Twenty-Five-Year Prediction of Stroke Deaths in the Seven Countries Study
The Role of Blood Pressure and Its Changes
Background and Purpose This report explores the prediction of long-term stroke mortality in cohorts of the Seven Countries Study.
Methods Sixteen cohorts of men aged 40 to 59 years at entry were examined at years 0, 5, and 10, with mortality follow-up through 25 years.
Results Stroke death rates in 25 years were high in rural Serbia, Croatia, and Japan; intermediate in Italy, Greece, and urban Serbia; and low in Finland, the Netherlands, and the United States. Age and blood pressure were powerful predictors of 25-year stroke mortality in almost all cohorts and countries. Proportional hazards regression coefficients were .0232 increase in stroke death hazard per millimeter of mercury (t=14.60) for systolic blood pressure and .0409 (t=13.41) for diastolic blood pressure. Moderate blood pressure increases from low usual levels were associated with lower stroke mortality rates in years 10 to 25. Increases of blood pressure starting from high usual levels were associated with increased rates of stroke mortality. Systolic blood pressure was associated with stroke mortality at given levels of diastolic pressure, but diastolic blood pressure was not predictive of stroke mortality at given levels of systolic blood pressure.
Conclusions Associations of systolic and diastolic blood pressure with stroke mortality were similar in cultures with different stroke mortality rates. Increases in blood pressure were associated with subsequent excess stroke mortality only in those who started from high usual levels; this study finds lower stroke risk in those men whose blood pressure increased moderately from low usual levels. Diastolic blood pressure is not independently associated with stroke risk in these populations.
The Seven Countries Study was begun in the late 1950s principally to answer major questions regarding the epidemiology and causes of coronary heart disease (CHD). Its findings have been reported in a number of monographs and articles.1 2 3 4 5 6 7 However, data on stroke incidence and mortality have been presented only in preliminary reports.8 9 10 Follow-up for 25 years shows that stroke remains a major cause of death in these countries. In fact, in several cohorts stroke deaths exceed those from CHD.
This report deals with factors that have predictive power for stroke. The hypotheses were (1) that no differences exist among cohorts in the factors predictive of stroke and (2) that increases in blood pressure predict subsequent increases in mortality from stroke. A further aim of this report was to study whether systolic and diastolic blood pressures were related to stroke mortality risk independently of each other. No attempt was made to explain differences in stroke mortality among cohorts.
Subjects and Methods
The Seven Countries Study was started in 1958 when 16 cohorts with a total of 12 763 men aged 40 to 59 years were enrolled: 1 cohort in the United States, 2 in Finland, 1 in the Netherlands, 3 in Italy, 2 in Croatia (formerly Yugoslavia), 3 in Serbia (formerly Yugoslavia), 2 in Greece, and 2 in Japan (named in Table 1⇓). Details on the samples are given elsewhere.1 2 4 The entry survey examinations were performed between 1958 and 1964 with an average participation rate of 90%, ranging from 75% in the US Railroad to 100% in Tanushimaru, Japan.4
Measurements were made according to standard protocols. For this analysis a measurement subset common to all the cohorts was used. Age in years was rounded to the nearest birthday. Body mass index (BMI) was expressed as weight (kilograms) divided by height (meters) squared. Cigarette smoking, elicited by a questionnaire, was expressed as the average number of cigarettes currently smoked per day. Serum total cholesterol, expressed in milligrams per deciliter, was measured in a casual blood sample following the method described by Anderson and Keys.11 Physical activity (primarily occupational), elicited from a questionnaire, was expressed as four activity levels: 1, bedridden; 2, sedentary; 3, moderate; and 4, heavy. Systolic blood pressure, expressed in millimeters of mercury, was measured with a mercury sphygmomanometer with the participant in a supine position at the end of a physical examination by trained physicians, following the method later described in the World Health Organization manual Cardiovascular Survey Methods12 ; two measurements of each of systolic and fourth and fifth phase diastolic blood pressures were recorded 1 minute apart. In this analysis the mean values of the two measurements of systolic and diastolic fifth phase blood pressures were used.
Similar examinations were repeated in surveys of the cohorts 5 and 10 years after entry, with the exception of the Japanese cohorts, in which data for the 5-year examination were unavailable, and the US Railroad, which was not reexamined after 10 years.
Collection of data on vital status and causes of death over the next 25 years was based on review of death certificates and collection of medical information from hospitals, clinical records, interviews of physicians and relatives of the deceased, or any other witnesses of the fatal event. After the 10th year of follow-up in the US Railroad, the 15th year of follow-up in Finland, and irregularly in other areas, only review and coding of official death certificates were performed. All these data enabled central reviewers (H.B. and A.M.) to allocate the final cause of death, following standard criteria and using the World Health Organization’s International Classification of Diseases, Eighth Revision.13 In cases of coexisting causes of death, violence, cancer in advanced stages, and CHD took precedence over stroke in the hierarchy that assigned underlying cause of death. After 25 years the vital status of all but 56 men was known.
The diagnostic criteria for validation of death by stroke were as follows: (1) history and objective evidence of clinical signs of focal disturbance of cerebral function, either acute or chronic, with no apparent cause other than vascular, with paralysis, paresis, or aphasia; or (2) history and evidence of global cerebrovascular insufficiency (a syndrome characterized by small, usually multifocal episodes of cerebral insufficiency, likely due to small hemorrhages) leading directly or indirectly to death.
From these data, which for the most part preceded the advent of CT, it was impossible to differentiate thrombotic from hemorrhagic stroke. Approximately 90% of cases were acute stroke, diagnosed when the event developed in minutes or hours, even if followed by a chronic stage leading to death.
At entry, prevalent cases of stroke were identified following the same criteria described above. Other cardiovascular diseases were identified at baseline and/or follow-up examinations. CHD was defined by a combination of information from a standard questionnaire on angina pectoris and myocardial infarction,12 plus findings of the resting electrocardiogram read by the Minnesota Code; clinical judgment was used when suspected CHD manifested as heart failure (severe dyspnea, gallop rhythm, edema, distended jugular veins) or chronic arrhythmias; details are reported elsewhere.2 4 Other heart diseases were hypertensive heart disease, rheumatic heart disease, valvular heart disease of uncertain etiology, and pulmonary heart disease. Hypertensive heart disease was defined by the presence or documented history of hypertension (systolic blood pressure ≥160 mm Hg and/or diastolic blood pressure ≥95 mm Hg and/or use of specific antihypertensive drugs) plus two or more of the following: history or evidence of congestive heart failure, severe arrhythmias, enlarged heart (apex ≥2 cm outside left midclavicular line), gallop rhythm, a murmur 3/6 or more, after exclusion of definite valvular origin, or use of digitalis. Rheumatic heart disease was diagnosed in the presence of history of rheumatic fever or equivalent information acceptable to the observer, plus any grade 2/6 or more apical or aortic systolic murmur; or any diastolic murmur alone when there was no evidence of luetic, congenital, or hypertensive etiology; or a mitral opening snap. The diagnosis of valvular heart disease of uncertain etiology was made when the rheumatic fever history was absent and when auscultation findings were atypical; practically it applied only to certain cases of aortic valvular disease in relatively elderly people. Pulmonary heart disease (chronic cor pulmonale) was diagnosed in the presence of severe pulmonary disease of long duration in the absence of other clear heart disease, accompanied by definite involvement of the heart with failure and cyanosis. The diagnoses of other heart diseases were made taking into account medical documents supplied by the subject but ignoring electrocardiographic findings collected during the field examination. Peripheral arterial disease was diagnosed by the presence of definite intermittent claudication as elicited by the standard Rose Questionnaire12 or, in the absence of claudication, by the presence of pulse deficit in lower limb(s) or specific trophic lesions. Other rarer heart diseases, such as congenital heart disease or cardiomyopathies, were diagnosed by accepting the clinical judgment of the examining physician.
The baseline data for this project were collected in the era (pre-1960) before written informed consent was required. Men were invited to participate either in writing or through personal contact with local leaders.1 The study and procedures were verbally explained to all participants. In subsequent examinations (circa 1965 and 1970), all study groups obtained additional consent, including written consent to obtain follow-up records in areas in which this was required.
Death rates (per 1000) from stroke were computed with direct age adjustment, with the overall Seven Countries quinquennial age distribution as the reference population.
Initially, predictive analysis was based on the Cox proportional hazards model,14 with stroke mortality in 25 years as the end point and five risk factors as predictors. Some models were solved by combining the cohorts within eight countries (separating Croatia from Serbia) to enlarge the size of samples and number of events. In these cases, dummy variables were added to identify the cohorts and adjust for stroke rate differences, taking as reference the one with the lower or lowest stroke death rate. Multivariate coefficients of systolic blood pressure were compared between cohorts with the use of t tests. The coefficients were pooled by a meta-analytic procedure, weighting by the inverse of the variance.15 The role of diastolic blood pressure was considered in separate multivariate solutions from systolic blood pressure because of high correlation between the two pressures.
The analysis dealing with blood pressure change and later stroke mortality was performed by adopting a linear model that was justified by its flexibility and the simplicity of translating the solutions into projected rates. For all cohorts (except for the US Railroad at year 10 and the two Japanese cohorts at year 5), measurements of blood pressure were available at years 0, 5, and 10. Blood pressure change during the first 10 years of follow-up (5 years for the US Railroad) was analyzed to predict stroke mortality occurring between years 10 and 25. A general linear model was solved with a dummy variable for stroke mortality as the end point. The mean of the available blood pressures at years 0, 5, and 10 was used as the baseline, and changes in 10 years (5 years for the US Railroad) were expressed as the difference of blood pressure at year 10 minus that at year 0. Use of average blood pressure measurements taken at years 0, 5, and 10 reduces dilution bias due to regression toward the mean. Further covariates were age at year 0, the interaction term of blood pressure with its changes, and the dummy variables for cohorts when data were pooled. Separate analyses were made for continuous systolic and diastolic blood pressure variables. Models were used to compute rates adjusted by age, cohort, average blood pressure, changes in blood pressure, and interaction of average with changes in blood pressure. Similar models were made with baseline blood pressure and changes expressed in four and three classes, respectively. Models were run for all cohorts, large geographic pools, and single countries. The geographic pools were composed of North American and northern European cohorts (US Railroad, East Finland, West Finland, and Zutphen, Netherlands) that had low stroke mortality rates; southern European cohorts (Crevalcore, Montegiorgio, and Rome Railroad in Italy; Dalmatia and Slavonia, Croatia; Velika Krsna, Zrenjanin, and Belgrade, Serbia; Corfu and Crete, Greece) that had intermediate or high stroke mortality rates; and Japanese cohorts (Tanushimaru and Ushibuka) that had high stroke mortality rates.
Systolic blood pressure categories were <125, 125 to 139, 140 to 159, and ≥160 mm Hg; diastolic blood pressure categories were <80, 80 to 89, 90 to 94, and ≥95 mm Hg. Categories of 10-year change in systolic blood pressure were 0 or decrease, 1 to 19 mm Hg increase, and ≥20 mm Hg increase. Those for diastolic blood pressure were 0 or decrease, 1 to 9 mm Hg increase, and ≥10 mm Hg increase. The choice of classes for average blood pressure and blood pressure change was based on convention after several options were tested, with both balanced cells and the biological meaning of categories considered.
The occurrence of broadly defined cardiovascular disease at any time during the study was defined by noting the presence of any of the diseases mentioned above occurring at either baseline, year 5, or year 10 examination.
The prevalence of stroke at entry examination was 2.1/1000 (27/12 763 men aged 40 to 59 years).
Table 1⇑ provides the age-adjusted death rates from stroke in 25 years in each of the 16 cohorts. Stroke death rates varied threefold across cohorts. They are lower in North American and northern European areas (36 to 50/1000), intermediate in most of southern Europe (44 to 77/1000), and high in rural Serbia and Croatia and Japan (83 to 107/1000). With the exception of Belgrade, stroke mortality rates in the former Yugoslavia were comparable to those in Japan. Of 28 possible comparisons of stroke death rates in eight countries, 20 were significantly different.
Comparisons were also made between stroke death rates of cohorts of the same country. No significant differences were found within Finland, Greece, and Japan, but rates were statistically significantly higher in Montegiorgio than Crevalcore and the Rome Railroad (Italy); in Slavonia than Dalmatia (Croatia); and in Velika Krsna and Zrenjanin than Belgrade (Serbia).
Stroke mortality represents less than 10% of total mortality in the North American and northern European areas and close to 20% in southern Europe and Japan.
Multivariate Analysis of Risk Factors for Stroke Death
Proportional hazards models were solved with stroke death as the end point and age, systolic blood pressure, physical activity, cigarette smoking, BMI, and serum total cholesterol as covariates for the 16 cohorts (Table 2⇓). Age was a positive predictor in all (significant except in Belgrade), and systolic blood pressure was positively associated in all (significant except in Zutphen, Slavonia, and Belgrade). The coefficients for systolic blood pressure and stroke death were similar between cohorts: of 120 possible comparisons of these coefficients between cohorts, only 12 were statistically significant, and none of these differences occurred between cohorts of the same country. Furthermore, the magnitudes of the age and systolic blood pressure coefficients for a cohort were unrelated to the stroke death rate for the cohort. The pooled estimate of the systolic blood pressure coefficient, weighing inverse to variance, was .0232, with a standard error of .0016 and a t value of 14.60.
The Cox solutions were replicated, including diastolic instead of systolic blood pressure. Multivariate results were similar. The pooled estimate for diastolic blood pressure coefficients was .0409, with a standard error of .0031 and t value of 13.41.
In the multivariate models, number of cigarettes smoked per day was positively associated with stroke mortality in 12 cohorts (statistically significant in the US Railroad, Slavonia, and Velika Krsna) and negative in 4 (statistically significant in Crete). The pooled coefficient was .0148, with a standard error of .0035 and a t value of 4.24. Coefficients predicting stroke mortality from physical activity, BMI, and serum total cholesterol were inconsistently positive or negative across cohorts.
Changes in Systolic and Diastolic Blood Pressure and Prediction of Subsequent Stroke Mortality
We studied the interaction of blood pressure level (average during 10 years) and its change (year 10 minus year 0) in predicting rates of stroke death occurring between 10 and 25 years of follow-up. For ease of visualization, we used a linear model with a dummy dependent variable for cumulative stroke death. Analyses were done separately for systolic and diastolic blood pressure, with adjustment for age. For simplicity, adjustment was omitted for cholesterol, BMI, physical activity score, and cigarette smoking because such adjustment had a negligible effect in analyses reported in Table 2⇑.
For all cohorts combined, when continuous blood pressure terms are used, there is a significant interaction between blood pressure level and blood pressure change (P=.017 for systolic and P=.029 for diastolic blood pressure). We elucidate this interaction in Figs 1⇓ and 2⇓, which show stroke death rates as a function of categories of average level and change in blood pressure, derived from linear models incorporating these categorical variables and their interactions. As in the models with continuous blood pressure terms, probability values for interaction are small in models with categorical terms for blood pressures (Table 3⇓), the entire study, southern Europe (diastolic blood pressure), and for Japan (systolic blood pressure). The predominant feature in both figures is a clear increase in stroke death rate with greater average systolic and diastolic blood pressure level. The effect of change in blood pressure, holding average level fixed, is smaller and varies depending on the average level. Changes in systolic blood pressure in the second intermediate class (140 to 159 mm Hg) and even more so in the highest class (≥160 mm Hg) are accompanied by an increased stroke mortality risk. In the first intermediate systolic class (125 to 139 mm Hg), changes in the first 10 years do not relate to stroke mortality in the next 15 years.
In contrast, in the lowest class of average systolic blood pressure (<125 mm Hg) an increase in blood pressure levels over the first 10 years is associated with a slight decrease in stroke mortality risk. This decrease in risk may be restricted to those whose blood pressure increase was moderate (ie, did not reach hypertensive levels). However, information was sparse on this point: only 21 men had average systolic blood pressure <125 mm Hg, an increase ≥20 mm Hg, and a year 10 systolic blood pressure in the hypertensive range (≥140 mm Hg).
The regression analysis in which categories of blood pressure were used (Figs 1⇑ and 2⇑) revealed another aspect not apparent from the linear interaction model. In those with the highest average blood pressure level (≥160 mm Hg), late stroke mortality was somewhat lower among those with the largest declines in blood pressure during the first 10 years.
An additional analysis assessed the influence of broadly defined cardiovascular disease occurring at any time during the 10 years of clinical observation on stroke mortality risk and blood pressure associations. The stroke mortality rate during years 10 to 25 was approximately one third higher in 1955 men with cardiovascular disease during the first 10 years than in 8308 men in whom cardiovascular disease was absent. The associations between stroke mortality and blood pressure depicted in Figs 1⇑ and 2⇑ were largely replicated in both men who had and did not have prevalent cardiovascular disease.
Trends vary slightly among the three large geographic groups. In particular, a rise in blood pressure in southern Europe and Japan was associated with higher stroke death rates, starting from intermediate baseline levels. It should be noted that the six very high death rates in Japan, in the two upper classes of average diastolic blood pressure, are based on small denominators (23, 9, 20, 28, 12, and 27 subjects, respectively). Other classes of diastolic blood pressure and its change contained at least 50 men. All except four classes of systolic blood pressure and its change included at least 50 men: average <125 mm Hg and change ≥20 mm Hg in northern Europe and North America (n=32) and Japan (n=17), average 140 to 159 mm Hg and change 1 to 19 mm Hg in Japan (n=42), and average ≥160 mm Hg and change 1 to 19 mm Hg in northern Europe and North America (n=36).
The means in the baseline classes of blood pressure were similar in the three geographic groups. Overall, the largest changes in both systolic and diastolic blood pressure in 10 years, by both classes of entry levels and classes of change, were seen in Japan, while southern Europe had intermediate levels of change and northern Europe and North America had the smallest.
Systolic Versus Diastolic Blood Pressure in Prediction of Subsequent Stroke Mortality
Table 4⇓ shows death rates from stroke in follow-up years 10 to 25 as a function of the four baseline classes of systolic crossed with diastolic blood pressure. Rates are adjusted by age and cohorts, with the use of a categorical model similar to that employed for blood pressure change analysis. In each class of diastolic blood pressure, increasing levels of systolic blood pressure were associated with increasing death rates. This was not present for increasing levels of diastolic blood pressure as the level of systolic blood pressure increased. The highest rates were found in cells with systolic blood pressure ≥160 mm Hg and diastolic blood pressure between 90 and 94 mm Hg. In this model as well as that dealing with the same variables treated continuously, the interaction of systolic with diastolic blood pressure in prediction of stroke mortality was not statistically significant.
We computed a similar 3 by 3 matrix for classes of systolic and diastolic blood pressure change, disregarding the baseline levels and again using a general linear model with categories of blood pressure changes. In Table 5⇓, for each class of diastolic blood pressure, increasing levels of change in systolic blood pressure were associated with increased stroke rates. However, greater changes in diastolic blood pressure were associated with higher stroke death rate only in the upper class of systolic change. The cell with the greatest stroke mortality risk was that corresponding to increases of systolic blood pressure ≥20 mm Hg and diastolic blood pressure ≥10 mm Hg. Nevertheless, the interaction term of systolic blood pressure change with diastolic blood pressure change was not statistically significant.
This analysis of the 25-year experience of stroke mortality and its predictors in the Seven Countries Study showed large differences in stroke mortality among cohorts, countries, and geographic groups. As reported by others,16 stroke mortality was higher where CHD mortality was lower.4 6 8
Baseline age and blood pressure were strong predictors of stroke mortality, as observed in other studies.17 18 19 20 21 22 23 In the present study there was great similarity among cohorts and countries representing different cultures in their multivariate coefficients for systolic and diastolic blood pressure in the prediction of stroke, regardless of model. Exceptions such as Belgrade were likely due to small numbers and the possible influence of better access to health facilities, which might have distorted the expected relationship.
Of four other possibly predictive factors studied, only cigarette smoking was predictive of stroke mortality, although it was not predictive in several cohorts. Cigarette smoking has frequently been reported as a risk factor for stroke.17 18 19 20 24 25 26 In this study, the absence of a positive relationship between serum total cholesterol level and stroke mortality can be explained by the impossibility of distinguishing thrombotic from hemorrhagic stroke; for example, a reasonable diagnostic attribution was possible in only two thirds of cases in two of the best-documented cohorts, and diagnoses were even less certain in other cohorts. In studies in which this separation has been possible, serum cholesterol revealed a predictive power for thrombotic but not hemorrhagic stroke.21 22 27 28 Physical activity (largely occupational) and BMI were also unrelated to stroke mortality, although physical inactivity has been observed to be related to stroke risk in other studies.21 29
This study contributes new information regarding the association of changes in blood pressure with stroke mortality. Previous reports from the Dutch and Italian cohorts of the Seven Countries Study found that increases in blood pressure were associated with increased stroke risk.9 10 The results of this study suggest that changes in blood pressure during 10 years were associated with a different risk of stroke mortality, depending on the usual blood pressure levels and the area but independent of whether prevalent cardiovascular disease was identified during 10 years of clinical observation. In general, a moderate increase of blood pressure, either systolic or diastolic, was associated with reduced risk when it occurred at low levels of usual blood pressure (<125 mm Hg for systolic and <80 mm Hg for diastolic). We cannot exclude the possibility that stroke mortality rate was increased when blood pressure increased from a low level into the hypertensive range, but this phenomenon rarely occurred in the Seven Countries Study. In the intermediate levels of usual blood pressure, little or no effect could be attributed to blood pressure change. In contrast, changes of blood pressure starting from the higher usual levels (≥140 mm Hg for systolic or ≥90 mm Hg for diastolic) were associated with excess risk when blood pressure increased, although the greatest risk was not associated with the greatest increase, except for systolic blood pressure in southern Europe. This observation could be explained in Japan by the small numbers involved in subclasses of blood pressure change and in northern Europe and North America by the possible protection of antihypertensive treatment, which may have been started after the higher blood pressure levels were detected in the last survey, at 10-year follow-up. This is plausible because the observational period of change ends in the mid 1960s for the US cohort and the late 1960s or early 1970s for the northern and southern European cohorts. The United States, Finland, and the Netherlands are among the countries that experienced a large decline in stroke mortality in the 1970s and 1980s, attributed in part to improvement of hypertension detection and treatment.30 31 32 33 This started later in other countries such as Italy,34 35 while for others, such as the former Yugoslavia, there has been a continuous increase in stroke mortality rates in those periods.35 Information on the use of antihypertensive drugs is scanty and not uniform in the Seven Countries Study database. However, this influence could have affected only a small proportion of men in the upper part of the blood pressure distribution, and that only toward the end of the follow-up period. In any case, we believe that these data show that reduced stroke risk is associated with a decrease or no change of blood pressure in men who usually have high blood pressure levels.
Additionally, this study makes a contribution concerning the relative importance of systolic versus diastolic blood pressure. For both usual blood pressure levels during 10 years and changes in these levels, systolic blood pressure was predictive of stroke within categories of diastolic blood pressure, but diastolic blood pressure was not predictive of stroke given the value of systolic blood pressure.
There are limitations in the diagnosis of stroke in this study. They are largely bound to the fact that early deaths occurred in 1958, when modern imaging of the brain was not available. However, the same standard was maintained for the 25 years of follow-up, which was concluded between 1983 and 1989, depending on the cohort. No validation based on modern techniques was possible, and the resulting misclassification may have attenuated the relationships reported in this article. An additional limitation of the study is that it was conducted only with men, at a time when female cohorts were seldom considered for population surveys of cardiovascular diseases. Generalizability of the findings to women should be done with caution.
The authors acknowledge Professor Ancel Keys (Minneapolis, Minn) for the promotion, organization, and construction of the Seven Countries Study for 25 years and also the contributions provided in the collection of data by Professor Martti Karvonen (Helsinki, Finland), Dr Sven Punsar (Espoo, Finland), and Drs Miodrag C. Ostojic and Miodrag Z. Grusic (Institute of Cardiovascular Diseases, University Clinical Center of Belgrade, Yugoslavia).
- Received September 15, 1995.
- Revision received November 27, 1995.
- Accepted November 27, 1995.
- Copyright © 1996 by American Heart Association
Keys A, Blackburn HW, Van Buchem FSP, Buzina R, Djordjevic BS, Dontas AS, Fidanza F, Karvonen MJ, Kimura N, Lekos D, Monti M, Puddu V, Taylor HL. Epidemiological studies related to coronary heart disease: characteristics of men aged 40-59 in seven countries. Acta Med Scand. 1967;460(suppl 180):1-392.
Keys A, ed. Coronary heart disease in seven countries. Circulation. 1970;41(suppl 1):1-211.
Keys A, Aravanis C, Blackburn H, Van Buchem FSP, Buzina R, Djordjevic BS, Fidanza F, Karvonen M, Menotti A, Puddu V, Taylor HL. Probability of middle-aged men developing coronary heart disease in five years. Circulation. 1972;45:815-828.
Keys A, ed. Seven Countries: A Multivariate Analysis of Death and Coronary Heart Disease. Cambridge, Mass: Harvard University Press; 1980:1-381.
Keys A, Menotti A, Aravanis C, Blackburn H, Djordjevic BS, Buzina R, Dontas AS, Fidanza F, Karvonen MJ, Kimura N, Mohacek I, Nedeljkovic S, Puddu V, Punsar S, Taylor HL, Conti S, Kromhout D, Toshima H. The Seven Countries Study: 2,289 deaths in 15 years. Prev Med. 1984;13:141-154.
Menotti A, Keys A, Kromhout D, Blackburn H, Aravanis C, Bloemberg B, Buzina R, Dontas A, Fidanza F, Giampaoli S, Karvonen MJ, Lanti M, Mohacek I, Nedeljkovic S, Nissinen A, Pekkanen J, Punsar S, Seccareccia F, Toshima H. Inter-cohort differences in coronary heart disease mortality in the 25-year follow-up of the Seven Countries Study. Eur J Epidemiol. 1993;9:527-526.
Menotti A, Keys A, Blackburn H, Aravanis C, Dontas AS, Fidanza F, Giampaoli S, Karvonen MJ, Kromhout D, Nedeljkovic S, Nissinen A, Pekkanen J, Punsar S, Seccareccia F, Toshima H. Twenty-year stroke mortality and prediction in twelve cohorts of the Seven Countries Study. Int J Epidemiol. 1990;19:309-315.
Keli S, Bloemberg B, Kromhout D. Predictive value of repeated systolic blood pressure measurements for stroke risk: the Zutphen Study. Stroke. 1992;23:347-351.
Menotti A, Lanti P, Seccareccia F, Gianpaoli S, Dima F. Multivariate prediction of the first cerebrovascular event in an Italian sample of middle-aged men followed up for 25 years. Stroke. 1993;24:42-48.
Anderson JT, Keys A. Cholesterol in serum and lipoprotein fractions: its measurement and stability. Clin Chem. 1956;2:145-159.
Rose G, Blackburn H. Cardiovascular Survey Methods. Geneva, Switzerland: World Health Organization; 1968:527-536.
WHO. International Classification of Diseases, Eighth Revision. Geneva, Switzerland: World Health Organization; 1965.
Cox DR. Regression models and life tables. J R Stat Soc. 1972;B34:187-220.
Pocock SJ, Thompson SG. Assessing the evidence: the role of meta-analysis. In: Marmot M, Elliott P, eds. Coronary Heart Disease Epidemiology: From Aetiology to Public Health. Oxford, England: Oxford University Press; 1992:383-396.
Reed DM. The paradox of high risk of stroke in populations with low risk of coronary heart disease. Am J Epidemiol. 1990;131:579-588.
Wilhelmsen L. Synergistic effects of risk factors. Clin Exp Hypertens Part A Theory and Practice.. 1990;12:845-863.
Harmsen P, Rosengren A, Tsopogianni A, Wilhelmsen L. Risk factors for stroke in middle-aged men in Göteborg, Sweden. Stroke. 1990;21:223-229.
Truelsen T, Lindenstrom E, Boysen G. Comparison of probability of stroke between the Copenhagen City Heart Study and the Framingham Study. Stroke. 1994;25:802-807.
Jamrozik K, Broadhurts RJ, Anderson CS, Stewart-Wynne EG. The role of lifestyle factors in the etiology of stroke: a population-based case-control study in Perth, Western Australia. Stroke. 1994;25:51-59.
Haheim LL, Holme I, Hjermann I, Leren P. Risk factors of stroke incidence and mortality: a 12-year follow-up of the Oslo Study. Stroke.. 1993;24:1484-1489.
Hu HH, Sheng WY, Chu FL, Lan CF, Chiang BN. Incidence of stroke in Taiwan. Stroke. 1992;23:1237-1241.
MacMahon S, Peto R, Cutler J, Collins R, Sorlie P, Neaton J, Abbott R, Godwin J, Dyer A, Stamler J. Blood pressure, stroke, and coronary heart disease, part I: prolonged differences in blood pressure: prospective observational studies corrected for the regression dilution bias. Lancet. 1990;335:765-774.
Benfante R, Yano K, Hwang LJ, Curb JD, Kagan A, Ross W. Elevated serum cholesterol is a risk factor for both coronary heart disease and thromboembolic stroke in Hawaiian Japanese men: implications of shared risk. Stroke. 1994;25:814-820.
Abbott RD, Rodriguez BL, Burchfiel CM, Curb JD. Physical activity in older middle-aged men and reduced risk of stroke: the Honolulu Heart Program. Am J Epidemiol. 1994;139:881-893.
Tuomilehto J, Bonita R, Stewart A, Nissinen A, Salonen JT. Hypertension, cigarette smoking, and the decline in stroke incidence in eastern Finland. Stroke. 1991;22:7-11.
Menotti A, Scanga M, and the Responsible Investigators of the RF2, OB43 and MICOL Research Groups. Trends in coronary risk factors in Italy. Int J Epidemiol. 1992;21:883-892.
Thom TJ, Epstein FH, Feldman JJ, Leaverton PE, Wolz M. Total Mortality and Mortality From Heart Disease, Cancer and Stroke From 1950 to 1987 in 27 Countries. Bethesda, Md: National Institutes of Health; 1992. Publication NIH 92-3088.