New statin use and left ventricular structure: estimating long-term associations in the Multi-Ethnic Study of Atherosclerosis (MESA)

Abstract

PURPOSE

Only small and short-term studies have evaluated statins in relation to changes in heart structure. We estimated the association between new statin use and 10-year remodeling of the left ventricle.

METHODS

The Multi-Ethnic Study of Atherosclerosis (MESA) collected data on statin use over approximately 10 years, conducting cardiac magnetic resonance (CMR) imaging at baseline and the 10-year exam. Participants were free of baseline cardiovascular disease (CVD), and we excluded users of statins at baseline. Statin initiation was defined as a report of current use at any of the four subsequent exams. Primary outcomes were the change in left ventricular mass index (LVMI; % predicted by height, weight, sex) and mass-to-volume ratio (MVR). Associations were estimated in a propensity score-matched analysis.

RESULTS

A total of 3113 participants (53% female; 40% European-American, 25% African-American, 22% Hispanic-American, 13% Chinese-American) were eligible; 2431 returned for a follow-up CMR after a median of 9.4 years. Statin therapy (moderate dose, 76%) was started by 36% of participants (N=872). We excluded 42 participants with incident myocardial infarction. Compared with non-use, statin use was associated with less 10-year progression in LVMI (−2.35 percentage points, 95%CI: −4.24, −0.47, p=0.01) and MVR (−0.03 absolute difference, 95%CI: −0.07, −0.00, p=0.02); effects were small in magnitude. A dose response was observed: higher statin dose was associated with less LVMI progression.

CONCLUSIONS

In contrast to previous small studies, we found very modest associations between statin use and indices of left ventricular remodeling over 10 years in this prospective study of a diverse cohort.

INTRODUCTION

Heart failure (HF) is a common¹ clinical syndrome characterized by structural and functional impairment of the left ventricle.² With high treatment costs due to repeat hospitalization, HF is associated with an increasingly high level of patient morbidity over time. Among many treatments explored for HF, statins (i.e. HMG-CoA reductase inhibitors) have been evaluated in both primary and secondary prevention. While an approximate 10% reduction in the risk of HF hospitalization has been shown in meta-analyses of statin primary and secondary prevention trials,^3,4 the mechanism of action is unclear. These agents may slow or reduce the cardiac remodeling,⁵ with some evidence suggesting a beneficial effect of statins on structure and function.^6,7

Recent clinical guidelines have increased the number of statin-eligible individuals.⁸ Those recommended statins may be young and expected to take these drugs for extended periods.^9,10 Understanding the potential benefits and harms of long-term treatment with statins is essential for clinical decision making. Among other side effects, there is evidence that statins may cause myopathy, more severe musculoskeletal dysfunction,¹¹ and diabetes.^12,13 While the short-term efficacy of statins is well-studied, few investigations have evaluated statins and long-term changes in cardiac structure, especially among healthy individuals. The purpose of this study was to estimate the association of new statin use with ten-year longitudinal changes in structure and function of the left ventricle in a diverse population free of clinical CVD at baseline. The primary outcomes were left ventricular mass (LVM), an important measure of LVH and predictor of cardiovascular events,¹⁴ and mass-to-volume ratio (MVR), a measure of concentric remodeling.¹⁵

METHODS

Participants

The Multi-Ethnic Study of Atherosclerosis (MESA) has been described previously.¹⁶ In brief, MESA is an ongoing prospective observational cohort study designed to investigate the pathogenesis of cardiovascular disease in four racial/ethnic groups, where all participants were free of known CVD at enrollment defined as any previous physician-diagnosed hypertension, myocardial infarction (MI), angina or use of nitroglycerin, stroke or transient ischemic attack (TIA), or HF; current atrial fibrillation; and prior CVD procedures. Between 2000 and July of 2002, MESA enrolled 6814 participants ages 45 to 84 years from six locations in the U.S.: Baltimore City and County, Maryland; Manhattan and the Bronx, New York; Chicago, Illinois; Forsyth County, North Carolina; Los Angeles, California; and St. Paul, Minnesota. All MESA protocols were approved by the Institutional Review Board at participating study institutions.

MESA participants attended a baseline examination (exam 1) between July of 2000 and July of 2002 (24 months), and four subsequent exams. Exams 1 through 4 occurred at average intervals of 9 to 21 months after the baseline exam. Exam 5 occurred between April 2010 and December 2011 (21 months), approximately 10 years after the baseline exam. Demographic information, medical history, anthropometric measurements, and medication inventories were collected at each exam. Current medications were evaluated at all five MESA exams using a validated medication inventory¹⁷ in which interviewers transcribed names, strengths, and dosage from medication bottles brought to the exam by participants. Participants were also asked about actual medication use in the previous two weeks. Clinical cardiovascular events including stroke, TIA, MI, and HF were ascertained and adjudicated by experts in neurology, cardiology, and epidemiology.¹⁶ Cardiac magnetic resonance (CMR) imaging protocols have been described previously¹⁸ and were performed on a subset of MESA participants. Coefficients of variation quantified at exam 1 were small (approximately 6% for LVM and 4% for end-diastolic volume).¹⁸ Changes in CMR pulse sequence technology and software between exams required that exam 1 parameters be adjusted to be comparable to those at exam 5. Calibration was performed with a subset of participants at exam 5 who were imaged using both the original gradient echo and the newer steady state free precession (SSFP) techniques as well as the different software packages for reading images.¹⁸ A total of 5004 MESA participants (73%) completed the baseline CMR scan and had technically adequate data. The main reasons for not completing scans were: ineligibility due to metallic implant or device (7%), inability (14%), mostly due to claustrophobia, refusal (3%), mechanical problems with the scanner (0.4%), and unknown (1%).¹⁹ For this investigation, we included all MESA participants free of statin use at baseline and with two serial CMR at baseline and 10-year follow-up.

Statin use

Several variables were defined to summarize statin use. Any statin initiation during study follow-up was the primary exposure, defined as any positive report of a statin on the medication inventory at the second through the fifth clinical exams that occurred at unequal intervals over a 10-year period. Participants with one or more missing reports of statin use at a clinical exam (e.g. they did not bring medication to the appointment) and no reported statin use at other exams were classified as having unclear or missing statin use.

To summarize statin dose, we estimated the mean daily dose reported during the study period. National drug codes (NDC) recorded in the medication inventory at each clinical exam were used to group statins into low, moderate, and high dose according to the ACC/AHA Guideline on the Treatment of Blood Cholesterol.⁸ A new user study design, where the analysis is restricted to medication non-users at baseline, was selected to increase the likelihood that baseline characteristics were not impacted by current statin use.²⁰

Cardiac Magnetic Resonance Imaging

The primary cardiac measures of interest were the absolute change in left ventricular end-diastolic mass (LVM) and mass-to-volume ratio (MVR) over the study period, derived from the MESA CMR SSFP readings. Prior to calculating absolute change, baseline and follow-up LVM were indexed using the allometric height-weight-gender methods described in Brumback et al.²¹ Because height and weight changed between baseline and follow-up exams, we indexed baseline LVM by gender and baseline height and weight, and follow-up LVM by gender and follow-up height and weight. We studied body surface area (BSA)-indexed LVM and unindexed LVM as secondary outcomes. MVR is the most commonly used measure of concentric remodeling, and was calculated as the ratio of unadjusted end-diastolic mass to left ventricular end-diastolic volume (LVEDV). Additional secondary outcomes included the absolute change in LVEDV and LVEF, where LVEF was defined as the stroke volume divided by the end diastolic volume and multiplied by 100.¹⁴

Statistical analysis

We estimated descriptive statistics for baseline characteristics among statin initiators and non-initiators with complete covariate data. We estimated propensity scores for statin initiation using a logistic regression with baseline covariates. Covariates included age, gender, race and other traditional cardiovascular risk factors^22,23,19,24 (smoking status, systolic blood pressure, diastolic blood pressure, treated diabetes, body mass index (BMI), waist circumference, HDL cholesterol, triglycerides, and total cholesterol), antihypertensive medication use (indicator variables for angiotensin-converting-enzyme inhibitors, angiotensin type 2 agonists, beta blockers, calcium channel blockers, and diuretics), potential predictors of statin use (intentional exercise and health insurance status), and the Agatston coronary artery calcium (CAC) score (range: 0–5148) derived from the baseline computed tomography (CT) scan. All CAC scores + 1 were transformed by the natural-log to approximate a normal distribution among those with positive CAC. We restricted the analysis to the much larger subset of individuals without an incident MI during the study period because the presence of an MI may predict both statin treatment and alterations in left ventricular structure and function. In addition, as individuals were able to initiate statins at any of the four clinical exams during the study period, we were unable to determine the timing of starting statins in relation to the timing of the incident MI in many instances. We did not exclude individuals with other clinical cardiovascular events.

Propensity score overlap was visually assessed, and covariate balance between statin initiators and matched non-initiators was evaluated using t-tests and with Rubin’s B and R statistics.²⁵ The primary analysis estimated the average treatment effect in the treated (ATT) from a propensity score-matched (PSM) model²⁶ of statin initiation using 1-to-1 nearest-neighbor matching within 0.025 caliper of propensity score with replacement. We also modeled the relationship using traditional multivariable linear regression. We ran three separate PSM models to evaluate the dose of the initiated statin, using the same group of non-users of statins as the comparison group. Additional multivariable linear models for statin dose are presented. Two sensitivity analyses tested whether these results were robust to 1) the use of multiple imputation for missing data on covariates and outcomes and 2) inverse probability of censoring weighting to account for any loss-to-follow up, as only a subset of participants returned for a follow-up CMR scan approximately 10 years after baseline. All analyses were completed using Stata version 14.0. The PSM analysis was completed using the psmatch2 Stata module (v4.0.11).²⁷ A two-tailed p-value less than 0.05 was considered statistically significant. We determined that the 95% confidence intervals (CIs) generated with a bootstrapping approach (N=1000) were conservative compared with those adjusting for large sample variance using the methodology of Abadie & Imbens (2016).²⁸

RESULTS

Baseline Characteristics

At baseline, of 6814 MESA participants, 4265 MESA participants were not statin users and had valid CMR measures (Figure 1). Among 4265 participants, 4234 had complete covariates at baseline. A total of 1832 (43%) participants had no statin use over the study period, 1281 (30%) initiated statins, and 1121 (27%) had unclear or missing statin use based on the study definitions. Those with unclear or missing statin use did not differ substantially from the population with complete covariates at baseline (N=4234), though they were slightly more likely to be uninsured (15% vs. 10%).

Figure 1.

Study flow for MESA participants

* Consists of all participants in the MESA cohort.

¥ Covariates include age, gender, race, smoking status, BMI, diabetes status, hypertension status, hypertension treatment, waist circumference, blood pressure, HDL cholesterol, triglycerides, and total cholesterol, intentional exercise, and health insurance at baseline and MI during the study period.

† 3 participants were missing follow-up data on height and weight and so it was not possible to index the follow-up left ventricular mass.

Baseline clinical and demographic characteristics are shown in Table 1 along with baseline cardiac indices for the subset of individuals with complete follow-up data. In general, non-initiators of statins were more likely to be younger, female, of Chinese-American ethnicity, and have lower blood pressure and total cholesterol compared with statin initiators. Rates of diabetes and hypertension, and CMR indices of LVM were higher among new users of statins. Statin initiators were also more likely to be using antihypertensive agents. Characteristics of the PSM sample are also shown in Table 1. In the final analysis sample (n=2389), most statin initiators (76%) started a moderate dose statin as shown in Suppl. Table 1. Only 13% of statin initiators used hydrophilic statins (i.e. pravastatin or rosuvastatin) during the study period. Further information on statin formulation and dose used during the study are described in Suppl. Table 2.

Table 1.

Baseline characteristics of participants who were statin initiators and non-initiators

Characteristic	Statin Initiators	Non-initiators	p-value	Matched Non-initiators	p-value
N	835	1554		835
Propensity score	0.49 (0.2)	0.27 (0.2)		--
Age	60.6 (8.8)	58.2 (9.5)	<0.001	61.3 (9.87) **	0.15
Male, N (%)	394 (47.2)	729 (46.9)	0.90	377 (45.2)	0.40
Ethnicity, N (%)
European-American	358 (42.9)	618 (39.8)	0.14	346 (41.4)	0.55
Chinese-American	94 (11.3)	223 (14.4)	0.03	95 (11.4)	0.94
African-American	206 (24.7)	391 (25.2)	0.79	241 (28.9)	0.05
Hispanic-American	177 (21.2)	322 (20.7)	0.79	153 (18.3)	0.14
Smoking status, N (%)
Never	438 (52.5)	836 (53.8)	0.53	433 (51.9)	0.81
Former	311 (37.3)	529 (34.0)	0.12	311 (37.3)	1.00
Current	86 (10.3)	189 (12.2)	0.17	91 (10.9)	0.69
BMI (kg/m2)	28.2 (4.8)	27.3 (5.0)	<0.001	27.9 (4.8)	0.11
Glucose status, N (%)
Normal	609 (72.9)	1327 (85.4)	<0.001	621 (74.4)	0.51
IFG	109 (13.1)	167 (10.8)	0.09	131 (15.7)	0.13
Diabetes, treated	31 (3.7)	16 (1.0)	<0.001	20 (2.4)	0.05
Diabetes, untreated	86 (10.3)	44 (2.8)	<0.001	63 (7.5)	0.12
Waist circumference (cm)	97.7 (12.6)	94.5 (13.6)	<0.001	97.2 (13.0)	0.41
Hypertension treatments, N (%)
Diuretic	119 (14.3)	112 (7.21)	<0.001	121 (14.5)	0.89
Calcium channel blocker	112 (13.4)	118 (7.6)	<0.001	89 (10.7)	0.08
Beta-blocker	72 (8.6)	90 (5.8)	0.009	74 (8.9)	0.86
ACE inhibitor (w/o diuretic)	99 (11.9)	85 (5.5)	<0.001	87 (10.4)	0.35
Angiotensin type 2 agonist	31 (3.7)	26 (1.7)	0.002	33 (4.0)	0.80
SBP (mmHg)	126.4 (18.8)	120.6 (20.3)	<0.001	127.2 (22.0) **	0.40
DBP (mmHg)	72.9 (10.0)	71.3 (10.4)	<0.001	73.0 (10.5)	0.75
HDL (mg/dL)	49.6 (13.6)	52.5 (16.0)	<0.001	49.1 (14.6) **	0.44
Total cholesterol (mg/dL)	211.6 (35.9)	188.2 (31.2)	<0.001	212.9 (36.5)	0.48
Triglycerides (mg/dL)	146.3 (85.2)	116.0 (63.8)	<0.001	147.4 (87.0)	0.78
Intentional exercise (met-min/week)	6396 (7117)	6126 (5813)	0.32	6218 (5851) **	0.58
No health insurance, N (%)	67 (8.0)	130 (8.4)	0.77	66 (7.9)	0.93
Natural log of Agatston CAC	2.2 (2.5)	1.3 (2.0)	<0.001	2.2 (2.4)	0.96
CMR Indices
LVM (g)	122.1 (28.9)	119.0 (28.1)	--	117.9 (28.0)	--
LVMI*	86.6 (14.1)	85.6 (13.0)	--	85.3 (13.5)	--
LVM-BSA indexed (g/m2)	64.8 (11.9)	63.8 (10.9)	--	63.4 (11.1)	--
LVEDV (mL)	129.5 (28.3)	131.4 (28.9)	--	126.4 (26.1)	--
MVR (g/mL)	0.96 (0.18)	0.92 (0.16)	--	0.94 (0.17)	--
LVEF (%)	62.8 (5.9)	62.4 (5.6)	--	62.4 (5.6)	--

This table reflects the population with complete covariates and no incident MI. P-values are from t-tests for equality of means in the two samples.

^*LVM is indexed by height, weight, and gender and multiplied by 100 using methods defined in the MESA cohort; an individual’s LVMI of 125 suggests that LVM is 25% greater than height, weight, and gender would predict.

^**Variance ratio outside [0.87; 1.15]. The variance ratio for the matched group should equal 1 if there is perfect balance between covariates. Covariates are flagged if variance ratios exceed the 2.5^th and 97.5^th percentiles of the F-distribution with number of matched treated minus 1 degrees of freedom.

Unadjusted Ten-year Change in Cardiac Indices

Follow-up CMR data were available for 2431 participants (57% of the original sample) after a median of 9.4 years. Of these individuals, 42 experienced an incident MI over the study period. Participants lost to follow-up (N=682) were generally similar to the population at baseline (N=4234); they were slightly less likely to be male (45% vs. 48%), and more likely to be Hispanic (27% vs. 23%), former smokers (38% vs. 35%), treated diabetics (12% vs. 7%), and taking an ACE inhibitor (14% vs. 9%), and had slightly higher mean triglyceride levels (144 vs. 129 mg/dL).

Among those with complete follow-up, the unadjusted mean changes in measures of left ventricular structure and function are shown in Table 2 and exclude those who experienced an MI. The average time between CMR scans was similar in statin initiators and non-initiators. Over this time, average mass and volume increased for both groups, while average LVEF declined. In general, indices of mass change were higher among non-initiators.

Table 2.

Unadjusted change over 10 years in height, weight, and cardiac magnetic resonance indices of left ventricular structure and function and incident MI, among statin initiators and non-initiators

	Statin Initiators	Non-initiators
N	835	1554
Time between CMR (years)	9.5 (0.5)	9.5 (0.5)
Height (cm)	−1.5 (1.4)	−1.5 (1.4)
Weight (lbs.)	−0.9 (15.1)	−1.7 (14.7)
Δ LVMI, %	2.11 (14.25)	2.55 (12.75)
Δ LVM unindexed, g	2.56 (20.35)	2.95 (17.86)
Δ LVM-BSA indexed	1.79 (10.64)	2.10 (9.43)
Δ MVR (unindexed)	0.12 (0.22)	0.12 (0.20)
Δ LVEDV, ml	11.01 (22.96)	11.16 (21.47)
Δ LVEF, %	−0.56 (7.75)	−0.73 (7.18)

All values are mean (SD) unless otherwise indicated. All CMR indices are unadjusted.

Propensity Score Matching

Figure 2 depicts the full overlap between the propensity scores of the statin initiators and non-initiators where all individuals were matched. Statistics for PSM diagnostics suggest mostly good balance of covariates in the statin initiators and matched non-initiators (Table 1, Rubin’s’ B=26.1, R=1.00).

Figure 2.

Distribution of propensity scores among 835 statin initiators and 1554 non-initiators

Covariates included in the logistic model to derive propensity scores include age, gender, race, smoking status (former, never, current), BMI, diabetes status (normal, impaired fasting glucose, untreated diabetes, treated diabetes), waist circumference, antihypertensive agent use (yes/no for diuretics, calcium channel blockers, beta-blockers, ace-inhibitors, and angiotensin type 2 antagonists), systolic and diastolic blood pressure, HDL cholesterol, triglycerides, total cholesterol, intentional exercise defined as moderate and vigorous physical activity total (met-min per week), health insurance status (yes/no), and the Agatston CAC Score as the ln(score + 1).

Association Between Statin Initiation and Change in Cardiac Indices

In the primary PSM analysis, excluding those who experienced an MI, statin initiation was statistically significantly associated with less progression of LVMI (−2.35 percentage points, 95%CI: −4.24, −0.47), LVM-BSA indexed (−1.58 grams/BSA, 95%CI: −2.96, −0.19), and MVR (−0.03, 95%CI: −0.07, −0.00), though the effects were small in magnitude (Table 3). Results from traditional multivariate regression, inverse probability of censoring weighted regression, and multiply imputed regression models were similar; all estimates were closer to the null than the estimates from the PSM analysis (Suppl. Table 3).

Table 3.

Propensity score matched models estimating the association between statin use vs. non-use and 10-year change in CMR parameters of left ventricular structure/function

Estimate	95% CI*	p-value*
Δ LVMI (percent predicted by height, weight, and gender)
−2.35	(−4.24, −0.47)	0.01
Δ LVM unindexed (grams)
−2.03	(−4.67, 0.61)	0.13
Δ LVM-BSA indexed (grams/BSA)
−1.58	(−2.96, −0.19)	0.03
Δ MVR (no units)
−0.03	(−0.07, −0.00)	0.02
Δ EDV (mL)
−1.35	(−4.32, 1.63)	0.38
Δ LVEF (percent)
0.57	(−0.45, 1.60)	0.27

The PSM models used nearest-neighbor matching within 0.025 caliper with replacement to estimate the average treatment effect in the treated. Propensity scores for statin initiation were derived from a logistic regression using the following at baseline: age, gender, race, smoking status (former, never, current), BMI, diabetes status (normal, impaired fasting glucose, untreated diabetes, treated diabetes), waist circumference, antihypertensive agent use (yes/no for diuretics, calcium channel blockers, beta-blockers, ace-inhibitors, and angiotensin type 2 antagonists), systolic and diastolic blood pressure, HDL cholesterol, triglycerides, total cholesterol, intentional exercise defined as moderate and vigorous physical activity total (met-min per week), health insurance status (yes/no), and the Agatston CAC Score as the ln(score + 1).

^*Normal-based confidence intervals were bootstrapped with 1000 repetitions.

Association Between Statin Dose and Change in Cardiac Indices

In the dose stratified PSM analyses, results for LVMI and LVEDV were numerically consistent with a dose response (Table 4). There was less progression of LVMI as the statin dose increased, while there was less progression in LVEDV as the dose decreased. Moderate dose statins were associated with statistically significantly less progression of LVMI relative to never use (−2.50, 95% CI: −4.60, −0.41, p=0.02). Low dose statins were associated with statistically significantly more decline in LVEF relative to never use (2.35 percentage points, 95% CI: 0.33, 4.37, p=0.02). Results from traditional regression models were similar (Suppl. Table 4).

Table 4.

Association between statin use vs. non-use and 10-year change in CMR parameters of left ventricular structure/function, estimated using propensity score matching and stratified by statin dose

Dose	Estimate	95% CI*	p-value*
Δ LVMI (percent predicted by height, weight, and gender)
Low	−0.66	(−4.48, 3.16)	0.74
Moderate	−2.50	(−4.60, −0.41)	0.02
High	−3.72	(−8.90, 1.46)	0.16
Δ LVM unindexed (grams)
Low	−0.18	(−5.93, 5.58)	0.95
Moderate	−2.35	(−5.17, 0.46)	0.10
High	−2.18	(−10.06, 5.70)	0.59
Δ LVM-BSA indexed (grams/BSA)
Low	−0.33	(−3.23, 2.57)	0.83
Moderate	−1.71	(−3.23, −0.19)	0.03
High	−2.39	(−6.55, 1.77)	0.26
Δ MVR (no units)
Low	−0.05	(−0.11, 0.01)	0.12
Moderate	−0.03	(−0.06, 0.00)	0.06
High	−0.04	(−0.13, 0.04)	0.33
Δ EDV (mL)
Low	−5.38	(−11.17, 0.41)	0.07
Moderate	−0.75	(−3.95, 2.46)	0.65
High	−0.14	(−9.46, 9.18)	0.98
Δ LVEF (percent)
Low	2.35	(0.33, 4.37)	0.02
Moderate	0.59	(−0.49, 1.67)	0.27
High	−2.27	(−5.24, 0.70)	0.13

The PSM models used nearest-neighbor matching within 0.025 caliper with replacement to estimate the average treatment effect in the treated. The low dose statin model consisted of 1672 individuals, of whom 38 untreated individuals did not overlap on propensity scores across groups. The moderate statin dose model consisted of 2194 individuals, and there was complete overlap on propensity scores across the treatment and comparison group. The high dose statin use model consisted of 1631 individuals, of whom 268 untreated individuals did not overlap on propensity scores across groups.

^*Normal-based confidence intervals were bootstrapped with 1000 repetitions.

DISCUSSION

This investigation demonstrated modest association between new statin use and long-term changes in cardiac structure and function in a diverse population with no clinical CVD at baseline. While there was some evidence for an association between statin use and reduced progression in left ventricular mass, all effect sizes were small. This contrasts with previous studies which have demonstrated larger changes, and a growing body of literature on statin pleiotropic effects. Ultimately, findings suggest long-term safety of statins that may include a small cardioprotective effect on heart structure. However, these findings are unlikely to fully explain the protective effects of statins on heart failure hospitalization in statin primary prevention trials (RR: 0.89, 95%CI: 0.67, 1.17).⁴ The mechanism of the beneficial effect of statins on stroke outcomes is still being elucidated. For example, a recent analysis found that statins reduce microembolic signaling in acute cerebral ischemia attributed to large artery atherosclerosis.²⁹

The cholesterol-independent effect of statins on left ventricular mass is postulated to occur through activity on both cardiomyocytes and fibroblasts, which account for 30% and 70% of the myocardium, respectively.³⁰ Among cardiomyocytes, statins may prevent or reduce hypertrophy via the inhibition of small GTPase signaling pathways (i.e. Rho, Rac, and especially, Ras).³¹ In animal models of CVD^32–38 as well as early stage hypertension³⁹ and hypercholesterolemia,⁴⁰ statins have been shown both to prevent development of cardiac hypertrophy and to induce regression of established hypertrophy. Early observational studies in humans have mostly evaluated current or short-term duration of statin use.^41,42 Randomized studies of statin use and cardiac structure and/or function have often focused on small populations with cardiovascular diseases including cardiomyopathy,^43–48 congenital aortic stenosis,⁴⁹ and heart failure.^50–52 These investigations have not consistently demonstrated statin effects on LVM and other indices in addition to having short time frames and limited statistical power.

Observational studies and trials in populations similar to MESA (Table 5) have tended to show much larger magnitudes of LVM regression^42,53 or reduced progression of LVM⁵⁴ than this investigation. In the longest and largest of these studies, the Hypertension High Risk Management Trial (HYRIM), drug-treated hypertensive patients randomized to fluvastatin had significantly reduced two-year progression of LVM compared with placebo-treated patients.⁵⁴ However, the HYRIM study included a markedly less healthy population than MESA, as evidenced by the magnitude of increase in LVM over two years among placebo-treated participants (approximately 30 grams vs. 3 grams among non-statin users in MESA). A different randomized investigation of fluvastatin versus placebo in hypertensive patients, demonstrated a similar magnitude of LVMI decline between fluvastatin (−17 g) and placebo (−16 g) randomized groups over 1 year, though this study was not powered to detect a difference between groups.⁵⁵ A randomized study of rosuvastatin did not find a significant effect of six months of treatment on LVMI relative to placebo among patients with hypertension and left-ventricular hypertrophy.⁵⁶ Collectively, these conclusions are supported by the smaller estimated magnitude of statin effects in our investigation.

Table 5.

Comparison with literature effect sizes

Study	Population	Sample	Timeframe	Statin	Effect estimate (g/m2)
Cohort42	Hyperlipidemia/essential hypertension	60	6 m	Pravastatin 10 mg	−16
Cohort6	Hypertension/LVH (elderly)	220	12 m	Pitavastatin 1–2 mg	−6
RCT7	LVH/essential hypertension	41	12 m	Simvastatin 10 mg	−21
RCT55	Essential hypertension	39	12 m	Fluvastatin 20 mg	−1
RCT56	Hypertension/LVH	142	6 m	Rosuvastatin 20 mg	2
RCT54	Hypertension/high BMI	368	24 m	Fluvastatin 40 mg	−11*
RCT53	Hypercholesterolemia	50	6 m	Pravastatin 10/20 mg	−14
This study	No CVD, require statins	2389	120 m	Statins	−2

The majority of these studies were not designed to detect differences between statins and placebo for indexed LVM. The estimates are derived and presented to inform a comparison of the magnitude of the effect.

^*This value is estimated from the change in grams. It assumes BSA remains constant and uses DuBois’ formula to calculate BSA from BMI, weight, and height.

MESA is a unique setting for investigating statin exposure and long-term changes in cardiac structure and function because the cohort is a population of relatively healthy individuals followed for medication use and subclinical measures, as well as clinical events. In addition, while a majority of previous investigations have relied upon two-dimensional echocardiographic measures of ventricular structure and function, MESA used CMR, which has higher resolution for structure, requiring fewer geometric and modeling assumptions. MESA provides the largest and longest investigation of statin exposure and heart structure to date. This investigation evaluated change over approximately 10 years, while few previous studies have evaluated statin use and cardiac structure/function for periods longer than one year.^46,49,54 Short-term studies are useful because cardiac remodeling may be a relatively fast process. Preclinical studies have demonstrated regression of cardiac mass within a few weeks of statin administration.⁵⁶ The long-term benefits and harms of statin use are also important. Statins have been associated with notable adverse side effects, particularly muscle dysfunction. As we did not detect strong or consistent associations between statin use and measures of structure and function of the left ventricle even when accounting for known confounders using PSM, our findings add support to studies suggesting that attenuation of structural remodeling is not the main (non-cholesterol mediated) mechanism through which statin therapy could produce a benefit on HF clinical outcomes. Alternative mechanisms may involve improved cardiovascular function (possibly occurring via reduced inflammation),⁵⁷ though we did not observe any statistically significant associations between statin use and the one evaluated functional measure, change in LVEF.

This study has several limitations. Firstly, results are subject to exposure misclassification due to the structure of the statin use reporting. We note that across several approaches of modeling statin use, the magnitudes of the effects were relatively consistent. In this investigation, individuals who had the greatest cardiac changes may have been be more likely to have CV events and less likely to return for a second scan. Thus, loss to follow-up bias may attenuate the association between statins and cardiac structure. However, re-weighting data by the inverse probability of censoring⁵⁸ and using multiple imputation for missing covariates did not produce a material difference in the results. This provides evidence against selection bias and suggests that either the effect of loss to follow-up bias may be minimal or that the mechanism is dependent on unmeasured covariates. An additional limitation may be residual confounding, especially via incomplete consideration of time-varying factors. Statin use was quantified based on all clinical exams, though other clinical covariates were considered only at baseline. Several important variables that changed during the study follow-up may have impacted results. These include absolute weight changes,¹⁵ new use of antihypertensive agents,⁵⁹ changes in blood pressure and/or kidney function, and the occurrence of clinical events (particularly MI). Lastly, findings are susceptible to indication⁶⁰ and healthy user biases⁶¹; however, these biases may be less concerning because of the new-user design, the sub-clinical outcome measures, and the adjustment for characteristics that could differ between statin users and non-users (i.e. exercise, health insurance, and other medication use) using the PSM approach.

CONCLUSION

We found modest associations between statin use and long-term changes in indices of cardiac remodeling, when accounting for age, sex, race, and other cardiovascular risk factors. Results are consistent with literature suggesting that an effect on cardiac remodeling is not likely the main mechanism of statin benefit in primary prevention of HF. The results may be consistent with short-term or modest dose-related effects of statins on cardiac structure; however, these findings suggest that there is minimal long-term effect in the primary prevention clinical setting. Future long-term studies may elucidate the causal relationship between statins and heart failure outcomes.

KEY POINTS.

• To date, small and short-term studies have investigated statins and heart structure.

• This study in a large, diverse cohort demonstrated a modest association between statin use and reduced remodeling of the left ventricle.

• Overall findings suggest a statin effect on cardiac remodeling is not the main mechanism of statin benefit in the primary prevention of heart failure.

ABBREVIATIONS

Abbreviation	Definition
ACC	American College of Cardiology
ACE	Angiotensin converting enzyme
ADA	American Diabetes Association
AHA	American Heart Association
BMI	Body mass index
BSA	Body surface area
CAC	Coronary artery calcium
CI	Confidence interval
CMR	Cardiac magnetic resonance
CT	Computed tomography
CVD	Cardiovascular disease
DBP	Diastolic blood pressure
HDL	High density lipoprotein
HF	Heart failure
IFG	Impaired fasting glucose
IPCW	Inverse probability of censoring weighted
JNC	Joint National Committee
LDL	Low density lipoprotein
LVEDV	Left ventricular end diastolic volume
LVEF	Left ventricular ejection fraction
LVH	Left ventricular hypertrophy
LVM	Left ventricular mass
LVMI	Left ventricular mass index
MESA	Multi-Ethnic Study of Atherosclerosis
MI	Myocardial infarction
MVR	Mass-to-volume ratio
NDC	National drug codes
PSM	Propensity score-matched
SBP	Systolic blood pressure
SD	Standard deviation
SE	Standard error
TIA	Transient ischemic attack
XR	Extended release