To examine the relationship between dietary carbohydrates, glycemic load and high-density lipoprotein cholesterol (HDL-C) concentrations in Asian Indians, a high-risk group for diabetes and premature coronary artery disease.
The study population comprised of 2043 individuals aged 20 years randomly selected from Chennai Urban Rural Epidemiological Study (CURES), an ongoing population-based study on a representative population of Chennai (formerly Madras) city in southern India. Participants with self-reported history of diabetes or heart disease or on drug therapy for dyslipidemia were excluded from the study. Dietary carbohydrates, glycemic index and glycemic load were assessed using a validated interviewer administered semiquantitative Food Frequency Questionnaire (FFQ).
Both dietary glycemic load (P<0.0001) and total dietary carbohydrate intake (P<0.001) were significantly associated with higher serum triglyceride levels and lower serum HDL-C levels. For the lowest to highest quintile of glycemic load, the multivariate-adjusted mean HDL-C values were 44.1 mg per 100 ml and 41.2 mg per 100 ml (6.6% difference, P for trend<0.001), while for total carbohydrate it was less (5% difference, P for trend=0.016). The pattern of decrease in HDL-C for the lowest to highest quintile of glycemic load was more pronounced among men (1st vs 5th quintile: adjusted HDL-C: 4.3 mg per 100 ml decrease (10.3%)) than women (1st vs 5th quintile: adjusted HDL-C: 3.2 mg per 100 ml decrease (6.9%)).
Our findings indicate that both total carbohydrates and dietary glycemic load intake are inversely associated with plasma HDL-C concentrations among Asian Indians, with dietary glycemic load having a stronger association.
Asian Indians have a typical dyslipidemia characterized by low high-density lipoprotein cholesterol (HDL-C) and high triglyceride levels which predisposes them to premature coronary artery disease (CAD) (Reddy and Yusuf, 1998; Mohan and Deepa, 2004; Gupta, 2005). This is part of the so-called ‘Asian Indian Phenotype’, the other features of which include increased insulin resistance and high prevalence of diabetes which are associated with excess central, but not generalized, obesity (McKeigue et al., 1991; Ramachandran et al., 1997; Mohan et al., 2003).
Low levels of HDL-C are an important risk factor for CAD (Beckles et al., 1986; Enas et al., 1996). Plasma concentrations of HDL-C are influenced by both genetic and lifestyle factors. Although it is difficult to alter genetic factors, modifiable environmental factors such as diet could be targeted in interventions aimed at raising HDL-C. Studies linking diet and heart disease have focused largely on dietary fats (Gordon, 1988). The role of dietary carbohydrates is less well recognized, but needs to be studied as carbohydrates provide the bulk of calories of the Asian Indian diet (Gopalan et al., 2004). High-carbohydrate diets are known to be associated with hypertriglyceridemia (Grundy and Denke, 1990; Hellerstein, 2002).
The traditional dietary advice is to increase intake of complex carbohydrates while limiting refined carbohydrates. However, recent studies indicate that this view of carbohydrates is too simplistic. Both the quantity and quality of carbohydrates influence the metabolic response to the ingestion of carbohydrate (FAO/WHO, 1998). Studies have shown that an increased intake of carbohydrates can lower serum HDL-C concentrations (Mancini et al., 1973; Yagalla et al., 1996; Anwar et al., 2007). In addition to the quantitative relationship between carbohydrate intake and HDL-C concentration, recent studies suggest that glycemic load, which depends on both the quality as well as the amount of dietary carbohydrates, may also influence HDL-C concentrations (Ford and Liu, 2001).
In this paper, we look at the association of both dietary carbohydrates and glycemic load with HDL-C concentrations in an urban south Indian population and is the first study to our knowledge in south Asians who are at high risk of diabetes (Mohan et al., 2003) and premature CAD (McKeigue, 1992; Reddy and Yusuf, 1998).
Subjects and methods
Study subjects were recruited from the Chennai Urban Rural Epidemiology Study (CURES), an ongoing epidemiological study conducted on a representative population (aged20 years) of Chennai city (formerly Madras) in southern India with a population of about 5 million people. The methodology of the study has been published elsewhere (Deepa et al., 2003). Our website http://www.drmohansdiabetes.com (under the link ‘publications’) provides details of the sampling frame. Briefly, in Phase 1 of the urban component of CURES, 26 001 individuals were recruited based on a systematic random sampling technique.
Phase 2 of CURES deals with studies on prevalence of microvascular and macrovascular complications of diabetes. Phases 1 and 2 are not discussed further in this article.
In Phase 3 of CURES, every tenth subject recruited in Phase 1 (n=2600) was invited to our center for detailed anthropometric measurements and biochemical tests. Of these, 2220 participated in the dietary assessment study (2220/2600 samples: response rate=85.3%). Subjects with self-reported history of diabetes or cardiovascular diseases or on drug therapy for dyslipidemia were excluded (n=177) because therapy could have altered the lipids including the HDL-C levels. The remaining 2043 subjects were included in the present study.
All study subjects underwent an oral glucose tolerance test (OGTT) using 75 g glucose load. HDL-C was measured enzymatically (direct method, polyethylene glycol-pretreated enzymes, Roche Diagnostics, USA). The intra-assay and the interassay co-efficient of variation for direct HDL-C levels were 2.0 and 3.6%, respectively. Values are reported in milligrams per deciliter (100 ml); to convert into millimoles per liter, multiply by 0.0259. Anthropometric measurements including weight, height, waist and hip measurements were obtained using standardized techniques as described earlier (Deepa et al., 2003). Demographic and socioeconomic characteristics, medical history, medications, family history, smoking (current smokers: yes or no) and alcohol consumption (current drinker: yes or no) were also obtained (Deepa et al., 2003). Details on physical activity were assessed using a previously validated physical activity questionnaire (Mohan et al., 2005). The protocol for the study was approved by the Institutional Ethics Committee of the Madras Diabetes Research Foundation and informed consent was obtained from all study subjects.
Dietary intakes were assessed using a previously validated and published interviewer administered semiquantitative food frequency questionnaire (FFQ) (meal-based) containing 222 food items to estimate food intake over the past year and took between 20 and 30 min to complete (Sudha et al., 2006). The nutritionists responsible for the data collection were well trained in the methodology to be used before the field work started. Individuals were asked to estimate the usual frequency (number of times per day, week, month or year/never) and the usual serving size of the portion of the various food items in the FFQ. Common household measures such as household cups, bowls, ladles, spoons (for the cooked foods like vegetables) wedges, circles of different diameter and visual atlas of different sizes of fruits (small, medium, large) were shown to assist the individuals in estimating portions. Information on total energy intake and macronutrients were obtained using the same FFQ. Sucrose and fructose included those present naturally as well as that added in cooking and in processed foods. Sucrose and fructose content of the foods were obtained from published values (Betty et al., 2002). A detailed description of the development of FFQ and the data on reproducibility and validity had been published (Sudha et al., 2006).
Assessment of glycemic load
The glycemic index of the food was defined as the 2-h incremental area under the blood glucose–response curve after consumption of a food portion containing a specific amount (50 g) of available carbohydrate, divided by the corresponding area after consumption of a portion of a reference food such as glucose or white bread containing the same amount of available carbohydrate, and multiplied by 100 to be expressed as a percentage (Jenkins et al., 1981; Wolever et al., 1991). To determine the glycemic index, each food item on the FFQ was directly matched to foods in the international table of glycemic index (Foster-Powell et al., 2002) and in several publications on the glycemic index of Indian foods (Raghuram et al., 1987; Mani et al., 1990, 1992, 1993, 1997) The glycemic index of the mixed meals could be predicted from the glycemic indices of the individual foods as proposed by FAO/WHO (1998) (weighted by their proportion of carbohydrate contents). Precise glycemic index values may vary because of differences in methods, intra subject variation and processing and cooking of the food. To minimize such variation in the glycemic index values, we calculated glycemic index in the current study on the basis of mean of glycemic index values from different studies measuring the glycemic index of similar foods (Foster-Powell et al., 2002). In the present study, glucose was used as the reference (glycemic index for glucose=100). The white bread-based glycemic index values were transformed into glucose-based glycemic index values by multiplying the white bread-based glycemic index by 0.7 as discussed previously (Foster-Powell et al., 2002).
Calculation of dietary glycemic load
Glycemic load is the arithmetic product of glycemic index and the total carbohydrate (g) and has been physiologically validated for glucose response in lean adults and overweight subjects (Salmeron et al., 1997). Dietary glycemic load of the individual serving was calculated by multiplying the carbohydrate content (grams per serving) of each food by its glycemic index value. We then multiplied this glycemic load value by the frequency of consumption and summed these products to obtain average daily dietary glycemic load.
Dietary glycemic load and total carbohydrates intake were adjusted for total energy by linear regression with the nutrient as the outcome and total energy intake as the predictor. The residuals from this model were added to the expected value of the nutrient at an average energy intake (Willet et al., 1997).
Responses to the individual food items were converted to average daily intake to calculate the macronutrients and selected micronutrients for each participant using an in-house EpiNu India database and software (Nutritional Epidemiology, Food and Nutrient database, version 1.0, Chennai), developed by the Department of Nutrition and Dietetics Research of the Madras Diabetes Research Foundation. This database contains comprehensive data for 1500 recipes (both commercial and non-commercial) and 60 nutrients.
All analysis were conducted using the statistical software package SPSS (10.0 version; SPSS Inc., Chicago, IL, USA). Subjects were divided into quintiles of glycemic load and the mean (±s.d.) or percentages of each was reported and compared for the descriptive characteristic. Univariate regression analysis was carried out using HDL-C as dependent variable to identify the risk factors for low HDL-C. To evaluate possible confounders for dietary variables; we further adjusted separately for intake of protein (% energy), fat (% energy), dietary fiber (g day−1), sucrose and fructose (g day−1). Multivariate regression analysis was carried out to determine the association of HDL-C with energy adjusted glycemic load and dietary carbohydrates after accounting for the possible confounders that related to HDL-C, glycemic load and total carbohydrates. The model constructed was controlled for age (in years), sex (dichotomous), low-density lipoprotein (LDL) cholesterol (mg per 100 ml), serum triglyceride (mg per 100 ml), sucrose (g day−1), fructose (g day−1), dietary protein (% of energy), dietary fat (% of energy), saturated fat (% of energy), monounsaturated fat (% of energy), polyunsaturated fat (% of energy), dietary fiber per 1000 kcal (g day−1), body mass index (BMI) (continuous variable), smoking (categorical), alcohol (categorical) and physical activity (regular/strenuous=0; no exercise and sedentary=1). The mean (±s.d.) values for HDL-C were calculated according to quintiles of dietary glycemic load and total carbohydrates after adjustment for potential confounding variables using one-way analysis of variance (with linear trend significant differences). All tests of significance were two-tailed and a P-value of <0.05 was considered significant.
The study included 2043 subjects, 886 men; mean age 41.1±13.2 years (range=20–84 years) and 1157 women; mean age 40.2±12.3 years (range=20–80 years). The mean glycemic load was 386.8±123.4 for men and 341.1±111.4 for women.
Table 1 presents selected baseline characteristics according to unadjusted quintiles of glycemic load. The mean dietary glycemic load intake ranged from 214 (lowest quintile) to >535 (highest quintile). Higher intake of dietary glycemic load was associated with younger age (P<0.0001), greater BMI (P<0.0001), waist circumference (P<0.0001), waist-to-hip ratio (P<0.0001), HDL-C (P<0.0001), serum triglyceride (P=0.002) and total energy intake (P<0.0001) but with lower prevalence of smoking (P=0.002) and alcohol (P<0.0001). Higher intake of dietary glycemic load was also associated with higher physical inactivity (P=0.005), higher intake of sucrose (P<0.0001) and fructose foods (P=0.003) and lower intake of total dietary fat (P<0.0001), saturated fat (P<0.0001), polyunsaturated fat (P<0.0001), monounsaturated fat (P<0.0001) and protein (P<0.0001).
Dietary glycemic load and total carbohydrate intake was inversely related to HDL-C concentrations (r-value=−0.142 vs −0.217 P<0.0001) and positively related to fasting plasma triglyceride (r-value=0.262, P=0.021 vs r-value=0.285, P=0.021) even after adjustment for age, sex, physical activity, BMI, waist-to-hip ratio, smoking, alcohol, total energy, total dietary fat (%), protein (%), dietary fiber per 1000 kcal, sucrose and fructose. Higher dietary glycemic load (r=value=−0.070; P=0.003) and total carbohydrate (r=−0.089 P<0.001) were also associated with lower HDL–LDL ratio. However, while total carbohydrate was inversely associated with total cholesterol (r-value=−0.058, P=0.022), no such association was observed for dietary glycemic load. No significant association was observed with LDL cholesterol for either glycemic load or total carbohydrates.
In the univariate regression analysis, HDL-C concentrations showed a significant negative association with age (P<0.001), BMI (P<0.001), waist-to-hip ratio (P<0.001) smoking (P<0.001), alcohol (P<0.05), serum triglyceride (P<0.001), total energy (P<0.001), energy adjusted glycemic load (P<0.001), total carbohydrates (P<0.001) and percentage of energy from protein (P<0.01) and a positive association with female sex (P<0.001), LDL cholesterol (P<0.001), physical activity (P<0.001), dietary fiber (P<0.001) and fructose (P<0.001). Dietary fat (% energy) and sucrose were not significantly related to HDL-C. Univariate analysis, using HDL-C as the dependent variable and energy adjusted glycemic load as independent variable was performed and other variables that were associated with glycemic load and HDL-C were entered along with dietary glycemic load. Adjusting for covariates such as age, sex, BMI, waist-to-hip ratio, physical activity, LDL cholesterol, serum triglyceride, smoking, alcohol, total dietary fat (% energy), dietary protein (energy %), dietary fiber per 1000 kcal did not alter its association (β=−0.013, P<0.001), while adding sucrose and fructose changed the regression coefficient slightly but did not abolish the significant relationship with HDL-C. (β=−0.023, P<0.001).
The unadjusted mean serum HDL-C levels decreased by 7.6 mg per 100 ml, from 47.4 mg per 100 ml in the lowest quintile to 39.8 mg per 100 ml in the highest quintile of energy adjusted dietary carbohydrate (P for trend <0.001). After multivariate adjustment, mean HDL-C concentration was 44.2 mg per 100 ml in the lowest and 42.0 mg per 100 ml in the highest quintile (5% difference, P for trend <0.016). The pattern of decrease in HDL-C was more pronounced among men (adjusted HDL-C: 4.2 mg per 100 ml in the lowest and 37.9 mg per 100 ml in the highest quintile (10.2% decrease), P for trend=0.024) than women (adjusted HDL-C: 46.0 mg per 100 ml in the lowest and 45.2 mg per 100 ml in the highest quintile (1.7% decrease); P for trend=NS) (Table 2).
Mean concentrations of HDL-C according to quintiles of glycemic load intake are presented in Table 3. As the glycemic load intake increased, the unadjusted mean serum HDL-C levels decreased by 2.5 mg per 100 ml, from 43.8 mg per 100 ml in the lowest quintile to 41.3 mg per 100 ml in the highest quintile (5.7% difference, P for trend <0.0001). For the lowest and highest quintile of glycemic load intake, the corresponding multivariate adjusted mean HDL-C concentration were 44.1 mg per 100 ml and 41.2 mg per 100 ml (1st vs 5th quintile: adjusted HDL-C: 6.6% decrease, P for trend <0.001). The pattern of decrease in HDL-C was more pronounced among men (1st vs 5th quintile: adjusted HDL-C: 4.3 mg per 100 ml (10.3% decrease)) than in women (1st vs 5th quintile: adjusted HDL-C: 3.2 mg per 100 ml (6.9% decrease)).
To our knowledge, this is the first epidemiological study to examine the relationship between dietary glycemic load with HDL-C concentration in an Asian Indian population. Studies have demonstrated an association between quality as well as quantity of dietary carbohydrates in Europeans (Ford and Liu, 2001; Liu et al., 2001) and one in an Asian (Japanese) population (Amano et al., 2004; Murakami et al., 2006). This study differs from the earlier studies in that it was done in Asian Indians who are at much higher risk of premature CAD (McKeigue, 1992; Reddy and Yusuf, 1998).
Complex carbohydrates diets have been shown to protect against cardiovascular disease and type II diabetes (Ravussin et al., 1994), and this was thought primarily to be the result of the high fiber and micronutrients content of these diets (Burkitt and Trowell, 1977). However, due to advances in food processing technology and milling process, today's fiber rich grains are losing substantial amount of fiber and these complex carbohydrates are becoming fiber-depleted starches (Schmidhuber and Shetty, 2005). Studies have shown that these refined grains and sugar products nearly always maintain much higher glycemic index and evidence showing that long-term consumption of high glycemic load carbohydrates may increase the risk of obesity, coronary heart disease and type II diabetes (Thorburn et al., 1987; Ludwig, 2002).
The main finding in this study is that higher consumption of foods with high glycemic load was associated with lower HDL-C and increased serum triglyceride in this population. Dietary glycemic load appeared to capture the combined effect of quantity and quality of carbohydrate consumed, better than total carbohydrates. The differential risk of low HDL was significantly higher in 5th quintile of glycemic load (OR=1.66, 95% CI=1.93, 2.31, P=0.003) compared to highest quintile (5th) of total carbohydrate (OR=1.12, 95% CI=0.81, 1.55, P=NS).
Our results suggest that both quality and quantity of carbohydrates (glycemic load) has significant effect on HDL-C levels compared to quantitative effect of carbohydrates alone. Overall, carbohydrate intake was weakly but significantly related to HDL-C in the multivariate adjusted model and showed borderline significance in men while no such association was observed for women. Our findings are consistent with those from earlier studies. Liu et al. (2001) showed an inverse relationship to HDL-C and a positive association with serum triglyceride and quality and quantity of dietary carbohydrate in 280 healthy post-menopausal women and reported that glycemic load best captured the effect, particularly in those prone to insulin resistance.
A similar effect was also observed in a large cross-sectional study, in the United States (Ford and Liu, 2001). The percentage decrease in HDL-C concentration was more among men (adjusted 8.06%) than women (adjusted 5.5%) in their study and this is similar to that observed in the present study (men vs women: adjusted difference 10.3 vs 6.9%) but the difference was higher than in the American study. However, differences such as dietary pattern and the population studied limit further interpretation of these study findings. In an intervention study, HDL-C decreased by 3% in subjects consuming diets containing <25% of energy from fat (P<0.05) (Knoop et al., 1997). A recent cross-sectional study (SHARE) which included Asians among other ethnic population (Europeans, South Asians in Canada and Chinese), did not show an inverse association between HDL-C concentration and glycemic load (Anwar et al., 2007).
Observational studies (Gordan et al., 1989) report that for every decrease of about 0.026 mmol l−1 (1 mg per 100 ml) in HDL-C concentration, the risk of coronary heart disease increased by 1.9% in men and 3.2% in women. Thus, the 0.1112 mmol l−1 (4.3 mg per 100 ml) decrease in HDL concentration in men and 0.0828 mmol l−1 (3.2 mg per 100 ml) in women when comparing the multivariate adjusted lowest and highest quintile of glycemic load in this study would theoretically translate into a 8.2 and 10.2% increase in CAD risk among men and women respectively due to increase in glycemic load in this Asian Indian population.
The glycemic load was much higher in our study (quintile: 1st 259 vs 5th 461) compared to that reported in European (quintile: 1st 117 vs 5th 180) (Liu et al., 2001) or other Asian populations (quintile glycemic load per 1000 kcal: 1st 31.1 vs 5th 148.5) (Murakami et al., 2006) and this was also associated with a lower HDL-C concentration in our population. In western studies, dietary glycemic index and glycemic load intake are determined by variety of food items. However, in Asian countries like India, diets are cereal-based and hence rich in carbohydrates which constitute about 60–80% of the total calories, especially in southern India, where ‘rice’ is the staple food.
Higher glycemic load is hypothesized to encourage a proatherogenic profile by elevating triglyceride and small dense LDL concentration, while reducing the HDL-C concentration by inducing fatty acid production in the liver and inhibiting the action of lipoprotein lipase through increased apolipoprotein CIII production, particularly in the presence of insulin resistance (Grundy, 1998; Grundy et al., 2002; Liu and Willett, 2002). This is critically important in Asian Indians, in the setting of insulin resistance and dyslipidemia.
This study has several limitations. First, being a cross-sectional study, it does not allow us to make a cause and effect inference. Prospective studies or randomized clinical trials are best suited to evaluate the role of carbohydrates both in terms of quality and quantity and the development of type II diabetes and coronary heart disease. Second, the calculation of the glycemic index was made by using values from the glycemic index tables of Foster-Powell et al. (2002), which are derived from foods (like fruits and vegetables) consumed in the United States, Canada and Australia as we do not have adequate data from India. Development of glycemic index databases in India, especially in the context of mixed meals is an urgent need. Third, the ability of the FFQ to measure sucrose and fructose content of Indian foods was limited largely due to incomplete food composition values available for both natural and processed foods. However, in this study substitutions of fructose and sucrose content for available foods were obtained from using the published USDA values. Finally, as the study population consisted of generally healthy persons, the clinical relevance of our findings to other groups, for example, people with diabetes and CHD remains to be elucidated.
However, this study also has several strengths: We investigated the association in a relatively large representative population of Chennai and hence the results can be extrapolated to urban India. Second, these are the first results on the relationships of quality as well as the quantity of carbohydrates with HDL-C concentration in a high-risk Asian Indian population. Third, the FFQ used in this study showed high reproducibility and validity for dietary glycemic index, glycemic load and total carbohydrates (Sudha et al., 2006). Fourth, misclassification of dietary exposure, which is always an important concern, was also minimized as the FFQ was designed to assess the long-term dietary intake (Gibson, 1990).
In conclusion, in Asian Indians, dietary glycemic load was inversely related to plasma HDL-C concentration and this was independent of total energy intake, dietary protein, dietary fat, dietary fiber, sucrose, fructose, alcohol and other known non-dietary risk factors such as age, sex, smoking, BMI, waist-to-hip ratio, serum triglyceride, LDL-C and physical activity. It is predicted that India would lead the world with highest rates of cardiovascular deaths in the future and it already leads the world in diabetes (Sicree et al., 2006). Therefore, it would be advisable to substitute foods with lower glycemic index as this could substantially reduce the glycemic load and therefore possibly the risk of diabetes and cardiovascular disease in the future.
We thank the Chennai Willington Corporate Foundation, Chennai for the CURES field studies. Special thanks to Ms R Tamilselvi and Ms M Ezhilarasi, research dietitians for their help with the manuscript. This is the 48th publication from Chennai Urban Rural Epidemiology Study (CURES 48).
About this article
High DHA dosage from algae oil improves postprandial hypertriglyceridemia and is safe for type-2 diabetics
International Journal of Diabetes in Developing Countries (2013)