of variation of approximately 2%. Creatinine measurements were externally validated(13) in two other laboratories, indicating that serum creatinine values measured in ALLHAT were comparable to the MDRD study measurements. Thus, use of the MDRD study equation in ALLHAT was justified.(12)

Estimated GFR was calculated using the simplified MDRD equation(12):

$$\text{eGFR}(\text{mL}/\text{min}/1.73{\text{m}}^2)=186.3*{\left(\text{creatinine}\right)}^{-1.154}*{\left(\text{age}\right)}^{-0.203}*1.212\left(\text{Black}\right)*0.742\left(\text{female}\right).$$

Serum creatinine measurements were repeated at one month, one year, two years, and then every other year of follow-up.(14) To calculate eGFR change during the trial, we subtracted baseline eGFR from the eGFR measured after at least 48 months (4 years) of follow-up. If 48-month eGFR measurement was not available, we subtracted baseline eGFR from the eGFR measured after at least 24 months (2 years) of follow-up. Participants with missing follow-up eGFR measurements were excluded. Next, we normalized eGFR change by the baseline eGFR value, to obtain relative eGFR change. To investigate whether the time between two eGFR measurements affected the study results, we conducted sensitivity analyses for 2-year and 4-year eGFR change, separately.

Outcome: Incident heart failure

The primary outcome in the current study was incident symptomatic congestive HF, as defined by the ALLHAT investigators. Symptomatic congestive HF was diagnosed in the presence of both: (1) Paroxysmal nocturnal dyspnea, dyspnea at rest, New York Heart Association class III symptoms or orthopnea, and (2) rales, ankle edema (2+ or greater), sinus tachycardia of 120 beats per minute or more after 5 minutes at rest, cardiomegaly by chest X-ray, chest X-ray characteristic of congestive HF, S3 gallop or jugular venous distention. In the current study, the secondary outcome was hospitalized/fatal HF. Both outcomes were validated by the ALLHAT HF validation study.(15–17)

Covariates

The ALLHAT investigators obtained the baseline medical history by a combination of chart review and questioning during a routine office visit. The hypertension history included categories of participants who were treated for at least two months, or less than two months/untreated. Baseline coronary heart disease history included known (including silent) myocardial infarction, angina, cardiac arrest, angiographically defined coronary stenosis more than 50%, reversible perfusion defects on cardiac scintigraphy, or prior coronary revascularization procedures. The presence of major ST-segment depression or T wave elevation on an ECG in the past two years was recorded. History of other atherosclerotic CVD included documented peripheral arterial disease or cerebrovascular disease.

Diabetes subgroup analysis was prespecified in ALLHAT.(8) Type II diabetes was defined as fasting plasma glucose > 140 mg/dl [7.77 mmol/L] or non-fasting plasma glucose >200 mg/dl [11.1 mmol/L] in the past two years and/or current treatment with insulin or oral hypoglycemic agents. History of HDL cholesterol < 35 mg/dl (0.91 mmol/l) on two or more determinations within the past five years was identified. The current smoking history was included. Baseline ECG-LVH or echocardiographic LVH (Echo-LVH) within the past two years was defined as previously described.(18)

Baseline blood pressure (BP) was assessed as an average of two BP determinations taken at least one day apart, with each determination being an average of 2 measurements. At every visit (every three months for the 1^st year and every four months thereafter), BP was recorded as an average of two measurements. We estimated the BP-lowering by subtracting baseline BP from the BP obtained at the latest in-trial study visit available at year sixth, fifth, fourth, third, second, or first.(18)

Statistical analysis

Antihypertensive treatment groups were defined per the intention-to-treat randomization assignment. Continuous variables are reported as means±standard deviation (if normally distributed) or as the median and interquartile range (IQR) if skewed distribution. For unadjusted comparison of clinical characteristics in participants with three tertiles of relative eGFR change, we used ANOVA and χ²test. To determine the association of clinical characteristics with relative eGFR change, we used multivariable linear regression models adjusted for age, sex, and race/ethnicity.

Associations of relative eGFR change and HF risk were also evaluated using fully adjusted (as described below) logistic regression models incorporating cubic splines with 4 knots at −30.7, −10.9, −0.6, and 25.1% of relative eGFR change.

We conducted causal mediation analysis(19), using counterfactual definitions of direct and indirect effects in parametric regression models, as implemented by VanderWeeleet al(20), which allows for treatment-mediator interaction. We estimated two models: (1) a linear model for the mediator conditional on treatment and covariates, and (2) a logistic model for the outcome conditional on treatment, the mediator, and covariates. In this study, treatment randomization eliminated exposure-outcome and exposure-mediator confounding. Relative change in eGFR over the course of the trial was used as a mediator. We adjusted for mediator-outcome confounders(16,21): demographic (age, sex, race, and ethnicity) and clinical characteristics known to be associated both with kidney function and HF: levels of baseline systolic and diastolic BP, in-trial systolic and diastolic BP lowering as previously defined,(18) length of antihypertensive treatment before enrollment, ECG- or echo-LVH, history of myocardial infarction, stroke, or other CVD, coronary revascularization, major ST depression or T-wave inversion, HDL<35 mg/dL twice in the past five years, body mass index, smoking, diabetes, use of aspirin, participation in the lipid-lowering ALLHAT trial, baseline levels of total cholesterol, eGFR, and potassium, and geographic region.

Considering the clinical importance of diuretics and ACEI, we merged doxazosin and amlodipine groups and used them together as a base/control group. We quantified the extent to which the effect of lisinopril and chlorthalidone (as compared to amlodipine and doxazosin) on incident HF is mediated by relative eGFR change.

We calculated four types of effects: the total, natural direct, natural indirect, and controlled direct effects. Natural direct effect represents the influence of antihypertensive treatment that is independent of eGFR changes, in the absence of eGFR changes (e.g., via blood pressure-lowering or drug-specific pharmacodynamics). A natural indirect (mediated) effect represents the influence of an antihypertensive drug that can be explained by its influence on dynamic eGFR changes over the course of the trial. To characterize treatment-mediator interaction, we calculated the controlled direct effect of the antihypertensive drug at eGFR increase and decrease by 10, 20, 30, 40, and 50%.

To quantify the extent of mediation, we calculated two metrics: proportion mediated (PM) and the proportion eliminated (PE) as follows:

Proportion mediated PM = || DE*(ME-1)/(DE*ME-1)||, where DE is a natural direct effect, and ME is a mediated effect. It captures what would happen to the effect of treatment if we disable the pathway from the treatment to the mediator, setting mediator to a single value.

Proportion eliminated PE(m) = (TE – CDE(m))/(TE-1), where TE is the total effect, and CDE(m) is the controlled direct effect at the level of mediator m. It captures what would happen to the effect of treatment on the outcome if we were to fix the mediator to the same value M = m for all persons, which is important in the case of treatment-mediator interaction.

Considering the importance of diabetes for incident HF and eGFR change, we conducted subgroup analysis in participants with and without diabetes.

To test the robustness of our findings, we conducted a sensitivity analysis for 2-year and 4-year eGFR change, separately.

Statistical analyses were performed using STATA MP 16.1 (StataCorp LLC, College Station, TX), and open-source code is provided at https://github.com/Tereshchenkolab/statistics.

Results

Study population. Relative change in eGFR during the trial.

Clinical characteristics of participants stratified by the tertiles of relative eGFR change are reported in Table 1. Baseline cardiovascular risk factors and cardiovascular outcomes of ALLHAT participants in different treatment groups stratified by baseline eGFR(12,22), and those with different renal outcomes(14) have been reported in detail previously.

Table 1.

Clinical characteristics of study participants by tertiles of eGFR change

Characteristic	Q1 eGFR change (n=9,307)	Q2 eGFR change (n=9,309)	Q3 eGFR change (n=9,302)	P
Relative eGFR change, median(range)	−18.0(−94.9 to −11.01)	−0.7(−11.0 to −0.6)	+12.3(−0.55 to +335)
Age(SD), y	66.5(7.5)	65.3(6.7)	67.8(7.6)	<0.0001
Black race, n(%)	3,179(34.2)	2,806(30.1)	2,984(32.1)	<0.0001
Men, n(%)	4,815(51.4)	5,792(62.2)	5,221(56.1)	<0.0001
HTN treated ≥ 2months, n(%)	8,064(86.6)	8,038(86.4)	8,277(89.0)	<0.0001
BMI(SD), kg/m2	29.7(6.0)	29.7(5.7)	29.5(5.6)	0.029
Baseline SBP(SD), mmHg	147.8(15.4)	145.4(15.5)	144.3(15.7)	<0.0001
Baseline DBP(SD), mmHg	84.0(10.1)	84.3(9.9)	83.2(10.0)	<0.0001
Hx of MI/stroke, n(%)	2,151(23.1)	2,175(23.4)	2,100(22.6)	0.427
Hx of revascularization, n(%)	1,299(14.0)	1,336(14.4)	1,243(13.4)	0.145
Hx of ST-T or other CVD, n(%)	2,932(31.5)	2,878(30.9)	3,080(33.1)	0.004
Diabetes, n(%)	3,732(40.1)	2,888(31.0)	2,908(31.3)	<0.0001
HDL<35mg/dL, n(%)	995(10.7)	1,330(14.3)	1,249(13.4)	<0.0001
Current smoking, n(%)	2,143(23.0)	2,054(22.1)	1,752(18.8)	<0.0001
Baseline ECG/Echo LVH, n(%)	1,809(19.4)	1,835(19.7)	1,914(20.6)	0.127
Baseline total cholesterol(SD)	216.7(43.4)	213.9(41.2)	215.9(41.9)	<0.0001
Baseline potassium(SD)	4.32(0.68)	4.34(0.60)	4.36(0.65)	0.0007
Baseline eGFR(SD)	83.5(20.8)	79.1(17.5)	71.6(16.6)	<0.0001
SBP change(SD), mmHg	−12.9(20.3)	−11.6(19.8)	−10.1(19.9)	<0.0001
DBP change(SD), mmHg	−8.6(11.9)	−8.0(11.6)	−7.3(11.7)	<0.0001
Midwest region, n(%)	1,761(18.9)	1,906(20.5)	2,017(21.7)	<0.0001
ITT chlorthalidone, n(%)	4,043(43.3)	3,404(36.6)	3,049(32.8)	<0.0001
ITT lisinoptil, n(%)	2,099(22.6)	2,027(21.8)	1,751(18.8)
ITT amlodipine, n(%)	1,588(17.1)	2,035(21.9)	2,543(27.3)
ITT doxazosin, n(%)	1,586(17.0)	1,843(19.8)	1,959(21.1)

LVH=left ventricular hypertrophy; SD=standard deviation; HTN=hypertension; BMI=body mass index; SBP=systolic blood pressure; DBP=diastolic blood pressure; MI=myocardial infarction; Hx=history; ST-T = major ST depression or T-wave inversion on ECG; CVD=cardiovascular disease; CHD=coronary heart disease; HDL=high density lipoprotein

In most of the study participants, relative eGFR changes were very small (Figure 2). On average, eGFR reduced by less than one percent (median −0.7; IQR −13.3 to 8.4%). Unadjusted analysis (Table 1) showed that baseline clinical characteristics had both linear and U-shaped relationships with relative eGFR change. Alongside the relative eGFR changes (decreased - unchanged - increased eGFR), there were gradually declining baseline eGFR, baseline BP, BP-lowering, and a number of participants in lisinopril and chlorthalidone groups, but gradually rising baseline potassium, and a number of participants in doxazosin and amlodipine groups. Study participants in the middle tertile of relative eGFR changes, as compared to both extremes, were slightly younger, more likely white men with a lower level of baseline total cholesterol.

Figure 2. Risk of heart failure across the distribution of relative eGFR change during the trial.

Adjusted odds of (A) symptomatic congestive heart failure (CHF) and (B) hospitalized or fatal HF associated with relative eGFR changes, in all participants. Restricted cubic spline with a 95% confidence interval shows a change in odds ratio (Y-axis) in response to relative eGFR change (X-axis). The 50th percentile of eGFR change is selected as the reference. Knots are at −30.7, −10.9, −0.6, and 25.1% of relative eGFR change.

In a linear regression analysis adjusted for demographic characteristics (Table 2), treatment of hypertension for at least two months prior to the start of the trial, baseline HDL<35 mg/dL, Hispanic ethnicity, and Midwest or Puerto Rico residence were associated with an increase in eGFR during the trial. As expected, most of the cardiovascular risk factors (age, systolic and diastolic BP, diabetes, smoking, history of myocardial infarction or stroke) and Black race were associated with a decrease in eGFR during the trial (Table 2).

Table 2.

Associations of clinical characteristics with eGFR change in linear regression

Characteristic	Relative eGFR change(95%CI), %	P
Age, per 1 y increase	−0.05(−0.08 to-0.02)	<0.0001
Race/ethnicity: White non-Hispanic	Reference
Black non-Hispanic	−0.70(−1.18 to −0.21)	0.005
White Hispanic	+1.39(0.68–2.10)	<0.0001
Black Hispanic	+1.82(0.45–3.18)	0.009
Male sex	+0.05(−0.38 to 0.48)	0.810
Hx of antihypertensive Rx ≥ 2months	+1.82(1.10–2.53)	<0.0001
BMI, per 1 kg/m2 increase	−0.02(−0.06 to 0.01)	0.206
Baseline SBP, per 1 mmHg increase	−0.13(−0.14 to −0.12)	<0.0001
Baseline DBP, per 1 mmHg increase	−0.06(−0.08 to −0.04)	<0.0001
Hx of MI/stroke	−0.87(−1.38 to −0.37)	0.001
Hx of revascularization	−1.05(−1.68 to −0.42)	0.001
Hx of ST-T on ECG or other CVD	+0.47(0.12–0.92)	0.044
Total cholesterol, per 1 mg/dL increase	+0.002(−0.004 to 0.007)	0.540
Hx of Aspirin use	−0.74(−1.18 to −0.29)	0.001
Diabetes	−3.10(−3.55 to −2.66)	<0.0001
HDL<35mg/dL	+1.29(0.66–1.93)	<0.0001
Current Smoking	−0.96(−1.48 to −0.43)	<0.0001
Baseline potassium, per 1 mmol/L increase	+0.19 (−0.13 to 0.52)	0.247
Baseline eGFR, per 1 ml/min/1.73 m2increase	−0.22(−0.23 to −0.21)	<0.0001
Baseline ECG- or echo-LVH	+0.40(−0.13 to 0.94)	0.140
Geographic region: East	reference
Midwest	+1.30(0.59–2.02)	<0.0001
South	+0.48(−0.15 to 1.12)	0.135
West	+0.91(0.06 to 1.75)	0.035
Canada	+1.37(−0.15 to 2.90)	0.077
Puerto Rico	+2.34 (0.88 to 3.79)	0.002

LVH=left ventricular hypertrophy; CI=confidence interval; BMI=body mass index; SBP=systolic blood pressure; DBP=diastolic blood pressure; MI=myocardial infarction; Hx=history; ST-T = major ST depression or T-wave inversion on ECG; CVD=cardiovascular disease; HDL=high density lipoprotein.

Mediation of Heart Failure by eGFR changes

After a median of 3.2 years follow-up in the doxazosin group, and 5.0 years in the other three groups, there were 1,769 incident HF outcomes, including 1,359 hospitalized/fatal HF events. The risk of both HF outcomes linearly increased with relative eGFR decline, whereas eGFR increase showed no statistically significant effect on HF outcomes (Figure 2).

Chlorthalidone reduced the relative risk of symptomatic HF by 28%, and hospitalized/fatal HF by 15%, as compared to doxazosin and amlodipine (Table 3). Lisinopril reduced the relative risk of symptomatic HF by 23%, and hospitalized/fatal HF by 14%, as compared to doxazosin and amlodipine. The relative change in eGFR during the trial mediated up to 40% of the effect of lisinopril and up to 25% of chlorthalidone effect on the incident symptomatic HF and hospitalized/fatal HF (Table 3). Direct and mediated effects of both lisinopril and chlorthalidone opposed each other. Mediated by eGFR change effect of lisinopril and chlorthalidone increased risk of HF, whereas the direct effect of lisinopril and chlorthalidone reduced HF risk.

Table 3.

Causal mediation results. Fully adjusted effect of antihypertensive treatment on incident heart failure (total), through eGFR changes (mediated), and independent of eGFR changes (direct).

HF Outcome	Treatment	Total effect RR(95%CI)	Direct effect RR(95%CI)	Mediated effect RR(95%CI)	% Mediated
CHF	Chlorthalidone	0.72(0.64–0.80)	0.67(0.59–0.75)	1.08(1.05–1.10)	−17.8
	Lisinopril	0.85(0.74–0.97)	0.81(0.70–0.92)	1.06(1.03–1.09)	−33.3
Hosp/fatal	Chlorthalidone	0.77(0.69–0.89)	0.71(0.63–0.82)	1.08(1.05–1.11)	−25.2
HF	Lisinopril	0.86(0.72–0.98)	0.80(0.67–0.91)	1.07(1.04–1.11)	−40.6
CHF Diabetes(+)	Chlorthalidone	0.69(0.58–0.84)	0.64(0.53–0.77)	1.08(1.04–1.13)	−16.8
	Lisinopril	0.84(0.66–1.05)	0.77(0.61–0.96)	1.10(1.05–1.16)	−47
Hosp/fatal	Chlorthalidone	0.75(0.60–0.90)	0.70(0.56–0.84)	1.08(1.03–1.13)	−21.9
HF Diabetes(+)	Lisinopril	0.82(0.62–1.04)	0.74(0.57–0.93)	1.11(1.06–1.17)	−43
CHF Diabetes(−)	Chlorthalidone	0.72(0.62–0.84)	0.68(0.58–0.79)	1.07(1.03–1.11)	−16.6
	Lisinopril	0.87(0.74–1.04)	0.85(0.72–1.03)	1.03(0.96–1.07)	−16.7
Hosp/fatal	Chlorthalidone	0.78(0.65–0.93)	0.72(0.60–0.86)	1.08(1.04–1.13)	−24.9
HF Diabetes(−)	Lisinopril	0.90(0.73–1.10)	0.87(0.70–1.09)	1.04(0.99–1.10)	−36.1

RR=relative risk. Proportion mediated=||DE*(ME-1)/(DE*ME-1)||, where DE is direct effect and ME is mediated effect.

We observed significant treatment-mediator interaction (Figure 3). The controlled direct effect at different levels of eGFR change was considerably different; it gradually increased consistently with relative eGFR reduction. If eGFR was reduced by 50%, chlorthalidone reduced the relative risk of symptomatic HF by 42%, and hospitalized/fatal HF by 40%, as compared to doxazosin and amlodipine, via the direct effect. Accordingly, if eGFR was reduced by 50%, lisinopril reduced the relative risk of symptomatic HF by 33%, and hospitalized/fatal HF by 29%, as compared to doxazosin and amlodipine, via the direct effect. Thus, eGFR change had two opposing effects: it mediated an increase in HF risk, and also was associated with the increased strength of the direct HF-protective effect of lisinopril and chlorthalidone.

Figure 3. Controlled at a particular level of mediator effect.

Controlled direct effect of lisinopril (A,C), and chlorthalidone (B,D), relative to doxazosin and amlodipine at eGFR increase and decrease by 10, 20, 30, 40, and 50%. Relative Risk of incident symptomatic congestive HF (A, B) and hospitalized/fatal HF (C, D).

Proportion eliminated (Central Illustration) was remarkably high for all treatment groups and both outcomes. In lisinopril arm, a relative change in eGFR explained up to 100% of symptomatic HF, and up to 60% of hospitalized/fatal HF. In chlorthalidone arm, a relative eGFR change explained up to 50% of symptomatic and up to 70% of hospitalized/fatal HF.

Central Illustration. The proportion of effect eliminated.

Proportion eliminated of the effect of lisinopril (A, C), and chlorthalidone (B, D) on incident symptomatic congestive HF (A, B) and hospitalized/fatal HF (C, D) at eGFR increase and decrease by 10, 20, 30, 40, and 50%.

Comparison of participants with and without diabetes

We observed heterogeneity of treatment-mediator interaction in participants with versus without diabetes (Figure 4). In participants with diabetes, the HF-protective controlled direct effect of chlorthalidone gradually increased with eGFR reduction, reaching 58% relative risk reduction for hospitalized/fatal HF, and 42% relative risk reduction for symptomatic HF in those with eGFR decline by 50%. In opposite, in diabetes-free participants, the HF-protective controlled direct effect of chlorthalidone gradually decreased with eGFR reduction, up to only 8% relative risk reduction for hospitalized/fatal HF, and 24% relative risk reduction for symptomatic HF in those with eGFR decline by 50%.

Figure 4. Subgroup analysis by diabetes.

A controlled direct effect of lisinopril and chlorthalidone (relative to doxazosin and amlodipine) and proportion eliminated of their effect on symptomatic CHF and hospitalized/fatal HF at eGFR increase and decrease by 10, 20, 30, 40, and 50%.

The relative change in eGFR mediated 47% of the effect of lisinopril on the incident symptomatic HF in participants with diabetes, in contrast to only 17% in diabetes-free participants (Table 3). Treatment-mediator interaction was stronger for diabetes-free participants than for those with diabetes in the lisinopril group. In diabetes-free patients, the protective controlled direct effect of lisinopril gradually increased with eGFR reduction, reaching 34% relative risk reduction for symptomatic HF in those with eGFR decline by 50%. In contrast, in participants with diabetes, the controlled direct effect of lisinopril for both HF outcomes did not change significantly with eGFR changes (Figure 4).

Sensitivity analysis

Mediation by eGFR change measured during 2 versus 4 years of follow-up was similar (Supplemental Table 1).

Discussion

This causal mediation analysis of the largest randomized controlled trial of antihypertensive treatment revealed that in hypertension patients with normal baseline kidney function, the cardiorenal syndrome was the main mechanism of incident HF. Even as small as 10% relative eGFR changes explained up to 50% of HF mechanisms. Large (±50%) eGFR change was frequently the main and only HF mechanism. There were differences in the degree of treatment-mediator interaction - between treatment groups and in diabetes subgroups. Relative eGFR change had two opposing effects: it mediated an increase in HF risk but also was associated with the increased strength of the direct HF-protective effect of lisinopril and chlorthalidone. Diabetes added another layer of complexity, showing heterogeneity of treatment-mediator interaction. In participants with diabetes, the HF-preventing controlled direct effect of chlorthalidone notably increased with eGFR reduction, whereas in diabetes-free participants, the HF-protective controlled direct effect of chlorthalidone slightly decreased with eGFR reduction.

Observational cohort studies provided the foundation for the cardiorenal syndrome definition with its characteristic feature of tangled interconnected concomitant presence of kidney dysfunction, and HF.(23) Hypertension is the leading risk factor of both CKD and HF.(24) Our causal mediation analysis of the largest randomized controlled trial supported the notion of HF as a kidney disorder.(25) While hypertension is one of the major risk factors for HF(26), BP-lowering mediated only up to 13% of the effect of the antihypertensive medications on incident HF.(18) We found that the cardiorenal syndrome explains a large portion of the effect of antihypertensive medications on HF.

Hypertension is characterized by arterial stiffening and endothelial dysfunction, which can lead to nephrons loss, compensatory hyperfiltration of the remaining nephrons in an attempt to maintain eGFR, and eventually, renal dysfunction.(27,28) Concomitant volume retention can lead to the volume-overloaded state, increased central venous pressure, and eventually, pulmonary hypertension and congestion, and symptomatic HF.(29) Endothelial dysfunction as a result of renal impairment facilitates myocyte stiffening and subsequent fibrosis, and provoke the inflammatory cascade and oxidative stress, predisposing to myocardial dysfunction. Our finding suggests that HF prevention cannot be successful without targeting the main HF mechanism - cardiorenal syndrome, which is supported by stalled progress in HF prevention for several decades.(30)

An average ALLHAT participant had a normal or mildly reduced baseline kidney function and a low risk for the development of ESKD.(12,31) Only approximately 20% of ALLHAT participants had CKD (eGFR < 60 mL/min/1.73m²) at the time of randomization. Of note, at baseline, ALLHAT excluded individuals with a history of symptomatic or hospitalized HF and LVEF < 35%. Among participants with hospitalized HF outcome, 44% developed HF with preserved LVEF (≥50%), and 56% - HF with reduced LVEF (<50%).(16) The development of ESKD, which was the prespecified secondary outcome was not different between treatment groups.(10,14) Thus, the ALLHAT study population was well-suited for the study of kidney dysfunction as a hypothesized HF mechanism. Importantly, we assessed dynamic changes in eGFR during the trial and applied rigorous causal mediation analysis adjusted for confounders. Altogether the rigorous study design, study population, and analytical approach support the main study conclusion about kidney dysfunction as the major mechanism of HF development in hypertensive patients.

Strengths and Limitations

ALLHAT is the largest trial of antihypertensive treatment, allowing unbiased mediation analysis. The development of ESKD was a prespecified secondary outcome in ALLHAT(9), which ensured the careful collection of dynamic eGFR data. Randomization eliminated exposure-outcome and exposure-mediator confounding. The ALLHAT had prespecified subgroup analysis by diabetes, which permitted subgroup analysis by diabetes in this study.

However, limitations of this study should be taken into account. While we adjusted for known common causes of eGFR change and incident HF, unmeasured confounding cannot be completely ruled out. ALLHAT did not report urine albumin in the patient population, which hindered us from using the degree of albuminuria, an important marker for CKD in this study. The ALLHAT was a pragmatic randomized controlled trial, and, therefore, did not measure several important confounders (right ventricular function, hemoglobin, acute kidney injury). Therefore, the results of the study should be interpreted with caution. We conducted a posthoc analysis of the ALLHAT trial, and, therefore, our findings need to be validated in the prospective study.

Conclusion

In hypertensive patients with normal baseline kidney function, the cardiorenal syndrome is the main mechanism of incident HF. On the risk difference scale, change in eGFR accounts for more than 50% of the mechanism by which antihypertensive medications affect HF. By targeting kidney function and choosing antihypertensive medications that prevent eGFR decline, one can eliminate a significant proportion of HF risk.

Clinical Perspectives

In hypertensive patients, the cardiorenal syndrome is one of the main mechanisms affecting the risk of HF. The monitoring of eGFR change can be used to guide hypertension management. The use of antihypertensive medications preventing eGFR reduction can help to eliminate a substantial proportion of HF risk.

Translational Outlook

Future development of novel interventions for hypertensive patients should target kidney function preservation and improvement.

Abbreviations:

HF

heart failure

eGFR

estimated glomerular filtration rate

ALLHAT

the Antihypertensive and Lipid-Lowering Treatment to Prevent Heart Attack Trial

MDRD

Modification of Diet in Renal Disease trial

ESKD

end-stage kidney disease

LVEF

left ventricular ejection fraction

ACEI

angiotensin-converting enzyme inhibitor

CKD

chronic kidney disease

CCB

calcium channel blocker

CVD

cardiovascular disease

HDL

high density lipoprotein

LVH

Left Ventricular Hypertrophy

BP

blood pressure

IQR

interquartile range