Despite ezetimibe’s ability to reduce serum cholesterol levels, there are concerns over its vascular effects and whether it prevents or ameliorates atherosclerotic disease (AD). The aims of this study were to estimate the effect of ezetimibe use on major AD events and all-cause mortality and to compare these associations to those observed for hydroxymethylglutaryl coenzyme A reductase inhibitor (statin) use. A total of 367 new ezetimibe users were identified from November 1, 2002, to December 31, 2009. These subjects were aged ≥18 years and had no previous statin use. One to 4 statin user matches were identified for each ezetimibe user, resulting in a total of 1,238 closely matched statin users. Pharmacy data and drug dosage information were used to estimate a moving window of ezetimibe and statin exposure for each day of study follow-up. The primary outcome was a composite of major AD events (coronary heart disease, cerebrovascular disease, and peripheral vascular disease events) and all-cause death. Ezetimibe use (odds ratio 0.33, 95% confidence interval 0.13 to 0.86) and statin use (odds ratio 0.61, 95% confidence interval 0.36 to 1.04) were associated with reductions in the likelihood of the composite outcome. These protective associations were most significant for cerebrovascular disease events and all-cause death. Subgroup analyses by gender, race or ethnicity, history of AD, diabetes status, and estimated renal function showed consistent estimates across strata, with no significant differences between ezetimibe and statin use. In conclusion, ezetimibe appeared to have a protective effect on major AD events and all-cause death that was not significantly different from that observed for statin use.
A number of studies suggest that ezetimibe use reduces low-density lipoprotein (LDL) cholesterol levels without a commensurate improvement in carotid intima-media thickness (CIMT), a proxy measure of atherosclerotic disease (AD) burden. However, 1 clinical trial in patients with chronic kidney disease showed that patients randomized to using ezetimibe and a statin concomitantly had fewer major AD events compared to those assigned to placebo. Because of the seemingly conflicting findings regarding the ability of ezetimibe to reduce AD, it is crucial to examine the independent effect of ezetimibe. However, to our knowledge, there are no published studies of the effect of ezetimibe on either cardiovascular outcomes or all-cause mortality in the absence of concomitant statin use. Using data from a large health maintenance organization, we examined and compared the estimated effect of ezetimibe use and statin use on these events. The large patient population allowed us to closely match patients using each type of medication, and detailed longitudinal clinical information enabled us to account for changing levels of drug exposure over time.
Methods
This study was approved by the institutional review board at Henry Ford Health System and was in compliance with its Health Insurance Portability and Accountability Act policy. Study subjects were members of a large health maintenance organization that serves southeastern Michigan, including metropolitan Detroit. We identified all subjects with prescription coverage who had medication fills for ezetimibe or statins from November 1, 2002, to December 31, 2009. This time period was chosen to avoid the controversy over ezetimibe use starting at the end of 2009. The first prescription fill in either class was considered the index prescription (and the time of that fill was the index date), provided the patient had no record of earlier ezetimibe or statin use. Study subjects were aged ≥18 years on the index date and were continuously enrolled in the health plan for ≥6 months before that date. We excluded patients with previous diagnoses of liver disease and those who died or had major AD events within 3 months of the index prescription.
We adapted the method of Steiner and Prochazka to estimate medication exposure. Because study subjects were members of the health maintenance organization with prescription coverage, we had a near complete record of all medication fills. We have previously shown for another class of medications that we capture about 99% of prescriptions filled. In calculating the unweighted, continuous measure of medication exposure (CMME), we assigned every day of medication supply the value of 1. We then summed all days supplied in a 3-month window (ending on each day of observation) and divided by 90 days. Each unweighted CMME took into account prescription fills before, but extending into, the 3-month observation window, as well as prescription fills in which the supply ran past the observation window (i.e., to include and truncate the supply estimates, respectively). Given the longitudinal study design, we calculated either an ezetimibe or a statin CMME for each patient for every day of follow-up. Therefore, patients had multiple CMMEs, each of which represented a moving 3-month window of medication exposure for their particular lipid-lowering medications.
Unlike ezetimibe, which was available as a single dose (10 mg/day), statin medications included multiple drugs and multiple dose preparations per drug. To account for potential differences in the magnitude of the effect on outcome by drug and dose, we developed a weighting schema on the basis of the relative potency to reduce LDL levels. By request, we obtained data from a published meta-analysis by Weng et al. These data were used to model the relation between dose (natural log transformed) and LDL reduction for each of the statin drugs. We rescaled the response variable (LDL reduction) as a proportion of the maximum response (i.e., 55.7% reduction in LDL for the 40-mg dose of rosuvastatin). This resulted in a fitted weight for each statin drug dose which ranged from 0.3 to 1.0 (see Table E1 in the on-line supplement). The appropriate drug- and dose-specific weight was therefore assigned to each day of medication supply. Days without an available statin supply were given a value of 0; therefore, daily weights ranged from 0 to 1. The calculation of the weighted CMME was the sum of the weights over the preceding 3 months divided by 90 days. Again, a moving weighted CMME was calculated for each day follow-up for subjects using statin medications.
Although much less frequently used by patients, separate exposure measures were created for the following classes of cholesterol-lowering medications: bile acid sequestrants, fibrates, and niacin. Separate, unweighted CMMEs were calculated for these medication classes, and these measures were used to adjust our analytic models.
Available demographic information included patient age, gender, race or ethnicity, and median household income (for the census tract in which a patient resided). Laboratory data included measures of serum LDL cholesterol and creatinine levels. Creatinine level, age, gender, and race or ethnicity were used to derive estimated glomerular filtration rate (eGFR), a measure of kidney function, according to the Modification of Diet in Renal Disease (MDRD) formula. The creatinine level most proximal and within 1 year before and 3 months after the index date was used to calculate eGFR.
Inpatient and outpatient clinical diagnoses were used to identify baseline co-morbid conditions. Baseline refers to meeting the following criteria at any time before the index date. Subjects were considered to have diabetes if they had ≥2 recorded diagnoses of diabetes or were taking diabetes medications (i.e., an oral medication or insulin). Subjects were considered to have hypertension if they had ≥2 previous diagnoses of hypertension or were taking oral antihypertensive medication. Subjects were considered to have AD at baseline if they had previous diagnoses of coronary heart disease (i.e., myocardial infarction, unstable angina, or coronary artery revascularization), cerebrovascular disease (i.e., either a cerebrovascular accident or a transient ischemic attack), or peripheral vascular occlusive disease.
AD outcome variables were identified from de novo diagnoses and procedures, as we have done previously and which are described in Appendix E1 of the on-line supplement. We identified cardiovascular deaths from records maintained by the Division of Vital Records and Health Statistics, Michigan Department of Community Health, and through death records maintained electronically by the health system.
When possible, we performed 1:4 matching between ezetimibe users and statin users. Subjects were matched on age in years (categorized as ≤49, 50 to 59, 60 to 69, 70 to 79, and ≥80 years), gender, race or ethnicity (white, African American, and other or unknown), baseline LDL cholesterol levels (categorized as ≤70, 71 to 100, 101 to 130, 131 to 160, 161 to 190, and >190 mg/dl), presence of AD at baseline, presence of diabetes at baseline, presence of hypertension at baseline, and year of index prescription. When >4 statin initiator matches were available per ezetimibe initiator, 4 matches were randomly selected among the available pool. Of the 401 eligible patients taking ezetimibe, 257 (64.1%) had 4 matched subjects using statins, 110 (27.4%) subjects had <4 matches, and 34 (8.5%) did not have a matched statin user. Therefore, the analysis included 367 ezetimibe users and 1,238 statin user matches.
The primary study outcome was a composite of major AD events and all-cause death. The former included fatal and nonfatal acute myocardial infarction, unstable angina, coronary revascularization, fatal and nonfatal cerebrovascular accident, cerebrovascular revascularization, transient ischemic attack, and peripheral vascular occlusive disease events (i.e., hospitalization, revascularization, amputation, or death from peripheral vascular occlusive disease). Secondary outcomes included separate analyses for each of the following: major AD events, coronary heart disease events (i.e., fatal and nonfatal acute myocardial infarction, unstable angina, and coronary revascularization), cerebrovascular disease events (i.e., fatal and nonfatal cerebrovascular accident, transient ischemic attack, and cerebrovascular revascularization), peripheral vascular occlusive disease events, and all-cause death.
As has been described by Allison, our analyses used logistic regression to perform the time-dependent survival analysis, whereby each day of observation per patient was treated as a separate record. This approach has the benefit of more easily managing the multiple daily updated exposure variables over Cox regression models. We adjusted for matching by treating each of the matching categories as strata as implemented in PROC LOGISTIC in SAS (SAS Institute Inc., Cary, North Carolina). We also accounted for other potential confounders, such as the use of other lipid-lowering drugs and household income. The assessment period began 3 months after the index date (i.e., the earliest time possible for assessing outcomes given the 3-month drug exposure windows) and continued until the patient experienced the outcome being assessed, disenrolled from the health system, switched medications (i.e., ezetimibe to a statin or vice versa), or died.
We performed 3 separate analyses to assess and compare the relation between ezetimibe and statin use on major AD events and all-cause death. For our first set of comparisons, we simply coded whether subjects were ezetimibe or statin users with a single dichotomous variable (i.e., ezetimibe use = 1, statin use = 0). This provided a direct comparison of the 2 therapies, but it did not account for changing use over time, nor did it assess the effect of increasing levels of exposure within class.
The second set of analyses had separate continuous and time-updated measures for ezetimibe and statin use (i.e., unweighted CMME) for each day of study follow-up. These non-negative, continuous exposure measures mostly ranged from 0 to 1 (occasionally greater) and represented the range of daily use. Likewise, the effect estimates for ezetimibe and statins each represented the effect of going from no use to daily use. Differences in the effect estimates for the 2 treatments were compared using the Wald test. This approach accounted for changing patterns of use over time, measured the effect of each drug class, and compared differences in effect between drug classes.
The final, principal set of analyses as shown in the tables in the “Results” section also had separate CMME estimates for ezetimibe use and statin use. However, the CMME estimates for statin use were weighted to account for differences in dosage and differences in the relative strength of the various statin preparations. Therefore, these weighted exposure measures represented the range of no use to daily use of the strongest statin at the highest dose. Because ezetimibe had a single daily dose, the unweighted CMME estimates for ezetimibe use had the same interpretation. As a result, the effect estimates for the 2 drug classes represented the effect of going from no use to daily use at maximum strength (i.e., a similar scale for the within–drug class effects of increasing exposure). Between-treatment differences (i.e., ezetimibe vs statin) were compared using the Wald test. Therefore, this final, analytic approach accounted for changing patterns of medication use over time, measured the within–drug class effect while accounting for differences in dose and preparation, and compared differences in effect between drug classes. This approach was also used in analyzing the secondary outcomes and for the subgroup analyses. In the subgroup analyses, we assessed the relation between ezetimibe and statin use on the primary outcome after stratifying within the following categories: gender, race or ethnicity (African American and white), history of AD, diabetes status, and baseline eGFR (<60 and ≥60 ml/min/1.73 m 2 ).
On the basis of the available sample and application of an exemplary data set calculation, we estimated 80% power to detect an odds ratio (OR) of 0.26 for the ezetimibe dose effect and an OR of 0.47 for the statin dose effect on the composite outcome. The OR for comparing the ezetimibe and statin dose effects (i.e., the ratio of the dose effects) would need to be 0.24 for a difference between them to be detectable with 80% power. All calculations assumed a 2-sided analysis and a significance level (α) of 0.05. Analyses were performed using SAS version 9.2.
Results
As shown in Figure 1 , we identified 401 new ezetimibe users who met the inclusion criteria. Of these subjects, 367 (92%) had ≥1 suitable statin user match. We identified 21,872 potential statin initiators who similarly had no histories of previous or concomitant ezetimibe use, liver disease, or major AD events within 3 months of the first statin fill. We selected up to 4 statin user matches for each ezetimibe user, and we randomly selected among the available statin users when >4 potential matches were available. This process resulted in 1,238 statin user matches.
The baseline characteristics of the ezetimibe and statin users are listed in Table 1 . As can be seen, these subjects were similar for all of the matching characteristics. However, we did observe a significant difference in the estimated median income for the 2 groups. Although infrequent, the use of other lipid-lowering agents was significantly more common among ezetimibe users compared to statin users (11% vs 2%, respectively, p <0.001). The average duration of follow-up for all study participants was 51 ± 29 months (range 3 to 105).
| Characteristic | Patients Receiving Ezetimibe | Patients Receiving Statins | p Value |
|---|---|---|---|
| (n = 367) | (n = 1,238) | ||
| Age (yrs) | 58.9 ± 12.6 | 58.5 ± 12.4 | 0.585 |
| Women | 211 (57%) | 712 (58%) | 0.995 |
| Race/ethnicity | 0.350 | ||
| African American | 72 (20%) | 232 (19%) | |
| White | 278 (76%) | 967 (78%) | |
| Other/unknown | 17 (5%) | 39 (3%) | |
| Household income ($) ∗ | 59,000 ± 24,000 | 56,000 ± 25,000 | 0.023 |
| LDL cholesterol level (mg/dl) † | 0.354 | ||
| ≤70 | 10 (3%) | 15 (1%) | |
| 71–100 | 21 (6%) | 61 (5%) | |
| 101–130 | 80 (24%) | 284 (24%) | |
| 131–160 | 121 (36%) | 445 (38%) | |
| 161–190 | 88 (26%) | 295 (25%) | |
| >190 | 20 (6%) | 61 (5%) | |
| AD ‡ | 101 (28%) | 302 (24%) | 0.225 |
| Diabetes mellitus § | 87 (24%) | 250 (20%) | 0.147 |
| Hypertension || | 270 (74%) | 911 (74%) | 0.995 |
| eGFR (ml/min/1.73 m 2 ) ¶ | 78 ± 23 | 81 ± 37 | 0.088 |
| Use of ≥1 additional lipid-lowering medication # | 41 (11%) | 29 (2%) | <0.001 |
| Type of statin medication used | |||
| Atorvastatin | — | 287 (23%) | |
| Fluvastatin | — | 3 (<1%) | |
| Lovastatin | — | 60 (5%) | |
| Pravastatin | — | 29 (2%) | |
| Rosuvastatin | — | 7 (1%) | |
| Simvastatin | — | 852 (69%) | |
| Type of additional lipid-lowering medication used | |||
| Bile salts | 5 (1%) | 4 (<1%) | 0.034 |
| Fibrates | 33 (9%) | 21 (2%) | <0.001 |
| Niacin | 4 (1%) | 4 (<1%) | 0.086 |
∗ Based on the median household income in the census tract for the listed primary residence.
† Defined as the value in the year before and closest to the index prescription.
‡ Subjects were considered to have AD at baseline if they had previous diagnoses of coronary heart disease (i.e., myocardial infarction, unstable angina, or coronary revascularization), cerebrovascular disease (i.e., cerebrovascular accident, transient ischemic attack, or carotid revascularization), or peripheral vascular occlusive disease at any time before their index prescriptions for ezetimibe or statins.
§ Subjects were considered to have diabetes if they had ≥2 recorded diagnoses of diabetes or were taking diabetes medications (i.e., an oral medication or insulin).
|| Subjects were considered to have hypertension if they had ≥2 previous diagnoses of hypertension or were taking oral antihypertensive medications.
¶ Estimated using the MDRD formula.
# Denotes whether patients were taking any other class of lipid-lowering medication at the time of their initial prescription fills for ezetimibe or statins.
Assessing exposure dichotomously (i.e., ezetimibe use vs statin use; data not shown), we observed a nonsignificant association favoring ezetimibe use for reducing the primary composite outcome of major AD events or all-cause death compared to statin use (OR 0.88, 95% confidence interval [CI] 0.59 to 1.33). When assessing both drug exposures continuously but without accounting for statin dose and strength, we found a protective association between increasing ezetimibe use and the primary composite outcome (OR 0.38, 95% CI 0.16 to 0.95, p = 0.038) and a borderline significant relation for statin use (OR 0.74, 95% CI 0.50 to 1.03, p = 0.074).
Accounting for the strength and dose of statins in the same model, we found similar relations as described previously for ezetimibe use (OR 0.33, 95% CI 0.13 to 0.86, p = 0.024) and statin use (OR 0.61, 95% CI 0.36 to 1.04, p = 0.068) on the primary outcome ( Table 2 ). However, the magnitude of the effect estimates for ezetimibe use and statin use were not significantly different (Wald test p = 0.231).
| Outcome | Model | Number of Events | OR (95% CI) | p Value | Wald Test p Value ∗ |
|---|---|---|---|---|---|
| Primary outcome † | 180 | 0.231 | |||
| Ezetimibe | 0.33 (0.13–0.86) | 0.024 | |||
| Statin | 0.61 (0.36–1.04) | 0.068 | |||
| Secondary outcomes | |||||
| Major AD events ‡ | 137 | 0.336 | |||
| Ezetimibe | 0.55 (0.20–1.52) | 0.250 | |||
| Statin | 0.93 (0.51–1.69) | 0.801 | |||
| Coronary heart disease § | 75 | 0.509 | |||
| Ezetimibe | 0.96 (0.28–3.32) | 0.944 | |||
| Statin | 1.48 (0.66–3.32) | 0.341 | |||
| Cerebrovascular disease || | 65 | 0.324 | |||
| Ezetimibe | 0.13 (0.02–0.99) | 0.048 | |||
| Statin | 0.37 (0.16–0.89) | 0.027 | |||
| Peripheral vascular occlusive disease ¶ | 12 | 0.543 | |||
| Ezetimibe | 3.78 (0.35–40.63) | 0.272 | |||
| Statin | 1.70 (0.17–17.37) | 0.655 | |||
| All-cause death | 71 | 0.242 | |||
| Ezetimibe | 0.03 (0.01–0.49) | 0.013 | |||
| Statin | 0.17 (0.07–0.42) | <0.001 |
∗ In all of the models, ezetimibe use and statin use are estimated as CMMEs. However, statin use is weighted to account for differences in statin dose and preparation. The models also account for household income and the use of other lipid-lowering medications (i.e., separate CMMEs for bile acid sequestrants, fibrates, and niacin).
† The primary study outcome was a composite of all-cause death and major AD events (i.e., acute myocardial infarction, unstable angina, coronary revascularization, cerebrovascular accidents, transient ischemic attacks, carotid revascularization, and peripheral vascular occlusive disease events).
‡ Fatal and nonfatal acute myocardial infarction, unstable angina, coronary revascularization, fatal and nonfatal cerebrovascular accident, transient ischemic attack, carotid revascularization, and peripheral vascular occlusive disease events.
§ Composite of fatal and nonfatal acute myocardial infarction, unstable angina, and coronary revascularization.
|| Composite of fatal and nonfatal cerebrovascular accident, transient ischemic attack, and carotid revascularization.
¶ Composite of hospitalization, peripheral revascularization, amputation, and death.
We observed similar relations for increasing ezetimibe use and increasing statin use on most secondary outcomes ( Table 2 ). However, most of these relations were nonsignificant, likely due in part to the much smaller number of analyzable events. Notable exceptions were a significant protective association between both ezetimibe use and statin use and cerebrovascular events. There was also a significant protective association for ezetimibe use and statin use on all-cause mortality. None of the effect estimates for ezetimibe use and statin use were significantly different.
Table 3 lists the types of deaths among the subjects in each treatment group. We did not observe a significant difference in the primary cause of death by group.
