点击显示 收起
1 From the Department of Epidemiology (YC, PF-L, GRH, and HA) and the Department of Environmental Health Sciences (PF-L and FP), Mailman School of Public Health, Columbia University, New York, NY; the Department of Environmental Medicine and the New York University Cancer Institute, New York University School of Medicine, New York, NY (YC); and the Herbert Irving Comprehensive Cancer Center, Columbia University, New York, NY (HA)
2 Supported by US National Institute of Environmental Health Sciences grants P42ES10349, P30ES09089, and ES000260 and National Institutes of Health grants R01CA107431 and R01CA102484. 3 Address reprint requests to Y Chen, Department of Environmental Medicine, New York University School of Medicine, 650 First Avenue, Room 510, New York, NY 10016. E-mail: y.chen{at}med.nyu.edu.
ABSTRACT
Background: The nutritional determinants of hypertension in Bangladesh and other low-income countries are largely unknown.
Objective: We assessed the associations of general hypertension with nutrient intakes and diet patterns in Bangladesh.
Design: This was a cross-sectional analysis of 11 116 participants enrolled in the Health Effects of Arsenic Longitudinal Study in Bangladesh. Dietary intakes were measured by use of a validated food-frequency questionnaire.
Results: Three major dietary patterns were identified by using principal component analysis: 1) the "balanced" pattern, which was characterized by rice, some meat, small fish, fruit, and vegetables; 2) the "animal protein" pattern, which was more heavily weighted on meat, milk, poultry, eggs, bread, large fish, and fruit; and 3) the "gourd and root vegetable" pattern, which consisted largely of squashes and root and leafy vegetables. Adjusted prevalence odds ratios for general hypertension in increasing quintiles of balanced pattern scores were 1.00 (reference), 0.81 (95% CI: 0.79, 0.97), 0.82 (0.68, 0.97), 0.79 (0.66, 0.94), and 0.71 (0.59, 0.85) (P for trend < 0.01). Prevalence odds ratios for general hypertension in increasing quintiles of animal protein pattern scores were 1.00 (reference), 1.30 (1.01, 1.52), 1.20 (1.01, 1.47), 1.22 (1.00, 1.44), and 1.21 (1.03, 1.49) (P for trend = 0.23). Markers of high socioeconomic status were positively associated with the animal protein pattern.
Conclusion: Our findings suggest the importance of dietary patterns in general hypertension in a low-income population undergoing the early stage of the epidemiologic transition.
Key Words: Nutritional epidemiology • hypertension • high blood pressure • cross-sectional study • diet patterns
INTRODUCTION
All societies are confronted with the problem of defining a strategy to control high blood pressure. Large, prospective epidemiologic studies unequivocally show a strong, direct relation between high blood pressure and mortality due to cardiovascular disease (CVD) (1). Although the relative contribution of CVD deaths to total mortality in developing countries is smaller than that in developed countries, developing countries, because of their large populations, contribute nearly twice as much as do developed countries to the global CVD burden (2). Because hypertension is the most common cardiovascular condition in the world, its prevention and treatment are important public health issues.
The shift of major causes of death from predominantly nutritional deficiencies and infectious diseases to those classified as degenerative (chronic diseases such as CVD) as the result of industrialization has been termed the epidemiologic transition (3). Although Bangladesh was classified as being in the earliest stage of this transition (4), a recent review of prevalence surveys conducted in Bangladesh indicated that the prevalence of hypertension has increased from <3% to 9% since 1976 (5). Parallel to this increase, the prevalence of chronic energy deficiency [body mass index (BMI; in kg/m2) < 18.5 on the basis of international criteria] in adults decreased by 14% from 1981 to 1996 (6).
Nutritional epidemiology in Bangladesh and other low-income countries with widely varying dietary practices faces the challenge of identifying prudent, affordable, and culturally acceptable diets. Recently, dietary pattern analysis has emerged as an alternative approach to studies of diet and chronic diseases. Instead of evaluating the influences of individual nutrients or foods, pattern analysis examines the effects of the overall diet. Major dietary patterns have been related to CVD risk in studies conducted in Western countries (7–13). Recent intervention trials such as the Dietary Approaches to Stop Hypertension (DASH) trial in the United States found short-term beneficial effects of the DASH diet (fruit, vegetables, low-fat dairy products, and reduced fat) on blood pressure in hypertensive and borderline hypertensive patients (14). However, no large epidemiologic studies have systematically evaluated associations of dietary factors or patterns with blood pressure in a low-income population.
Using the baseline data of the Health Effects of Arsenic Longitudinal Study (HEALS), we performed a cross-sectional analysis to examine the associations between high blood pressure and dietary factors in a population of rural Bangladesh. We first evaluated relations between the prevalence of general hypertension, defined as a diastolic blood pressure (DBP) 90 mm Hg or a systolic blood pressure (SBP) 140 mm Hg (15, 16) or currently using antihypertensive medication, and intakes of nutrients that have been associated with hypertension in the literature. We then performed principal component analysis to identify major dietary patterns in the population and examined the associations between these dietary patterns and the prevalence of general hypertension.
SUBJECTS AND METHODS
The Health Effects of Arsenic Longitudinal Study
The HEALS is an ongoing prospective cohort study in Araihazar, Bangladesh. The principal aim of the HEALS is to investigate the health effects of arsenic exposure from drinking water. Details of the study methods have been presented elsewhere (17) and are summarized here. Briefly, before subject recruitment, water samples and global positioning system data were collected for a set of 5966 contiguous wells in a well-defined geographic area of 25 km2 in Araihazar, Bangladesh. Demographic information on users of these wells was collected to create a sampling frame for the HEALS (18). Between October 2000 and May 2002, 11 746 men and women aged 18 y were recruited, with a participation rate of 97.5%. Extensive baseline interviews were conducted to collect information on history of well use, demographics, and lifestyle characteristics. Dietary intakes were assessed by using a newly developed and validated food-frequency questionnaire (FFQ) (19). In addition, trained physicians administered a comprehensive physical examination that included blood pressure measurements. Verbal consent was obtained from each eligible respondent who agreed to participate in the study. The study procedures were approved by the Columbia University Institutional Review Board and the Ethical Committee of the Bangladesh Medical Research Council.
Ninety-seven percent of the participants completed the FFQ without missing values on any food item. There were no significant differences between those who missed at least one food item and those who did not in terms of age, sex, and markers of socioeconomic status such as educational attainment (data not shown). Blood pressure measurements were available for 11 458 participants. Participants who declined the physical examination, mainly due to a lack of time availability, did not have their blood pressure measured. Participants with available blood pressure measurements were somewhat more likely to be more educated and to own land or a television (although not statistically significant) than were those without a blood pressure measurement (data not shown). The present analysis included the 11 116 participants with at least one blood pressure measurement at baseline and a complete FFQ.
Measurements of nutritional intakes
Dietary intakes were measured at baseline of the HEALS with a semi-quantitative, 39-item FFQ designed for the study population. Trained interviewers completed the FFQ through face-to-face interviews. Detailed information on the design and the validation of the FFQ are published elsewhere (19) and are summarized only briefly here. On the basis of pilot work conducted with focus groups, common food items were included in the FFQ, and food items with intake frequencies less than once per month during the past year were deemed to be insignificant. To assess the validity of the FFQ, two 7-d food diaries were completed in 2 separate seasons by trained interviewers for 189 of 200 participants randomly selected from the overall HEALS study population. Correlations between consumptions measured by the FFQ and those measured by the food diary indicated that the validity of the FFQ in measuring long-term intakes of common food items, macronutrients, and some micronutrients was good. Specifically, correlations for macronutrients and common micronutrients including total fat, monounsaturated fat, polyunsaturated fat, saturated fat, protein, carbohydrate, dietary fiber, sodium, potassium, vitamin B-6, vitamin B-12, riboflavin, manganese, thiamine, and iron ranged from 0.30 to 0.76. We used both the US Department of Agriculture Nutrient Database for Standard Reference (abbreviated version) (20) and an Indian food nutrient database (21) to convert food intakes to nutrient intake values.
Blood pressure measurements
Blood pressure was measured with automatic sphygmomanometers by trained physicians. Measurements were taken with subjects in a seated position after 5 min of rest, with the cuff around the upper left arm, in accordance with recommended guidelines. The adjustable adult arm cuff fits arms that are 9-13 inches (23-33 cm) in circumference. Also, a large adult arm cuff is available for additional comfort for arm sizes of 13-17 inches (33-43 cm). The model of the automatic sphygmomanometers used (Omron automatic blood pressure monitor, HEM 712-C; Omron Healthcare GmbH, Hamburg, Germany) has been validated to have 85% of readings falling within 5–10 mm Hg of the mercury standard (22). For respondents found to have SBP 140 or DBP 90 mm Hg at the first measurement, 2 additional measurements were taken after 2–3 min of rest, and the lowest reading was recorded.
In addition, information on the use of antihypertensive medicines was extracted from the questionnaires. The study participants were asked to show all medicines they were currently taking, and the interviewers recorded generic names. Antihypertensive medicines that were reported included diuretics such as spironolactone, ß blockers such as metoprolol and propranolol, angiotensin-converting enzyme inhibitors such as lisinopril, angiotensin II antagonists such as losartan, and Ca2+ channel blockers such as nifedipine. One hundred fourteen participants among the overall HEALS participants (111 in the present analysis) were identified as taking antihypertensive medicines at the time of the baseline interview. These patients were excluded from some analyses in the present study (see the statistical analyses section).
Statistical analyses
Descriptive analyses
We first conducted univariate analyses to describe the distribution of nutrient intakes and demographic attributes by use of antihypertensive medicine and classification of hypertension defined according to the Seventh Report of the Joint National Committee on Prevention, Detection, Evaluation, and Treatment of High Blood Pressure (15). Nutrient patterns considered in the analysis were defined on the basis of nutrients with interaction effects on hypertension, including the ratio of dietary sodium to potassium intake and the ratio of saturated to unsaturated fat intake. All nutrient intakes were adjusted for total energy intake by the residual method (23).
Food pattern derivation
All statistical analyses were conducted with SAS (version 9.1; SAS Institute Inc, Cary, NC). We performed principal component analysis by using the PROC FACTOR procedure in SAS to identify dietary patterns. To avoid any arbitrary decisions on food group definitions, food items were directly entered into the principal component analysis as average daily intakes in grams or milliliters. An orthogonal rotation (the varimax option in SAS) was used to derive factors (dietary patterns) uncorrelated to one another for better interpretability. To determine the number of meaningful diet patterns, conventional criteria for principal component analysis including eigenvalue, the scree test, proportion of variance accounted for, and the interpretation criterion were considered (24). A diet pattern with an eigenvalue > 1.00 is considered to account for a greater amount of variance than are patterns with eigenvalues 1.00 (24). We focused on diet patterns with eigenvalues 1.5 to limit the number of diet patterns and to better identify meaningful diet patterns. Each food item received a factor loading associated with each diet pattern, and the factor loading represents the correlation coefficient between the food item and the diet pattern. To indicate a subject's relative standing of each diet pattern in the population, a factor score, which is a linear composite of the optimally weighted food items by factor loadings, was constructed for each dietary pattern. Each subject received a factor score for each identified diet pattern.
Evaluation of associations between nutrient intakes and general hypertension
Adjusted prevalence odds ratios (PORs) for general hypertension (SBP 140 or DBP 90) were calculated by using logistic regression to compare participants in quintiles of nutrient intakes and quintiles of factor scores and factor scores adjusted for energy intake by using the residual method. In addition, the relations of SBP and DBP with nutrient intakes and diet patterns were evaluated by using multiple linear regression analysis. In linear regression models, values of nutrient intakes were log-transformed. Participants treated with blood pressure medicines were considered as cases of general hypertension in logistic regression analysis of general hypertension and were excluded from linear regression analysis of SBP and DBP. We first adjusted for age, sex, and total energy intake, and in separate models, we additionally controlled for 1) BMI, smoking status, and educational attainment, and 2) BMI, smoking status, educational attainment, and intakes of all other nutrients of interest. Although educational attainment, a marker of socioeconomic status, may be an antecedent of dietary factors, it may be related to hypertension through pathways other than its influence on diet. Therefore, we included educational attainment in the model. We performed stratified analysis to evaluate whether associations between the prevalence of hypertension and diet patterns differed by sex.
RESULTS
Associations between intakes of individual nutrients and nutrient patterns with hypertension
The proportions of television ownership and land ownership, average BMI, educational attainment, and most nutrient intakes differed by stage of hypertension (Table 1).
View this table:
TABLE 1. Distributions of demographics, socioeconomic variables, and dietary nutrition intakes by classification of blood pressure1
No consistent association was found between the ratio of dietary sodium to potassium intake and the prevalence of hypertension (Table 2). The positive associations of prevalence of hypertension with total fat intake, vitamin B-12 intake, and the ratio of saturated to unsaturated fat intake (model 1) were no longer apparent after additional adjustment for BMI, smoking status, and educational attainment (model 2). The prevalence of hypertension was positively associated with protein intake and was inversely associated with intakes of vitamin B-6, riboflavin, magnesium, fiber, and carbohydrate (P = 0.06 or P < 0.05, model 2). In model 2 of the linear regression analysis, 1-unit increases on the log scale in intakes of vitamin B-6 (mg), magnesium (mg), and fiber (g) were associated with changes of –3.70 (95% CI: –6.71, –0.69), –2.17 (95% CI: –3.72, –0.62), and –1.13 (95% CI: –1.79, –0.47) in DBP (mm Hg), respectively. The directions of the associations of these nutrients with SBP and the associations of other nutrients with both DBP and SBP were all consistent with the directions seen in the logistic regression analysis of general hypertension. However, none of the other associations were significant at P < 0.05 (results not shown). With more adjustments for intakes of other nutrients (model 3), protein intake remained positively associated with the prevalence of hypertension, whereas intakes of fiber, carbohydrate, and riboflavin were significantly or marginally significantly (for riboflavin and fiber) inversely associated with the prevalence of hypertension. For other nutrients, patterns of PORs in model 3 were not significantly different from those in model 2. None of the nutrients were significantly associated with SBP or DBP at P < 0.05 in linear regression analysis with additional adjustments for all other nutrients (results not shown).
View this table:
TABLE 2. Adjusted prevalence odds ratios (PORs) for general hypertension by quintile (Q) of nutrient intakes1
Associations of dietary patterns with nutrient intakes and socioeconomic status
Factor loadings for the 3 dietary patterns are presented in Table 3. Factor loadings represented correlation coefficients between individual food items and dietary patterns. Food items with strong loadings contribute to a dietary pattern. Three major dietary patterns were identified and named according to the foods that loaded most heavily on the pattern. The "balanced" pattern was characterized by intake of steamed rice, red meat, small fish, fruit, and vegetables. The "animal protein" pattern was heavily loaded with greater intakes of large fish, eggs, milk, poultry, red meat (beef and mutton), bread, and fruit. The "gourd and root vegetable" pattern largely consisted of greater intakes of a variety of squashes, pumpkin, sweet potato, radish, and yam.
View this table:
TABLE 3. Factor loading matrix for dietary patterns derived from principal component analysis1
Shown in Table 4 are the Spearman correlation coefficients between nutrient intakes relevant to blood pressure and the factor scores, which represent the relative standing of each person for each of the dietary patterns. The balanced pattern was more positively related to intakes of carbohydrate, fiber, folate, vitamin B-6, riboflavin, polyunsaturated fat, monounsaturated fat, and magnesium than were the other 2 diet patterns. On the other hand, the animal protein pattern was most positively associated with intakes of protein, sodium, potassium, total fat, vitamin B-12, saturated fat, ratio of sodium to potassium, and ratio of saturated to unsaturated fat. The gourd and root vegetable pattern was related to higher intakes of folate and fiber.
View this table:
TABLE 4. Spearman correlations between factor scores for dietary patterns and daily nutrient intakes1
The animal protein pattern was independently and positively associated with the prevalence of cigarette smoking and markers of socioeconomic status, including educational attainment, television ownership, and land ownership (P < 0.01; Table 5). The positive associations between the balanced pattern and markers of socioeconomic status were weaker (P = 0.10 for educational attainment, P = 0.05 for land ownership, and P = 0.01 for television ownership). The gourd and root vegetable pattern was inversely associated with television ownership (P < 0.01).
View this table:
TABLE 5. Sociodemographic characteristics by quintile (Q) of dietary factor scores
Associations between dietary patterns and high blood pressure
The balanced pattern was inversely associated with DBP (P < 0.05), SBP (P < 0.05), and the prevalence of hypertension, with PORs ranging from 0.78 to 0.70 comparing participants with higher quintiles with participants in the bottom quintile of factor scores (Table 6). The animal protein pattern was positively associated with DBP (P < 0.05), SBP (P < 0.05), and the prevalence of hypertension, with PORs ranging from 1.38 to 1.73 comparing higher quintiles with the bottom quintile of factor scores. These associations remained significant even after adjustment for BMI, educational attainment, and smoking status (model 2). However, the dose-response relation between the animal protein pattern and the prevalence of hypertension was no longer apparent (model 2, P = 0.23 for trend test). The gourd and root vegetable pattern was positively associated with SBP but was not associated with either DBP or the prevalence of hypertension in both model 1 and model 2. With additional adjustment for all nutrient intakes relevant to blood pressure, the inverse associations of the balanced pattern with DBP and the prevalence of hypertension remained significant or marginally significant, whereas the associations of the animal protein pattern with blood pressure were no longer apparent. The gourd and root vegetable pattern remained positively associated with SBP. Stratified analysis showed that associations of diet patterns and blood pressure were similar in men and women. Analysis results based on factor scores unadjusted for energy intake were not significantly different (results not shown).
View this table:
TABLE 6. Associations of dietary factor scores with general hypertension, systolic blood pressure (SBP), and diastolic blood pressure (DBP)1
DISCUSSION
In this cross-sectional analysis, we evaluated the associations of the prevalence of high blood pressure with major dietary nutrient intakes and dietary patterns in a rural Bangladeshi population. We identified 3 major diet patterns and found that the balanced pattern was inversely and the animal protein pattern was positively associated with the prevalence of hypertension.
Our goal was to identify actual eating patterns in this rural Bangladeshi population and to examine their relations with hypertension. After the analysis was controlled for BMI, smoking status, and educational attainment, the PDRs for hypertension associated with nutrients were mostly diminished. On the other hand, the PDRs associated with dietary patterns remained apparent. This finding indicates that compared with single nutrients, combinations of nutrients (dietary patterns) represent stronger independent risk factors of hypertension. The pattern of PORs suggests that participants in each of the higher 4 quintiles of the factor scores for the animal protein pattern were more likely to have hypertension than were participants in the lowest quintile. Furthermore, the dose-response association between the animal protein pattern and hypertension was not apparent (model 2), which suggests that a threshold may exist for the effect of animal protein pattern on hypretension in this population. The PORs associated with the animal protein pattern were no longer significant after adjustment for intakes of nutrients, which suggests that the effect of the animal protein pattern may be due to the joint effects of nutrients. On the other hand, the inverse association between the balanced pattern and hypertension remained significant or marginally significant, which indicates that the effect of the balanced pattern was more than what can be explained by single nutrients and that a nutrient-nutrient interaction may exist.
Although major food patterns and the meaning of a healthy diet differ by populations and cultures, our finding that a diet richer in meats and fat was positively associated with general hypertension is consistent with the literature. We also found an inverse association between the prevalence of hypertension and the balanced pattern, which reflects a diet similar to the DASH diet. A vegetarian diet has been found to be associated with some degree of protection against hypertension compared with nonvegetarian diets in Western populations (25, 26). However, we did not find an inverse association between adherence to the gourd and root vegetable pattern and the prevalence of hypertension in this lean population (average BMI = 19.8). The positive association between the gourd and root vegetable pattern and SBP requires future investigation.
We found that intakes of protein, carbohydrate, and fiber were significantly associated with the prevalence of general hypertension, even after the analysis was controlled for BMI and other nutrients. Dietary fiber may lower blood pressure by reducing the glycemic index of foods, and soluble fiber may improve mineral absorption in the gastrointestinal system, which may have an indirect favorable effect on blood pressure (27). Epidemiologic studies of the relations of hypertension with intakes of protein (28–30) and carbohydrate (29, 31) have shown conflicting results. The positive association between carbohydrate intake and hypertension may be due to carbohydrates with a high glycemic index (32). Although we did not categorize carbohydrate by glycemic index, it is clear that the food list consisted of foods with a low glycemic index. In an additional analysis, we found that the positive association between protein intake and hypertension was mostly due to that between animal protein intake and hypertension (data not shown).
Several other statistical methods have been used to identify dietary patterns in the population. The a priori approach involves constructing indexes for the consumption of food items on the basis of previous knowledge of a "healthy" diet. Diet patterns constructed by using the a priori approach are limited to current knowledge (33). In addition, in a low-income population such as the one of the present analysis, common foods were mostly considered healthy by Western standards, which poses a challenge to index construction. Principal component analysis, on the other hand, generates diet patterns a posteriori on the basis of foods that tend to be consumed together. However, principal component analysis involves several arbitrary decisions, including the number of factors to be extracted, the method of the rotation, and the interpretation of the derived patterns (33). We extracted the top 3 dietary patterns on the basis of standard criteria of principal component analysis. The simple diet of the study population and the smaller number of foods in the FFQ allowed us to avoid arbitrarily aggregating food items into food groups.
The present study was a cross-sectional study. However, health awareness and knowledge about the etiology of hypertension is very limited in rural Bangladesh, and persons with hypertension at an early stage are often unaware of their disease. Other potential confounders such as physical activity and psychosocial stress that may be related to both the risk of hypertension and dietary patterns were not fully controlled for in the analyses. Potential confounding due to physical activity and psychosocial stress was controlled for to the extent that BMI and educational attainment were related to these factors. Dietary intakes were measured by FFQ, and therefore measurement errors are expected. However, the validation study for the FFQ showed that the validity of the FFQ in measuring long-term intakes of macronutrients and common foods was good (19), and the major components of the identified dietary patterns were mostly common foods. Although national nutritional fortification programs have been proposed (34), they are not currently in effect. In addition, alcohol drinking, which affects nutritional absorption in the body, is rare in Bangladesh because of religious beliefs. These features strengthen the validity of the dietary intakes measured by the FFQ. The fact that the balanced pattern and the animal protein pattern were more associated with nutrients that have been related to a reduced risk and an increased risk of hypertension, respectively, further strengthens the study findings.
Although dietary determinants of high blood pressure in low- and middle-income populations may be similar to those in high-income populations, their associations with socioeconomic status are different. We found that the animal protein pattern was strongly positively associated with markers of socioeconomic status and the prevalence of cigarette smoking. This observation supports the conception that the study population was at an early stage of the acculturation process, during which the CVD epidemic would first affect members of the higher social classes, who are the first to change from a lower-risk to a higher-risk lifestyle characterized by diets rich in fat, a sedentary lifestyle, and smoking. The Bangladeshi population is likely to have a double burden of disease for different subsets of the population: a heavy communicable disease burden for the poor and an increasing noncommunicable disease burden for the more affluent. The task of CVD control in Bangladesh and in other similar low-income populations may therefore be more complex than in developed countries. Hypertension is easy to identify and is universally recognized as a determinant of CVD occurrence. Community- or provider-based health promotion programs carefully designed for the control of hypertension are needed.
ACKNOWLEDGMENTS
We thank our staff, the fieldworkers, and the study participants in Bangladesh, without whom this work would not have been possible. We also thank our colleagues Joseph Graziano, Paul Brandt-Rauf, and Lydia Zablotska for their helpful comments.
YC led the writing and was responsible for the concept and analysis of the study. HA, PF-L, and GRH help to interpret the data and revise the manuscript. FP was responsible for the collection, shipment, and processing of samples. None of the authors had a conflict of interest in relation to this study.
REFERENCES