Lacunar Infarcts Defined by Magnetic Resonance Imaging of 3660 Elderly People:  The Cardiovascular Health Study FREE

FISHER^1- 3 DEFINED lacunes pathologically as areas of infarction of less than 2 cm in size and resulting from occlusion of 1 of the small penetrating branches of large cerebral arteries. He described these small deep infarcts of the brain as common—being single or multiple, clinically silent or, less frequently, symptomatic. Although for many clinicians the diagnosis of lacunar infarct implies hypertensive small-vessel disease, risk factors specific for lacunar infarcts have not been investigated extensively.⁴ Some^5- 6 have suggested that the clinical syndrome and risk factors may differ for patients with single as opposed to multiple lacunes and that the same mechanisms in effect for nonlacunar infarcts, such as artery-to-artery emboli and cardiac emboli, may also play a role in lacunes.^7- 9 Large pathologic series might be useful in unraveling the risk factors for lacunar infarcts, but such studies are scarce due to the low early fatality rate in patients with lacunes.¹⁰ Consequently, in most studies of risk factors, lacunes have been defined clinically rather than pathologically. Many of these studies are flawed by small numbers of subjects, lack of appropriate control groups, and use of computed tomography that may underestimate the burden of small deep infarcts, relative to magnetic resonance imaging (MRI).⁴

The Cardiovascular Health Study (CHS) is a prospective, multicenter, epidemiological study of risk factors for coronary and cerebrovascular disease in older adults. As part of the study, 3660 participants have undergone cranial MRI. The imaging characteristics of lesions on MRI consistent with infarcts has been detailed previously.¹¹ Clinical manifestations associated with these lesions are common, regardless of whether the participant has a history of stroke.¹² Our article focuses on the subset of subjects in CHS with only lacunar infarcts on MRI. The CHS offers a unique opportunity to evaluate the potential risk factors for lacunar infarcts defined by MRI, to examine how risk factors may differ for single vs multiple lacunes and silent vs symptomatic lacunes, and to assess neurologic dysfunction in subjects with silent lacunes, namely those whose MRI shows 1 or more lacunes but who deny having had a transient ischemic attack (TIA) or stroke.

Members of the CHS cohort were recruited from a random sample of the Health Care Financing Administration Medicare eligibility lists in 4 US communities: Forsyth County, North Carolina; Sacramento County, California; Washington County, Maryland; and Pittsburgh (Allegheny County), Pennsylvania. Participants had to be 65 years or older, able to give informed consent, and able to respond to questions without the aid of a surrogate respondent. They could not be institutionalized, wheelchair-bound in the home, or under treatment for cancer. Details about the study design and characteristics of the 5888 participants are published elsewhere.^13- 15

Eligible and consenting participants underwent an extensive baseline evaluation including standard questionnaires, physical examination, and laboratory testing, as detailed elsewhere.^13- 14 Subjects' cognitive functions were also evaluated, using a modified Mini-Mental State Examination^16- 17 and the Digit-Symbol Substitution Test.¹⁸ Parts of the baseline evaluation have been repeated at various times since the initial examinations were conducted between June 1989 and May 1990 (year 2 of the study). For the analyses concerning potential risk factors and clinical manifestations, in general, we used information that was collected at the examination closest in time but preceding the MRI scan.

Coronary heart disease, myocardial infarction, congestive heart failure, stroke, TIA, claudication, hypertension, and diabetes were defined using patients' reports, hospital and clinic records, and data collected as part of the CHS study. Coronary heart disease included myocardial infarction, angina, coronary bypass graft, or angioplasty. For these analyses, a condition was said to be present at baseline if the subject reported having been told by a physician that he/she had that condition regardless of whether that report could be confirmed. More details on the criteria used to define these conditions were reported previously.^13- 14,19

Several sources of information were used to decide whether subjects reported having had a TIA with symptoms lasting less than 24 hours or a stroke with symptoms lasting 24 hours or more. Thus, subjects were classified as being free of TIA or stroke if they lacked a self-report of these conditions at baseline, whether confirmed or not, and lacked an incident TIA or stroke prior to the MRI that was performed as part of the study. Subjects were asked on a yearly basis whether they had experienced a TIA or stroke. Detailed information was collected from all giving a positive response and was used to adjudicate the event.¹⁹ Subjects with evidence on imaging, whether performed as part of the study or not, of brain infarcts but without a self-report of a TIA or stroke were classified as having silent infarcts.

Cranial MRI scans were performed during years 4, 5, and 6 of the study. Those without contraindication and who consented underwent imaging in a standard fashion.^20- 21 The scanning protocol included standard sagittal T₁-weighted images and axial T₁-weighted, spin density, and T₂-weighted images—all with 5-mm thickness and no interslice gaps. Imaging data were sent to a single reading center for interpretation by neuroradiologists with training in the CHS protocol and without knowledge of the subjects' age, sex, race, ethnicity, or other clinical information.

In this study, evidence for cerebrovascular disease showed on MRI scan was defined as an area of abnormal signal intensity in a vascular distribution that lacked mass effect.^11,20- 21 Briefly, nonhemorrhagic infarcts of the cortical gray and deep nuclear regions had to be brighter on spin density and T₂-weighted images than normal gray matter. The requirement for hyperintensity on spin density-weighted images was intended to distinguish small deep nuclear region infarcts from dilated perivascular spaces. Nonhemorrhagic infarcts in the white matter were similarly defined, except that they also had to be hypointense on T₁-weighted images, approaching the hypointensity of cerebrospinal fluid, thus distinguishing them from diffuse white matter disease. Hemorrhagic lesions had heterogeneous increased signal on T₁-weighted images, heterogeneous decreased signal on T₂-weighted images, or both findings. White matter changes were graded in a standard fashion on a 10-point scale.¹⁵

This article concerns only infarcts 3 mm in size or greater. For each subject up to 5 such infarcts were identified and classified according to size (maximum of the anterior-to-posterior, right-to-left, and rostral-caudal dimen-sions) and 16 anatomic locations.¹¹ In this article, all infarcts were further classified as being less than 20 mm in all dimensions or not and exclusively subcortical (cerebral and cerebellar) or not. For these analyses, a lacunar infarct or lacune was defined as exclusively subcortical and small, less than 20 mm in all dimensions.

To examine the association of lacunes with potential risk factors, we performed logistic regression analysis with age, sex, and each of the potential risk factors as the independent variables. The factors examined are listed later in a table. Because of the potential for age and sex to confound the relation between lacunes and potential risk factors, we included age and sex in all analyses. Five case definitions or dependent variables were examined, but all models included the same comparison or control group, those subjects who denied having had a TIA or stroke and whose MRI lacked evidence for an infarct of any type. Case groups included only those subjects with MRI evidence for lacunar infarcts. Excluded from these analyses were all subjects with other types of infarcts or combinations of lacunes with other types of infarcts, as well as all subjects without evidence for any infarct on MRI but with a self-report of a TIA or stroke. Subgroups of interest were defined by number of lacunes (single vs multiple) and by report of a TIA or stroke (silent vs symptomatic). Thus, analyses involved 5 dependent variables defined by a case group (≥1 lacunes) or 4 subgroups (single lacunes, multiple lacunes, silent lacunes, and symptomatic lacunes). Lacunes were classified as silent if the subject denied and as symptomatic if the subject reported having had a TIA or stroke at some time before the MRI scan.

After evaluating individually each risk factor, all risk factors were presented to an initial logistic regression analysis model to identify which were independently associated with lacunar infarcts. A stepwise method was used with a P value of .01. Next the analysis was rerun dropping from the list of candidates available to enter the model all the echocardiogram variables because none of these variables was significant in the prior model and because they were missing in a sizable proportion of the cohort. Finally, so as to maximize the number of subjects, the analysis was repeated making available only those variables identified as significant from the previous stepwise model. These steps were repeated for the 5 dependent variables defined above. Variables that had entered 1 or more of these stepwise models were then forced into logistic regression models for each of the 5 dependent variables to allow a comparison of the same set of variables across all dependent variables. In these final models, all the independent variables that were continuous were recoded into quartiles based on the distribution of the variable in the entire group of subjects eligible for these analyses.

We also examined the question of whether participants with silent lacunes demonstrated impaired function or clinical manifestations, using the clinical manifestation, rather than lacune, as the dependent variable and with age, sex, and silent lacunes, whether single or multiple, as the independent variables. The manifestations examined are listed later in a table. Additional analyses were performed on variables whose partial correlation coefficients for silent lacune adjusted for age and sex were significant. Adjustments were made for other variables that might also affect the function being evaluated. These additional variables included education (number of years of school up to 16 for a college graduate, and those with more education coded as 17) and the number of times the participant had performed the test being assessed prior to when the MRI was performed. For measures of cognitive function, a depression score was also added.²² Also factors independently associated with silent lacunes were forced into the models to control for potential confounding. Finally, given the association of many of these factors with white matter changes showed on MRI, as described previously,¹⁵ this factor was also included in all models.

Results of the regression analysis models described above are summarized in various ways. The simplest summary is with partial correlation coefficients and P values for the variables of interest. The greater the absolute value of the partial correlation coefficient, the stronger is the association. For the more detailed multivariate models, the regression coefficients are also presented. The regression coefficient indicates the average change in the dependent variable for a change of 1 unit in the independent variable, when the other independent variables are fixed. Finally, in the analyses where the same variables are forced into each model, the associations are summarized with odds ratios and 95% confidence intervals.

All these analyses were based on the updated CHS database that incorporates minor corrections through May 30, 1997.

Of 5888 members of the CHS cohort, 3660 (62%) underwent MRI scanning. Those who were scanned were significantly younger, more educated, more likely to have never smoked, and healthier than those who were not scanned, as detailed previously.^11,15 Considering all 3660 MRI scans, 2529 subjects (69.1%) had scans free of any infarcts (Table 1). The remaining 1131 subjects (30.9%) had MRI evidence for 1 or more infarcts, of which 841 subjects (74.4%) had only lacunes. Of 841 subjects with only lacunes, 554 had a single lacune (65.9%) and 751 had silent lacunes (89.3%), with the subject denying having had a TIA or stroke before the MRI.

Further analyses were restricted to 2403 subjects without report of TIA or stroke and without any infarcts by MRI (the comparison group) and the 841 subjects with only lacunar infarcts demonstrated on MRI (the case group). Of 841 with lacunar infarcts, 554 (66%) had 1 lacune, 184 (22%) 2, 71 (8%) 3, 25 (3%) 4, and 7 (1%) 5. Thus, the 841 subjects had among them 1270 total lacunes. Concerning these lacunes and the maximum of the right-left and anterior-to-posterior dimensions, 259 (20%) were 3 to less than 5 mm, 741 (58%) were 5 to less than 10 mm, 204 (16%) were 10 to less than 15 mm, and 66 (5%) were 15 to less than 20 mm in largest diameter. Of these lacunes, neuroradiologists coded 1109 (87%) as being in a single subcortical location, 132 (10%) in 2, 28 (2%) in 3, and 1 in 4.

The location of 1109 lacunes coded as being in a single location on MRI and whether they were silent or symptomatic is summarized in Table 2. Most of these lacunes were located in the lentiform nuclei (48.8%) or thalamus (18.8%) with those in the lentiform nuclei being slightly more likely to be present in subjects without reports of a TIA or stroke and those in the thalamus to be present in subjects with such reports.

Table 3 summarizes the associations between potential risk factors and lacunes for the case group and 4 subgroups. Age was strongly associated with all lacunes and with each subgroup, with partial correlation coefficients adjusted for sex ranging from 0.09 to 0.14 (for all, P <.001). For all lacunes and the 4 subtypes, strong associations, adjusted for age and sex, were seen for the ankle-to-arm ratio, the common and internal carotid artery wall thickness, serum creatinine, and various measures of blood pressure. Associations that were significant for all 4 subtypes but more varied included coronary heart disease at baseline, pack-years of smoking, and white blood cell count.

We performed additional analyses to examine the association with internal carotid artery stenosis. We examined the group with 1 or more lacunes on just one side of the brain and computed partial correlation coefficients as above, adjusting for age and sex, with ipsilateral and contralateral internal carotid artery stenosis of 50% or greater. For entirely right-sided lacunes, the partial correlation coefficient was 0.041 (total N=2589; P=.04) for ipsilateral and 0.015 (total N=2590; P =.45) for contralateral internal carotid artery stenosis. For entirely left-sided lacunes, the partial correlation coefficient was 0.059 (total N=2579; P <.01) for ipsilateral and 0.062 (total N=2580; P<.01) for contralateral internal carotid artery stenosis. Thus, the correlations were not consistently higher for ipsilateral than contralateral carotid artery stenosis.

We next turned to multivariate models to identify which of the variables in Table 3 were independently associated with lacunes in the case group and 4 subgroups. The results are in Table 4. Considering all groups, the strongest associations were for age. Having MRI evidence of 1 or more lacunes was independently associated with having a higher diastolic blood pressure, higher creatinine level, a maximum internal carotid artery stenosis of 50% or greater, more pack-years of smoking, diabetes present at baseline, and being a woman. Considering just the variables that are dichotomous, the odds ratios for a maximum internal carotid artery stenosis of 50% or greater was 1.81, for diabetes was 1.33, and for male sex was 0.74. Besides age and sex, which were included in all models (Table 4), the models for single and multiple lacunes had no factors in common. The same was true for the models for silent and symptomatic lacunes.

Table 5 shows the results when models for the case group and 4 subgroups are forced to have the same factors, namely all those factors entering into any of the models in Table 4. For these analyses, all continuous independent variables have been converted to quartiles and associations are summarized with odds ratios and 95% confidence intervals. The findings for all lacunes and the large and overlapping subgroups of silent lacunes and single lacunes are all similar, with the dominant risk factors being age, diastolic blood pressure, creatinine, maximum internal carotid artery stenosis, and coronary heart disease at baseline. The model for multiple lacunes differs from these other relatively similar models by showing stronger associations for sex and low-density lipoprotein cholesterol and not for maximum internal carotid artery stenosis. Finally, the model for symptomatic lacunes has stronger associations than in the other models for use of any diuretic in the last 2 weeks and diabetes at baseline. The differences among these models were minor with odds ratios being similar and 95% confidence intervals substantially overlapping.

Table 6 shows the partial correlation coefficients adjusted for age and sex between silent lacunes and a variety of potential clinical manifestations. Significant associations were present in all categories of function examined, including cognitive, upper extremity, lower extremity, and overall function. Some of the strongest associations were for the modified Mini-Mental State Examination, the Digit Symbol Substitution Test, time to put on and button a shirt, and time to walk 450 cm. Except for the modified Mini-Mental State Examination, the associations between impaired function and silent lacunes remained statistically significant after controlling for other potentially confounding variables as displayed in Table 7.

Consistent with other studies^23- 30 on brains of the elderly with imaging or autopsy, we found that infarcts are common. Of 3660 older adults who underwent MRI in the CHS, 1131 (31%) had findings consistent with an ischemic infarct 3 mm in diameter or greater. Of 1131 participants with infarcts, 841 (74%) had only lacunes, being small (<20 mm) and subcortical, and of 841 with only lacunes, 751 (89%) were silent, subjects denying a history of TIA or stroke. These findings suggest that, more often than not, MRI-defined lacunes either produce no symptoms or produce symptoms that are not recognized by patients or their physicians as attributable to a stroke. However, considering those who reported having had a TIA or stroke, we cannot easily determine if a particular infarct seen on the MRI explains their symptoms. Consequently, even a larger proportion of lacunes may have been silent.

The mechanism of lacunes has been debated.^{3,7- 10,31- 32} Are lacunes always due to disease of small penetrating vessels or can they be caused by emboli from larger arteries or the heart? Cross-sectional observational studies such as CHS can provide clues as to the causative risk factors, but cannot establish causality. Comparisons of CHS with other studies is difficult because they have dealt mostly with clinically defined lacunar stroke in symptomatic patients either with^33- 36 or without a stroke-free comparison group,^{24,26,37- 41} as opposed to CHS with MRI-defined infarcts in participants who mostly deny a history of TIAs or stroke. These studies agree in general about a lack of association with factors that may indicate a source of emboli in the heart. In CHS, the only such variable that showed a significant association in models adjusted for only age and sex was left atrial enlargement (Table 3). None of these factors, including left atrial enlargement, was independently associated in the detailed stepwise multivariate models.

In CHS, the factors independently associated with a greater risk of having 1 or more MRI-defined lacunes were increased age, diastolic blood pressure, and serum creatinine, and less so, increased pack-years of smoking, being a woman, maximum internal carotid artery stenosis of 50% or greater, and a report of diabetes mellitus at the baseline examination. Associations with age, hypertension, cigarette smoking, and diabetes have been previously described.^33- 36 The association with creatinine is not well recognized but has been reported.⁴² It may simply reflect the deleterious effects of hypertension on a vascular bed other than the brain, with the elevated creatinine level being a marker for target organ damage in the kidneys and the brain. Effects of renal insufficiency on small-vessel permeability have also been proposed.⁴³ The association with female sex is difficult to explain and prior studies⁴ have differed as to whether men or women are at greater risk for lacunes. Some population-based studies have supported an excess of symptomatic lacunes in women,^44- 45 while others have not.⁴⁶

Unresolved by this and other cross-sectional studies^4,47- 48 is whether stenotic lesions of the internal carotid artery play a role in the pathogenesis of lacunes—through artery to artery emboli or hemodynamic effects—or are simply markers for diffuse vascular disease. Internal carotid artery wall thickness was more strongly related to the risk of lacunes in models adjusted for only age and sex than was maximum internal carotid artery stenosis of 50% or greater (Table 3), but the variable for stenosis was the one that entered the stepwise multivariate models. In separate analyses restricted to those with lacunes affecting only one side of the brain, the correlations of stenoses were not consistently stronger for ipsilateral than contralateral lacunes, an observation arguing against a direct causal association. Instead, risk factors for lacunes, such as hypertension and cigarette smoking, may be risk factor for carotid stenosis too.

The Austrian Stroke Prevention Study⁴⁹ deserves special mention because it is population based and included MRI. In this study, 280 elderly volunteers randomly selected from the official register of residents of the city of Graz, Austria, underwent MRI. Rather than just lacunes, as in the current study, these investigators looked for correlates of microangiopathy-related cerebral damage, which included ischemic infarcts of any type and deep white matter lesions. In a stepwise multivariate model, age and hypertension were the most important associations followed by the apolipoprotein E e2/e3 genotype, despite this genotype being associated with a more favorable lipid profile and less cardiac disease. These findings suggest that genetic factors may influence the occurrence of lacunes, white matter changes, or both—a possibility that has not yet been addressed in CHS. Of interest in CHS was the finding of an association with having 1 or more lacunes that was stronger for diastolic than systolic blood pressure, the reverse having been true for white matter findings.¹⁵

Other investigators^5- 6 have suggested that the risk factor profiles of patients with lacunes may differ based on whether the patient had a single vs multiple lacunes and silent vs symptomatic lacunes. In the stepwise multivariate models of subgroups (Table 4), besides age and sex that were forced into all models, the model for single lacunes had no factors in common with the model for multiple lacunes, and similarly the one for silent lacunes had no factors in common with the one for symptomatic lacunes. While these models might be interpreted as supporting the hypothesis of distinct lacunar entities,^5- 6 the intercorrelations among the independent variables and the overlapping confidence interval for these associations (Table 5) do not support such an interpretation. Once a variable is selected by the stepwise procedure, highly correlated variables are unlikely to enter the model even though they may have had bivariate associations with the dependent variable almost as strong as the variable that was selected. When all the variables that had entered any of the stepwise models (Table 4) were forced into new models, the differences across subgroups were minimal, as reflected by similar odds ratios and overlapping 95% confidence intervals (Table 5). For example, the odds ratio for diastolic blood pressure, contrasting the lowest and highest quartile, was in fact greater for symptomatic (2.89) than silent (1.56) lacunes although the 95% confidence intervals for the latter (1.20-2.03)) were largely contained in the former (1.29-6.47). Overall, the results of the analysis for subgroups in CHS do not provide strong support for these subgroups having important differences in risk factor profiles.

Although the subjects with silent lacunes by definition denied a history of TIA and stroke, they were more likely than those whose MRI was free of any infarct to report problems with a broad array of physical and cognitive functions (Table 6). These findings are similar to those in a recent CHS report¹² examining all type of infarcts, despite having excluded in the current study more than 200 participants whose MRI showed large and small cortical infarcts and large subcortical infarcts. In our study, even after controlling for several potential confounding factors, we found that associations remained significant for Digit-Symbol Substitution Test, time to put on and button a shirt, and time to walk 450 cm. These models included white matter grade, which in CHS is correlated both with these measures of function and with the presence of infarcts.¹⁵ The findings of these multivariate models suggest that among participants without a history of TIA or stroke, white matter grade is associated more strongly than lacunes with cognitive impairment and lower extremity dysfunction. Thus, silent lacunes, white matter changes, or their combination may cause dysfunction that is not recognized by patients or their physicians as related to a stroke.

One of the major limitations of this and similar studies is the lack of direct pathologic confirmation that what we are calling lacunes are infarcts and not some other pathologic process that could relate to these dysfunctions. In addition, to decide if participants had a history of TIA or stroke, we relied heavily on their self-report. A study of residents of Rochester, Minn, suggested that self-report may underestimate the number of people whose medical records indicate that they have had a TIA or stroke.⁵⁰ Thus, we may have misclassified some of the participants as free of TIA or stroke and incorrectly concluded that their lacunes were silent.

In summary, in this cohort of largely healthy older adults, MRI-defined lacunar infarcts are common. The most important risk factors for lacunes include increased age, diastolic blood pressure, and creatinine. Differences among subgroups are not drastic enough to suggest a different mechanism of disease for single vs multiple lacunes or silent vs symptomatic lacunes. Despite subjects with silent lacunes denying a history of TIA or stroke, they were at increased risk of dysfunction involving cognition, upper extremities, and lower extremities. The importance of MRI-defined lacunes as indicators of subsequent risk for stroke⁵¹ and the predictors of the development of incident lacunes, whether symptomatic or silent, awaits further study.

Bowman Gray School of Medicine of Wake Forest University, Forsyth County, North Carolina: Gregory L. Burke, John Chen, Alan Elster, Walter H. Ettinger, Curt D. Furberg, Gerardo Heiss, Sharon Jackson, Dalane Kitzman, Margie Lamb, David S. Lefkowitz, Mary F. Lyles, Cathy Nunn, Ward Riley, Beverly Tucker. Electrocardiography Reading Center: Farida Rautaharju, Pentti Rautaharju. University of California, Davis, Sacramento County: William Bommer, Charles Bernick, Andrew Duxbury, Mary Haan, Calvin Hirsch, Lawrence Laslett, Marshall Lee, John Robbins, Richard White. The Johns Hopkins University, Washington County, Maryland: M. Jan Busby-Whitehead, Joyce Chabot, George W. Comstock, Adrian Dobs, Linda P. Fried, Joel G. Hill, Steven J. Kittner, Shiriki Kumanyika, David Levine, Joao A. Lima, Neil R. Powe, Thomas R. Price, Jeff Williamson, Moyses Szkio Melvyn Tockman. Magnetic Resonance Imaging Reading Center: R. Nick Bryan, Norm Beauchamp, Carolyn C. Meltzer, Douglas Fellows, Melanie Hawkins, Patrice Holtz, Naiyer Iman, Michael Kraut, Grace Lee, Cynthia Quinn, Larry Schertz, Earl P. Steinberg, Scott Wells, Linda Wilkins, Nancy C. Yue. University of Pittsburgh, Allegheny County, Pennsylvania: Diane G. Ives, Charles A. Jungreis, Laurie Knepper, Lewis H. Kuller, Elaine Meilahn, Peg Meyer, Reberta Moyer, Anne Newman, Richard Schulz, Vivienne E. Smith, Sidney K. Wolfson. Echocardiography Reading Center (Baseline), University of California, Irvine: Hoda Anton-Culver, Julius M. Gardin, Margaret Knoll, Tom Kurosaki, Nathan Wong. Echocardiography Reading Center (Follow-up), Georgetown MedicalCenter, Washington, DC: John Gottdiener, Eva Hausner, Stephen Kraus, Judy Gay, Sue Livengood, Mary Ann Yohe, Retha Webb. Ultrasound Reading Center, Geisinger Medical Center, Danville, Pa: Daniel H. O'Leary, Joseph F. Polak, Laurie Funk. Central Blood Analysis Laboratory, University of Vermont, Colchester: Edwin Bovill, Elaine Cornell, Mary Cushman, Russell P. Tracy. Respiratory Sciences, University of Arizona, Tucson: Paul Enright. Coordinating Center, University of Washington, Seattle: Alice Arnold, Annette L. Fitzpatrick, Bonnie K. Lind, Richard A. Kronmal, Bruce M. Psaty, David S. Siscovick, Lynn Shemanski, Will Longstreth, Patricia W. Wahl, David Yanez, Paula Diehr, Maryann McBurnie, Chuck Spieker, Scott Emerson, Cathy Tangen, Priscilla Velentgas. National Heart, Lung, and Blood Institute Project Office, Bethesda, Md: Diane E. Bild, Robin Boineau, Teri A. Manolio, Peter J. Savage, Patricia Smith.

Accepted for publication January 30, 1998.

This study was supported by contracts N01-HC-85079 and N01-HC-85086 from the National Heart, Lung, and Blood Institute, Bethesda, Md.

Reprints: W. T. Longstreth, Jr, MD, MPH, Department of Neurology, Box 359775, Harborview Medical Center, 325 Ninth Ave, Seattle, WA 98104-2499 (e-mail: wl@u.washington.edu).