Prospective Evaluation of Dietary and Other Predictors of Fatal Stroke in Shanghai, China
Background Although a number of risk factors for fatal stroke are well established in Western populations, this is less true for Asian countries, many of which have stroke mortality rates that are historically high. In a prospective study in Shanghai, China, we determined whether the same factors predict risk for fatal stroke as in the West. We also studied a number of potential dietary associations, particularly those with dietary antioxidants because these have been suggested to reduce atherogenesis.
Methods and Results Between 1986 and 1989, 18 244 men aged 45 to 64 years living in four geographically defined areas of Shanghai, China were recruited to participate in a prospective study of diet and cancer. All participants completed an in-person, structured interview and provided blood and urine samples. As of March 1994, fatal stroke accounted for 245 of the 980 observed deaths. The most important risk factor for stroke mortality was a history of hypertension (multivariate relative risk, 4.5; 95% confidence interval, 3.3, 6.2). Cigarette smoking was not strongly associated with risk, and alcohol consumption increased risk only in the extreme categories of lifetime consumption. Educational level was strongly, inversely associated with fatal stroke, and this could not be explained by adjustment for any other risk factors. No macronutrient was associated with risk, including total energy, fat consumption, or any component of fat. There also were no significant inverse associations for stroke mortality with several micronutrients of interest, including vitamin C, carotene, vitamin E, riboflavin, or calcium.
Conclusions Our data demonstrate that hypertension is by far the most important contributor to stroke mortality in Shanghai and that among dietary factors, only alcohol consumption shows any evidence whatsoever of being a risk factor.
The epidemiology of fatal stroke has been well studied in Western populations, and a number of risk factors have been identified and characterized. Cigarette smoking and a history of hypertension or diabetes are prominent among these.1 2
In some Asian countries, including China and, historically at least, Japan, the incidence and mortality rates from stroke have been higher than those in the United States.3 Recent studies have provided useful information on demographic risk factors for stroke (age, sex, and secular trends) in a number of these Asian populations,4 5 but there are few data available to identify high-risk groups on the basis of nondemographic characteristics.
Recently, it has been suggested that the antioxidant properties of certain micronutrients, in particular vitamins C and E and certain carotenoids, may reduce risk of ischemic heart disease6 7 by preventing oxidation of LDL,8 thereby interfering in atherogenesis. Although there exists some minimal evidence that foods rich in these micronutrients might reduce risk of stroke as well,9 10 11 there are no detailed epidemiological evaluations of these possible relationships. Other dietary factors, including alcohol12 and sodium,13 have been suggested to increase risk of stroke, whereas diets high in potassium,14 selenium,15 fat,16 and protein,17 have been suggested to reduce risk, but the epidemiological evidence to support any of these relationships is, to date, uncompelling.18
Since 1989, we have been monitoring a well-characterized cohort of 18 244 middle-aged men in Shanghai, People’s Republic of China, for various health outcomes. Until March 1, 1994, cerebrovascular accidents were by far the leading cause of death in this cohort. The annual age-adjusted mortality rate from stroke in the cohort to date is 220 per 100 000. Using similar age adjustment, rates in US men are <25% of this rate.19 We therefore took the opportunity to evaluate risk factors, both dietary and nondietary, for stroke mortality in this population.
Between January 1, 1986, and September 30, 1989, we invited all eligible male residents of four small, geographically defined communities scattered over a wide area of the city of Shanghai to participate in a prospective, epidemiological study of diet and cancer. These four regions were chosen to cover a broad geographic cross section of the city but were comparable in terms of social class, including education and family income. The eligibility criteria were ages 45 to 64 years and no history of cancer. In addition to collecting a 10-mL nonfasting blood sample and a single void urine sample, we interviewed each subject in person using a structured questionnaire that included level of education, usual occupation, current height and weight, history of tobacco and alcohol use, current diet (intake frequencies of 45 food items, see “Appendix” for the list of these food items), and medical history (ever diagnosed by a medical doctor for the following illnesses: hepatitis, hepatic cirrhosis, other liver disease, gallstones, other gallbladder disease, gastric ulcer, duodenal ulcer, diverticulosis, polyposis coli, ulcerative colitis, adenomatous polyps, diabetes, hypertension, tuberculosis, emphysema, asthma, schistosomiasis, other parasitic disease, and cancer). Interviews were administered by retired nurses who had been specially trained by one of us (J-M.Y.) with experience in epidemiological dietary studies. During the 3-year recruitment period, 18 244 men (≈80% of eligible subjects) enrolled in the study.
To determine the level of consumption of specific foods, food categories, and certain macronutrients and micronutrients, the following strategies were used: In the structured questionnaire, the respondent was asked to indicate the frequency (in number of times per day, week, month, or year) with which he usually consumed each of 45 food items or food groups as an adult. All common foods consumed in the local Shanghainese diet were covered by these categories. For seasonal foods, we obtained the frequency of consumption when the food was in season. The amounts of various cooking oils used each month by the household were also indicated.
To establish a standard portion size for each of the food items investigated, 100 subjects randomly selected from all cohort members were asked to indicate, using standard measuring instruments and food models, their normal serving size of each of the foods under study. For each food, the mean weight of all recorded servings was taken as the weight of the standard portion.
The macronutrient and micronutrient content of each food was determined from the Chinese Food Composition Tables.20 Although crude fiber represents only a portion of total dietary fiber, the Chinese Food Composition Tables do not provide values on dietary fiber, and the only available surrogate for fiber intake is crude fiber, which we used for all analyses related to fiber. For each subject, we computed the average yearly intake of each macronutrient or micronutrient by summing across all foods the cross product of intake frequency (in number of times per year), standard portion weight (in grams), and average nutrient level in the particular food. Mean daily intake levels (in grams) were calculated by dividing yearly totals by 365. Tertiles of exposure categories were constructed on the basis of the distribution of intake among all cohort members.
In terms of alcohol consumption, we asked each participant whether he had ever drunk alcoholic beverages at least once a week for ≥6 months. If the answer was yes, he was asked to provide the age at which he started to drink regularly and the usual amount of consumption of beer, wine, and hard liquor separately. For each subject, we computed the lifetime ethanol intake by summing the cross product of yearly ethanol consumption (grams per week times 52) from beer (3.5% ethanol by weight), wine (12.1% ethanol by weight), and hard liquor (43% ethanol by weight) and the number of years he drank regularly.
Although follow-up is achieved in a number of ways,21 22 vital statistics units within the Shanghai municipality are the principal sources for identifying deaths. The Shanghai municipality consists of 12 urban districts, each with its own such Vital Statistics Unit. Copies of death certificates, which include cause of death, for residents of the districts targeted for the cohort study are routinely ascertained and matched against the cohort master file. Because the vital status is currently known for >99% of the participants, follow-up in terms of mortality is essentially complete. Because of the substantial additional costs involved in ascertaining and confirming nonfatal health outcomes, these were not routinely tracked for any disease except cancer.
As of March 1, 1994, the 18 244 original participants in the cohort had contributed 105 375 person-years of follow-up and 980 deaths had been identified, including 245 for which cerebrovascular accident or stroke was listed as the principal cause of death. In other words, 1 of every 4 deaths was due to stroke. This was easily the number one cause of death in the cohort (the second and third leading causes of death were lung cancer and cardiovascular disease, with 130 and 123 deaths, respectively). Of the 245 cases of fatal stroke, 83% died in a hospital or had been recently hospitalized for stroke but died at home. Another 9% of patients had their death certificates completed by physicians affiliated with a city or district-level hospital.
For each of these 245 patients who had died of stroke, we selected from the remaining cohort 5 additional participants who were matched to the stroke patients within 1 year of age, within 1 month of interview, and for neighborhood of residence. These control subjects were randomly selected from among all cohort members who were enrolled in the study and were alive at the time of death of the index case subject.
Finally, we assessed urinary sodium, potassium, and creatinine (to adjust for volume of the spot sample) in case subjects and in one of the five (randomly selected) matched control subjects. Case-control comparisons were made by comparing sodium and potassium to creatinine ratios.
Estimates of relative risks (RRs) and their corresponding CIs were computed by use of the conditional logistic regression method23 to examine the relationship between various exposure indices and stroke death with or without adjustment for possible confounders. The nonparametric Mann-Whitney U test was used to compare nutrient and food consumption levels between stroke patients and control subjects.24 All probability values reported are two-sided.
Men dying of stroke were less likely than control subjects to have received a university education and also had a somewhat lower average income level than control subjects, although the latter association was not statistically significant. A large majority of both case and control subjects were currently married, but case subjects were substantially more likely to be widowers or never married than control subjects (Table 1⇓).
Risk of fatal stroke increased with increasing body mass index; those in the highest quartile had 1.5 times the risk of those in the lowest quartile (P=.03, test for trend) (Table 1⇑).
There was no strong relationship between any category of cigarette smoking and fatal stroke (Table 1⇑). Current smokers had a 10% but not statistically significant elevation in risk of stroke death compared with lifetime nonsmokers, and ex-smokers showed a small, statistically nonsignificant reduction in risk. There was a modest increase in risk with increasing lifetime cigarettes; after ≥30 years of smoking, risk was elevated 20% to 40% compared with lifetime nonsmokers. However, <30 years of smoking conveyed no elevation in risk. Moderate alcohol consumption was not associated with any increase in risk of stroke death (Table 1⇑). However, heavy drinking over a sustained period of time (≥1000 kg of ethanol over lifetime) was associated with a 2.4-fold increase in risk of stroke death relative to nondrinking and moderate alcohol drinking (95% CI=1.4, 3.9).
The strongest association between any of the variables evaluated and stroke mortality was for a history of hypertension. Approximately two thirds of case subjects reported such a history. The RR of fatal stroke among men with a history of hypertension was 4.1 (95% CI=3.1, 5.5). All but 4 of the case subjects and 10 of the control subjects reporting a history of high blood pressure had received treatment for it. Only 4 case subjects and 13 control subjects reported a history of diabetes (RR=1.5, 95% CI=0.5, 4.7).
Table 1⇑ also presents RRs for potential nondietary risk factors for stroke mortality, taking into account both the subject’s hypertensive status, because hypertension was such a dominant risk factor, and the other risk factors or potential risk factors included in the Table⇑. The association with body mass index disappeared in this multivariate analysis. The weak relationship between cigarette smoking and stroke mortality dissipated further after adjustment for a history of hypertension and these other variables, whereas the relationship with heavy alcohol consumption remained (adjusted RR=2.2 for >1000 kg lifetime consumption versus none; 95% CI=1.3, 3.9). Another association modestly affected by these adjustments was level of education; there was now an even more distinct gradient of decreasing risk with increasing level of education. As for the association between hypertension and stroke mortality, this was slightly enhanced by adjustment for the other variables presented in Table 1⇑.
Tables 2⇓ and 3⇓ show the relationships between selected nutrients and stroke mortality as median levels of daily intake in case and control subjects without adjustment for potential confounders and as tertiles of intake (based on distributions of the entire cohort) with adjustment for hypertension and other potential confounders. Neither fat nor any component of fat was associated with fatal stroke, nor were protein or carbohydrate consumption nor caloric consumption overall. Similarly, we found no differences between case and control subjects in intake of the micronutrients under investigation, including carotene, vitamin C, vitamin E, riboflavin, and calcium, either when evaluated as median levels of daily intake between case and control subjects without adjustment for potential confounders or as tertiles of intake with adjustment for hypertension and other confounders.
We also examined the relationship between stroke mortality risk and individual food items or food groups. No significant difference between case and control subjects was found in intake of any food items or food groups (see “Appendix” for the food list), when evaluated either as median levels of daily intake or as tertiles of intake frequencies (data not shown).
The laboratory analyses of sodium- and potassium-to-creatinine ratios were restricted to nonhypertensives because therapy for hypertension can profoundly affect urinary sodium and potassium excretion. No statistically significant differences in these ratios were apparent between case and control subjects, either in mean values or when divided into tertiles based on values in control subjects (data not shown).
This study has a number of important strengths as well as some limitations. Among the strengths are the prospective study design, eliminating the possibility of recall bias; the excellent follow-up achieved, minimizing the possibility of biased follow-up associated with any of the exposures under evaluation; the relatively large number of fatal stroke outcomes, providing substantial statistical power for evaluating exposure–fatal stroke associations; and the availability of detailed food composition tables, allowing calculations of individual macronutrient and micronutrient consumption.
The principal limitation of the study is the need to rely on death certificates as the sole diagnostic source of stroke mortality. The very strong relationship between hypertension (the most important risk factor for stroke in Western populations2 ) and stroke mortality in Shanghai, as well as the high proportion of men who died with stroke as the listed cause of death who also had a history of hypertension, provides some reassurance that the cause of death has been correctly attributed to stroke in the majority of instances. More importantly, as noted, we have routinely reviewed and documented circumstances of death of cohort members by reviewing hospital records or other supporting medical documents. Ninety-two percent of patients had been hospitalized for stroke or had their cause of death listed by a physician in a large, high-quality hospital. These diagnoses are highly likely to be accurate. Nonetheless, by relying on death certificate ascertainment, we were unable to distinguish fatal hemorrhagic events from thrombotic ones. In other major urban areas of China, hemorrhagic stroke has been reported to account for 38% of stroke deaths versus 58% for thrombotic stroke.5 Shanghai would likely have a similar distribution. To the extent that these two main types of fatal events have different causes and risk factors, our results might obscure some modest associations.
A second limitation of this study is inherent to all observational epidemiological studies of diet, ie, measurement error in the dietary assessment. Although misclassification will not create false associations, it will reduce possible real associations toward the null. However, even when dietary variables were evaluated categorically, no associations were found.
The baseline questionnaire did not elicit a history of stroke or transient ischemic attack. The underlying prevalence of these conditions was likely very low given the relatively young ages of the enrollees. We noted a deficit in the number of total deaths during the first year of follow-up compared with subsequent years,19 suggesting that very ill patients were less likely to enroll. Thus, the existence of nonfatal strokes among control subjects is unlikely to materially affect our results.
Our data demonstrate that hypertension is by far the most important risk factor for fatal stroke in Shanghai, as in most Western populations, but they do not support a strong role for smoking, another important risk factor in the West, in predicting risk in this population. In Shanghai, smoking is strongly associated with risk of other diseases known to be related to smoking in Western populations.19 Although all our risk factor data are self-reported, this would not result in false associations in a prospective study with internal comparisons. Hypertension in Shanghai, as in all populations, may go undiagnosed (although nearly all cohort members had healthcare coverage through their workplace). A strong relationship with body mass index was greatly attenuated when adjusted for the strong association between stroke mortality and hypertension. This observation is not surprising given the well-known association between obesity and hypertension, although it is of particular interest given the relatively low mean body mass index of men in this cohort (22.2 kg/m2, ≈4 kg/m2 less than US whites25 ). After adjustment for hypertension, the relationship between heavy consumption of alcohol and stroke mortality remained, the final result again being compatible with the association between alcohol intake and risk observed in Western populations. Alcohol intake is often underreported in epidemiological studies, but there is no apparent social stigma associated with alcohol use in China, and moreover, there is no reason to suspect differential underreporting of alcohol intake among subsequent stroke victims.
Self-reported diabetes was so uncommon in this cohort that we are unable to determine with certainty whether such a history represents an independent risk factor, although the limited data available suggest it was not. Quite unexpectedly, we noted a strong inverse relationship between risk of stroke death and level of education that could not be explained by any of the lifestyle or dietary factors investigated in the study.
This cohort provides little evidence that any macronutrient (in particular, fat or any component of fat and protein intake) or micronutrient (in particular, vitamins C, E, or A; carotene; riboflavin; or calcium) is associated with fatal stroke.
There are few epidemiological data available on the relationship between antioxidants and stroke, whether fatal or nonfatal, in Western populations. Gey et al,26 using data from the Basel Prospective Study, found some evidence of an inverse relationship between stroke mortality and low plasma carotene and low plasma vitamin C (especially in conjunction with low plasma carotene) but no relationship with circulating vitamin E or retinol. Gillman et al11 recently reported that on the basis of a 24-hour recall, the combined intake of fruits and vegetables was associated with a reduced risk of stroke in 832 men followed up prospectively for 20 years in Framingham, Mass. However, no calculations of specific nutrient intake were conducted. Epidemiological data are also scanty regarding a possible reduced risk of ischemic heart disease with high dietary intake of micronutrient antioxidants.6 7 Our data provide no support for a protective effect of these micronutrients on stroke mortality.
There has been much interest in the past decade in a possible inverse relationship between dietary calcium and blood pressure.27 Observational epidemiological data and clinical trials of supplemental calcium have provided inconsistent results on this relationship, but these studies as a group have certainly not excluded an effect of calcium in reducing blood pressure.27 A recent meta-analysis28 of randomized trials of the impact of calcium supplementation on blood pressure concluded that calcium supplements may lead to a small reduction in systolic but not diastolic blood pressure. In the current study, we find no evidence that dietary calcium reduces risk of fatal stroke despite the strong association between hypertension and stroke mortality.
In summary, hypertension was found to be the most important risk factor for stroke mortality in middle-aged men in Shanghai, China. Smoking was not a strong predictor of risk for fatal stroke in this high-risk population. Among dietary factors, heavy alcohol consumption was positively associated with risk of fatal stroke. Our data provided no support for a protective effect of dietary antioxidants on stroke mortality in middle-aged Shanghai men.
The 45 food items or food groups listed in the questionnaire are: (1) corn; (2) wheat or other grains (bread, steamed bun, noodles, etc); (3) dried beans or peas (broad beans, mung beans, red beans, peas, etc); (4) peanuts; (5) peanut butter; (6) rice; (7) fresh bean curd (tofu, partially dried tofu, tofu sheet, sugee, etc); (8) fresh fish of any kind; (9) fresh shrimp, clam, crab, or squid; (10) fresh fatty pork; (11) fresh lean pork; (12) fresh chicken or duck; (13) cow or goat milk; (14) fresh eggs of any kind; (15) liver of any kind; (16) fresh vegetables of any kind; (17) dark green leafy vegetables (bok choi, spinach, Chinese celery, mustard green, etc); (18) fresh green beans or peas (broad beans, peas, etc); (19) green squash or gourd; (20) light green/white vegetables (cabbage, cauliflower, wild-rice stem, eggplant, etc); (21) carrots; (22) tomatoes in season; (23) pumpkin or other orange winter squash in season; (24) sweet or hot red peppers in season; (25) sweet or hot green peppers in season; (26) white potato; (27) pale sweet potato in season; (28) dark sweet potato in season; (29) fresh fruit of any kind; (30) orange or tangerine,; (31) watermelon in season; (32) other melons (honeydew, sweet squash, etc) in season; (33) loquat in season; (34) persimmon in season; (35) apricots in season; (36) other fruits (apple, pear, banana, etc); (37) fermented bean curd (fuyu); (38) fermented bean paste; (39) salted vegetables or roots; (40) salted fish; (41) salted pork, shrimp, or crab; (42) sausage, ham, or other cured meat; (43) pickled vegetables; (44) salted eggs of any kind; and (45) lard.
This study was supported by grants R01 CA43092 and R35 CA53890 from the National Cancer Institute, Bethesda, Md. We thank Kazuko Arakawa for her assistance in data management and analysis.
Reprint requests to Dr Ronald Ross, University of Southern California/Norris Comprehensive Cancer Center, 1441 Eastlake Ave, Suite 803, Los Angeles, CA 90033.
- Received August 26, 1996.
- Revision received January 8, 1997.
- Accepted January 21, 1997.
- Copyright © 1997 by American Heart Association
Dyken ML, Wolf PA, Barnett HJM, Bergan JJ, Hass WK, Kannel WB, Kuller L, Kurtzke JF, Sundt TM. Risk factors in stroke: a statement for physicians by the Subcommittee on Risk Factors and Stroke of the Stroke Council. Stroke. 1984;15:1105-1111.
Shaper AG, Phillips AN, Pocock SJ, Walker M, Macfarian PW. Risk factors for stroke in middle-aged British men. Br Med J. 1991;302:1111-1115.
Gey KF. On the antioxidant hypothesis with regard to arteriosclerosis. Bibl Nutr Dieta. 1986;37:53-91.
Witztum JL, Steinberg D. Role of oxidized low density lipoprotein in atherogenesis. J Clin Invest. 1991;88:1785-1792.
Acheson RM, Williams DRR. Does consumption of fruit and vegetables protect against stroke? Lancet. 1983;8335:1191-1193.
Virtamo J, Valkeila E, Alfthan G, Punsar S, Huttunen JK, Karvonen MJ. Serum selenium and the risk of coronary heart disease and stroke. Am J Epidemiol. 1985;122:276-282.
McGee D, Reed D, Stemmerman G, Rhoads G, Yano K, Feinleib M. The relationship of dietary fat and cholesterol on mortality in 10 years: the Honolulu Heart Program. Int J Epidemiol. 1985;14:97-105.
Institute of Nutrition and Food Hygiene, Chinese Academy of Preventive Medicine. Chinese Food Composition Tables. 2nd ed. Beijing, PRC: New China Press; 1991.
Qian G-S, Yu MC, Ross RK, Yuan J-M, Gao Y-T, Wogan GN, Groopman JD. A follow-up study of urinary markers of aflatoxin exposure and liver cancer risk in Shanghai, People’s Republic of China. Cancer Epidemiol Biomarkers Prev. 1994;3:3-10.
Breslow NE, Day NE. Statistical Methods in Cancer Research: Vol 1, The Analysis of Case-Control Studies. Lyon, France: International Agency for Research on Cancer; 1980.
Siegel S. Nonparametric Statistics for the Behavioral Sciences. New York, NY: McGraw-Hill Book Co; 1956:116-127.
Gey KF, Stähelin HB, Eichholzer M. Poor plasma status of carotene and vitamin C is associated with higher mortality from ischemic heart disease and stroke: Basel Prospective Study. Clin Investig. 1993;71:3-6.