Sacco RL, Benson RT, Kargman DE, Boden-Albala B, Tuck C, Lin I, Cheng JF, Paik MC, Shea S, Berglund L. High-Density Lipoprotein Cholesterol and Ischemic Stroke in the ElderlyThe Northern Manhattan Stroke Study. JAMA. 2001;285(21):2729-2735. doi:10.1001/jama.285.21.2729
Author Affiliations: Department of Neurology (Drs Sacco, Benson, Kargman, and Lin and Ms Boden-Albala), Sergievsky Center (Dr Sacco), Department of Medicine, Division of General Medicine (Dr Shea), Preventive Medicine and Nutrition (Drs Tuck and Berglund), and Irving Center for Clinical Research (Dr Berglund), College of Physicians and Surgeons,and Divisions of Epidemiology (Drs Sacco and Shea), Socio-Medical Science (Ms Boden-Albala), and Biostatistics (Drs Lin, Paik, and Cheng) the Mailman School of Public Health, Columbia University, New York, NY.
Context Elevated high-density lipoprotein cholesterol (HDL-C) levels have been
shown to be protective against cardiovascular disease. However, the association
of specific lipoprotein classes and ischemic stroke has not been well defined,
particularly in higher-risk minority populations.
Objective To evaluate the association between HDL-C and ischemic stroke in an
elderly, racially or ethnically diverse population.
Design Population-based, incident case-control study conducted July 1993 through
Setting A multiethnic community in northern Manhattan, New York, NY.
Participants Cases (n = 539) of first ischemic stroke (67% aged ≥65 years; 55%
women; 53% Hispanic, 28% black, and 19% white) were enrolled and matched by
age, sex, and race or ethnicity to stroke-free community residents (controls;
n = 905).
Main Outcome Measure Independent association of fasting HDL-C levels, determined at enrollment,
with ischemic stroke, including atherosclerotic and nonatherosclerotic ischemic
Results After risk factor adjustment, a protective effect was observed for HDL-C
levels of at least 35 mg/dL (0.91 mmol/L) (odds ratio [OR], 0.53; 95% confidence
interval [CI], 0.39-0.72). A dose-response relationship was observed (OR,
0.65; 95% CI, 0.47-0.90 and OR, 0.31; 95% CI, 0.21-0.46) for HDL-C levels
of 35 to 49 mg/dL (0.91-1.28 mmol/L) and at least 50 mg/dL (1.29 mmol/L),
respectively. The protective effect of a higher HDL-C level was significant
among participants aged 75 years or older (OR, 0.51; 95% CI, 0.27-0.94), was
more potent for the atherosclerotic stroke subtype (OR, 0.20; 95% CI, 0.08-0.50),
and was present in all 3 racial or ethnic groups studied.
Conclusions Increased HDL-C levels are associated with reduced risk of ischemic
stroke in the elderly and among different racial or ethnic groups. These data
add to the evidence relating lipids to stroke and support HDL-C as an important
modifiable stroke risk factor.
Stroke is a major cause of death and disability in the United States
and is a growing public health concern. The latest projections estimate that
more than 1 million strokes will occur every year by the year 2010. Moreover,
stroke has a disproportionate impact on elderly, black, and Hispanic persons,
who are among the fastest growing segments of the US population. Many studies
have provided strong evidence for lipids as a risk factor for coronary artery
disease (CAD). These studies demonstrate a direct relationship between total
cholesterol, low-density lipoprotein cholesterol (LDL-C), and CAD and an inverse
relationship between high-density lipoprotein cholesterol (HDL-C) and CAD.1- 4 These
relationships have not been as clearly established for ischemic stroke with
some studies even questioning whether cholesterol is a risk factor for stroke.
The advent of statin agents, which significantly lower lipid levels
by inhibiting 3-hydroxy-3-methylglutaryl coenzyme A (HMG-CoA) reductase, has
focused more research on lipids in all vascular outcomes, in particular stroke.
Compelling data from statin trials show impressive reductions in stroke risk
among persons with CAD.5 Some researchers have
suggested that the improvement in cardiovascular and cerebrovascular end points
with statin agents cannot be completely explained by the baseline or treated
LDL-C levels alone.6- 8
Furthermore, several studies have suggested that low levels of HDL-C without
high levels of LDL-C characterizes 20% to 30% of patients with CAD in the
United States.9,10 The benefits
of statins for stroke reduction have reinitiated discussions of the role of
lipids as a stroke risk factor.
Few studies have examined the relation between fractionated cholesterol
and stroke. None of these studies analyzed data based on stroke subtype and
age category or in a population that included black, Hispanic, and white persons
all residing in the same geographic sampling frame. The purpose of this study
was to examine the association of specific lipoprotein classes and ischemic
stroke among elderly Hispanic, black, and white persons, residing in the same
community, as part of the Northern Manhattan Stroke Study (NOMASS).
We conducted a population-based case-control study in the northern Manhattan
community.11 Methods for case detection in
NOMASS have been previously described.12 Briefly,
688 incident cases were prospectively enrolled between July 1, 1993, and June
30, 1997, based on the following criteria: (1) diagnosed as having a first
cerebral infarction, fatal or nonfatal; (2) older than age 39 years at onset
of stroke; and (3) a resident of Northern Manhattan in a household with a
telephone. Patients with intracerebral or subarachnoid hemorrhage and transient
ischemic attack were excluded.
Case surveillance included screening of all admissions, discharges,
and head computed tomographic scans at the Columbia-Presbyterian Medical Center,
the only hospital in the community. Cases were also identified through discharge
lists from other hospitals and through a comprehensive community-based surveillance
system for nonhospitalized persons with stroke. Because lipid profiles were
not obtained from patients who were admitted at some of the other hospitals,
only 539 of the cases were analyzed in this study. To ensure that the 539
cases did not differ substantially from the entire NOMASS cohort, we compared
baseline variables available in both groups. We found that the study group
was representative of the entire sample with no significant differences in
the baseline demographics, risk factors, and total cholesterol levels.
Methods of control recruitment and enrollment have also been described
in previous publications.13,14
Stroke-free community subjects were identified by random-digit dialing using
dual frame sampling to identify both published and unpublished telephone numbers.
When a household was contacted, the research objectives were explained and
a resident aged 39 years or older was interviewed briefly to record age, sex,
race or ethnicity, and risk factors. These telephone interviews were conducted
by Audits and Surveys, Inc, New York, NY, using trained bilingual interviewers.
Approximately 46 453 numbers were dialed to reach 9608 households, 876
people refused an interview (telephone response rate 91%).
Telephone interview data from control-eligible subjects were downloaded
to the NOMASS computer system and assigned to cells defined by age, sex, and
race or ethnicity. Patients were randomly selected from cells matched to the
accumulating case group by age, sex, and race or ethnicity and were recontacted
by the NOMASS staff. The in-person participation rate for selected and matched
controls was 75%. Therefore, our overall control response rate was 68%. Appointments
were made for in person evaluations at the hospital or home for those who
could not come in-person (7% were done at home). Approximately 80% of the
cases were matched to 2 controls, and 12% were matched to 1 control; 8% were
matched to other case-control strata. The institutional review boards of Columbia-Presbyterian
Medical Center and the other hospitals approved the study, and written informed
consent was obtained at the time of the in-person visit.
Data were collected through interviews of cases and controls by trained
bilingual research assistants using standardized data collection instruments,
medical record review, physical and neurological examination by the study
physicians, in-person measurements, and fasting blood specimens for lipid,
glucose, and cholesterol level determinations as previously described.15 Race or ethnicity, hypertension, diabetes mellitus,
any cardiac disease, physical activity, smoking status, alcohol use, and body
mass index (BMI) were defined using standardized criteria as outlined in prior
When the subject was unable to answer questions, a proxy who was knowledgeable
about the patient's history was interviewed. Proxy respondents were used for
26% of cases and 1% of controls. Reliability studies between proxies and subjects
showed excellent concordance for specific risk factors questions including
alcohol consumption (r = 0.81; P<.001) and physical activity (r = 0.69; P<.01). Moreover, measurement of serum glucose and lipids
were used in the definitions of stroke risk factors.
Cases of ischemic stroke were classified into infarct subtype categories
based on the results of their neurovascular evaluation. Stroke subtype was
determined after review of all the available data by a diagnostic committee
as described in prior reports.16 In this analysis,
patients were subdivided into 2 groups: infarction due to atherosclerosis
that included extracranial or intracranial atherosclerosis, and nonatherosclerotic
infarction that included cardioembolism, lacune, and cryptogenic infarction.
Fasting blood samples were drawn within several days of study enrollment,
usually within 72 hours of admission. Total cholesterol and triglyceride levels
were determined using standard enzymatic procedures in an automated spectrometer
(Hitachi 705; Boehringer, Mannheim, Germany). Plasma HDL-C cholesterol levels
were measured after precipitation of apolipoprotein B (apo B) containing lipoproteins
by phosphotungstic acid. Low-density lipoprotein cholesterol concentrations
were calculated by the Friedewald formula.17
Univariate and multivariable conditional logistic regression models
for matched case-control data were used to calculate the odds ratios (ORs)
and 95% confidence intervals (CIs) for HDL-C level and ischemic stroke after
adjusting for potential confounding variables (hypertension, diabetes, heart
disease, current smoking status, BMI, physical activity, and education level
as a marker of socioeconomic status18). Several
analyses also adjusted for total cholesterol, LDL-C, and triglyceride levels.
High-density lipoprotein cholesterol level was examined continuously, dichotomously,
and as 3 categories, based on standard criteria set by the second report of
the National Cholesterol Education Program (NCEP).19
The categorical approach was preferred because it permitted the calculation
of ORs that gave more clinical interpretable measures of risk. Analyses were
conducted overall and stratified by age, sex, race or ethnicity, and ischemic
stroke subtype. All tests were 2 sided; significance was determined to be P < .05. Statistical analyses were performed using SAS
software (SAS Institute, Cary, NC).
We analyzed 539 ischemic stroke cases and 905 controls (Table 1). Among the cases, 67% were aged 65 years or older; 55%
were women; 53%, Hispanic; 28%, black; and 19%, white. Subjects in the "Other"
race or ethnic category were not included in the analyses discussed in this
article because of the small numbers.
The median total cholesterol, HDL-C, and LDL-C levels were lower in
cases than in the controls, while the median triglyceride levels were higher
in the cases (Table 2). A higher
percentage of participants in the control group than in the case group had
HDL-C levels of 50 mg/dL (1.29 mmol/L) or higher than those in the case group.
When HDL-C levels were examined dichotomously, unadjusted for other risk factors,
a protective effect was seen for ischemic stroke in patients with levels of
at least 35 mg/dL (0.91 mmol/L) or higher (OR, 0.50; 95% CI, 0.39-0.65) (Table 3). The protective effect was greater
for patients with HDL-C levels of 50 mg/dL (1.29 mmol/L) or higher than patients
with HDL-C levels between 35 and 49 mg/dL (0.91-1.27 mmol/L). In multivariable
analysis, adjusted for hypertension, diabetes, cardiac disease, current smoking
status, BMI, physical activity, and educational level, the protective effects
of HDL-C were similar. Although the use of aspirin and cholesterol-lowering
medications was low in this community sample (20% taking aspirin; 9%, cholesterol-lowering
medications), multivariate analyses adjusting for these medications did not
alter our model.
In multivariable analyses, adjusting for LDL-C levels of 130 mg/dL (3.4
mmol/L) or higher and triglyceride levels of 200 mg/dL (2.3 mmol/L) or higher
had no effect on the protective effect of HDL-C against ischemic stroke (Table 3). When total cholesterol levels
of 240 mg/dL (6.2 mmol/L) or higher were added to the multivariate model,
the ORs were unchanged (Data not shown). When HDL-C levels were examined as
a continuous variable in a multivariate model including stroke risk factors
and LDL-C and triglyceride levels, a protective effect was found with an OR
of 0.81 (95% CI, 0.77-0.86) for a 5-mg/dL (0.13-mmol/L) increase in the HDL-C
level. This translated into a 19% odds reduction for stroke.
The protective effects of HDL-C levels were analyzed by age, sex, and
race or ethnic subgroups and were found to be similar for all groups with
a statistically significant protective effect evident in patients aged 75
years or older (adjusted OR, 0.51; 95% CI, 0.27-0.94) (Table 4). A similar dose-response relationship was observed in all
3 race or ethnic groups (Figure 1).
When ischemic stroke was stratified by atherosclerotic ischemic stroke and
nonatherosclerotic ischemic stroke (nonatherosclerotic infarction) subtype
in the multivariable model, a clear protective effect was seen for both stroke
subtypes (Table 5). The protective
effect of HDL-C was significantly more pronounced for the atherosclerotic
stroke subtype than the nonatherosclerotic subtype (test for heterogeneity, z = 2.18; P = .02).
This population-based case-control study has demonstrated a protective
effect of greater HDL-C level for ischemic stroke in an elderly, multiethnic
population of men and women. Higher HDL-C levels were significantly more protective
against atherosclerotic ischemic stroke than nonatherosclerotic stroke subtypes.
The protective dose-response relationship was observed even when HDL-C levels
were adjusted for total cholesterol or LDL-C and triglyceride levels suggesting
that this protective effect is not mediated by relationships between HDL-C
and other measured lipid levels. Our effect estimates for HDL-C levels were
adjusted for multiple potential confounders and for other known stroke risk
factors including hypertension, diabetes, cardiac disease, current smoking
status, physical activity, BMI, and educational level as a marker of socioeconomic
status. These adjustments had little effect on the magnitude of the protective
effect of HDL-C levels for ischemic stroke. These results imply a need to
examine fractionated cholesterol levels when assessing stroke risk, since
HDL-C is a potentially modifiable stroke risk factor.
The relationship between abnormalities of serum lipids and stroke has
been less clear than for CAD.20 Some prospective
cohort studies including the Framingham Heart Study have found no association
between total serum cholesterol or HDL-C level and cerebral infarction.21 Others have found a modest relationship. In the Multiple
Risk Factor Intervention Trial, mortality from ischemic stroke was greater
among men with high total cholesterol levels.22
In the Honolulu Heart Program, there was a continuous and progressive increase
in thromboembolic stroke rates with increasing levels of total cholesterol
level, with a relative risk of 1.4 comparing highest and lowest quartiles.23 Meta-analyses among prospective studies have found
either no or only a minimally increased relative risk of stroke due to elevated
total cholesterol level.24,25
Consistent with these meta-analyses, we found no relationship between total
cholesterol levels and stroke risk. The absence of a consistent relationship
between total cholesterol levels and stroke in these other studies may be
partially explained by the heterogeneity of stroke, the reliance on total
cholesterol measurements instead of lipoprotein fractions, and the focus on
cardiovascular events rather than on stroke events, which occur more frequently
in those aged 65 years or older.
The protective effects of increased HDL-C levels on the risk of myocardial
infarction have been established by numerous epidemiologic studies.26,27 Few studies have examined the relationship
between HDL-C level and stroke. A few case-control studies have found the
concentration of HDL-C to be lower in persons who had a stroke, even after
controlling for other stroke risk factors.28- 32
A strong inverse relationship between HDL-C level and ischemic stroke was
observed among predominately white patients in an Israeli prospective cohort
and in the Copenhagen City Heart Study.33,34
Studies using carotid artery ultrasound technology have demonstrated an inverse
relationship between HDL-C level and extracranial carotid artery atherosclerosis.35- 39
No published epidemiological studies have evaluated the relationship between
HDL-C level and ischemic stroke risk in multiethnic groups or evaluated atherosclerotic
stroke subtypes as we have in our study.
Although the mechanisms by which HDL-C protects against ischemic stroke
are unclear, several possibilities have been suggested. High-density lipoprotein
cholesterol may prevent the oxidation of LDL-C, which has been linked to atherogenesis,
or HDL-C may increase the reverse transport of LDL-C from peripheral tissues
to the liver where degradation occurs.4 It
has also been suggested that HDL-C may transport antioxidants to LDL-C, making
LDL-C less susceptible to oxidation within the endothelium.40
Most likely, the protective effects of HDL-C levels are multifactorial and
other potential mechanisms remain to be elucidated. Furthermore, HDL-C levels
may be modifiable. Exercise,41 weight reduction,42 moderate alcohol consumption,43
smoking cessation,44 and statin agents45 have all been shown to increase HDL-C levels.
An interesting finding in our study was that HDL-C levels showed more
protection for atherosclerotic stroke than nonatherosclerotic infarction but
was significantly protective against both subtypes. The more pronounced effect
in the atherosclerotic stroke cases underscores the value of evaluating the
relationship of stroke risk factors in certain specific stroke subtypes. Not
all strokes are directly due to atherosclerosis; therefore, a weaker effect
for lipids in some prior epidemiological stroke studies could be due to the
lower prevalence of atherosclerotic stroke subtypes. Compared with other studies,
our cohort included more black and Hispanic persons who have been found to
have a greater incidence of intracranial atherosclerotic stroke.46
Plaque stabilization, regression, and protection of LDL-C from peroxidation,
or both, which has been proposed in the genesis of atheroma formation, may
explain why HDL-C levels would be protective against atherosclerotic stroke,
but it does not easily explain why HDL-C would be protective against nonatherosclerotic
infarction. Nonatherosclerotic infarction includes cardioembolic, cryptogenic,
and lacunar infarcts, each of which have a different mechanism. These patients,
however, could have atherosclerotic cardiovascular disease, microatheroma
of the cerebral vessels, or atherosclerotic plaque in large vessels not easily
detected by conventional vascular studies. High-density lipoprotein cholesterol
could possibly protect against nonatherosclerotic infarction by causing stabilization
or regression of plaque in some of these other conditions.
Clinical trials analyzing the efficacy of lipid-lowering strategies
with statins have demonstrated impressive reductions in stroke risk among
various high-risk populations with cardiac disease. In these studies, stroke
was either a secondary end point, or a nonspecified end point determined based
on post hoc analyses.7,8 Meta-analyses
of some of these trials have found significant reductions in stroke risk.5,47 Two large trials in which stroke was
prespecified as a secondary end point, Cholesterol and Recurrent Events (CARE)
and Long-term Intervention with Pravastatin in Ischemic Disease (LIPID), have
also shown significant reductions with pravastatin sodium among patients with
coronary artery disease and normal to only modest elevations of cholesterol
levels.48,49 Differences may exist
in the magnitude of the effect of HDL-C level in primary prevention trials.
In the West of Scotland study,8 for example,
a more modest, nonsignificant stroke risk reduction of only 11% was found.
Besides reducing total cholesterol and LDL-C levels, modest elevations of
HDL-C levels were reported among those treated with statins.
Moreover, the efficacy of certain statins for stroke prevention has
been demonstrated to be significantly greater among those with lower initial
levels of HDL-C.50 The Veterans Affairs–High
density lipoprotein cholesterol Trial (VA-HIT)51
showed an independent protective effect of HDL-C level for stroke and other
vascular outcomes over the 5 years of the study. This study examined the effect
of increasing HDL-C levels selectively with gemfibrozil, in men with CAD and
low levels of LDL-C. The investigators found a 24% reduction in the combined
outcomes of death due to CAD, nonfatal myocardial infarction, or stroke with
an average 6% increase in HDL-C levels in the first year.52
Our study demonstrated a 14% to 23% odds reduction in stroke with every 5-mg/dL
(0.13 mmol/L) rise in HDL-C, which is equivalent to a 12% increase. These
clinical trials demonstrate that stroke can be prevented with modern approaches
to lipid lowering and that some of this benefit may be mediated through effects
on HDL-C. Our data provide additional evidence for the importance of HDL-C
in determining stroke risk.
Several limitations of our study design deserve discussion. The population-based
approach and the matching by age, sex, and race or ethnicity, however, minimize
the potential biases often associated with case-control studies. Our HDL-C
measurements were performed after the stroke and may not accurately reflect
prestroke exposure. By design, however, we made special efforts to collect
fasting blood within 72 hours of admission. The stability of lipids after
stroke is controversial. A few studies have reported that some lipid levels
may not be stable after stroke; however, these studies have demonstrated that
HDL-C levels are more stable after stroke than triglyceride or total cholesterol
levels.53,54 In a small study,55 we also demonstrated the stability of HDL-C measurements
in the immediate period after stroke.
Some stroke patients enrolled in the study did not have a complete lipid
panel and were not included in the analysis of HDL-C; however, total cholesterol
values were no different in those included in this subset compared with those
without a complete lipid panel. Moreover, it is unlikely that any differential
selection of controls with elevated HDL-C could have occurred to bias our
results. Furthermore, despite the differential proportion of risk factor information
collected by proxy in our cases vs controls, the strong concordance rates
for proxy reliability and the reliance on laboratory results for HDL-C levels
limit any bias in the collection of lipids and stroke risk.
Our results were adjusted for several known stroke-risk factors; however,
some unknown or unmeasured confounders could have been operational. The strength
and consistency of the effect of HDL-C across multiple strata minimizes this
possibility. The effect of lipid treatments could not be fully evaluated.
Information was obtained about lipid-lowering agents, but because so few of
the patients in this community were taking these agents at the time of data
collection, this variable was not used in the final analyses.
Our study has important public health implications for minority groups
and elderly persons. Blacks and Hispanics have a greater mortality and incidence
of stroke.56,57 According to recent
US projections, the Hispanic population is one of the fastest growing minority
groups in this country. The Northern Manhattan Stroke Study has provided unique
information on the incidence of stroke among whites, blacks, and Hispanics,
and now has shown a significant modifiable risk factor that is pertinent to
each of these ethnic groups and both sexes.12
This study has also shown a protective effect of high HDL-C levels for ischemic
stroke among the "older-elderly," a finding significant in that several researchers
have questioned the need to diagnose or treat dyslipidemia in individuals
older than 70 years.58- 60
These studies, which were looking primarily at cardiac events, have suggested
that other risk factors may become more significant than lipids in persons
older than 70 years. Our data suggest that HDL-C is still strongly protective
against ischemic stroke among those older than 75 years.
The second report of the NCEP paid greater attention to HDL-C levels
as a risk factor.19 The NCEP recommended the
addition of HDL-C to initial cholesterol testing, the designation of high
HDL-C levels as a protective factor, and the increased emphasis on physical
activity and weight loss as components of the dietary therapy of high blood
cholesterol levels. In the third report of the NCEP, the level of HDL-C has
been changed to 40 mg/dL (1.03 mmol/L) and the goal for LDL-C–lowering
therapy has been modified for those with low levels of HDL-C.61
The National Stroke Association's stroke prevention guidelines to reduce the
risk of first stroke references the NCEP guidelines and recommends statin
agents for individuals with high cholesterol levels and atherosclerotic cardiac
This epidemiologic study provides further data to support these guidelines
to prevent stroke. More effective detection of low HDL-C levels and treatments
to modify this stroke risk factor could affect significantly the clinical
and public health burdens of stroke.