The Indian population has the highest prevalence of diabetes worldwide [¹] and exhibits high-risk metabolic profiles at younger ages and lower body mass index (BMI) than their Western counterparts [²,³]. Although genetic susceptibility is likely to play a role in chronic disease etiology, the strong evidence for diet and other environmental factors [^4-7] suggest that such an epidemic may be preventable.

Diets across India have not been widely investigated, yet many believe that India may be in the midst of a "nutrition transition," [^8-10] where changes in diet parallel an expanding industrial economy and a rapidly progressing epidemic of obesity and chronic, non-communicable disease. In this emerging at-risk population, the suspected access to and adoption of a less healthy diet and lifestyle, and/or deviance from traditional and potentially protective behaviors [⁹,¹¹,¹²], may be linked to anthropometric factors and biological markers of chronic disease risk [³,¹³]. However, few have closely examined the role of diet in these patterns and relationships across a large, diverse, Indian population.

Large cohort studies in the U.S. and Europe have used dietary patterns analyses to bring to light common eating behaviors and their relationships with risk of cardiovascular disease and cancer [¹⁴,¹⁵]; however, no such studies exist in India. Furthermore, diet is expected to vary considerably within India across North-South regions [¹⁰,¹³,¹⁶] and with the exception of national food surveys [⁹,¹⁷], few comprehensive and up-to-date assessments of regional Indian diets are currently available [¹¹,¹²,¹⁸].

Utilizing detailed diet histories collected in a cohort feasibility study conducted across three diverse regions of India, our objective was to aggregate foods in to regional dietary patterns and to investigate associations with cardio-metabolic risk factors, such as abdominal adiposity, hypertension, diabetes, and dyslipidemia. As little is known regarding regional differences, we used an exploratory approach, factor analysis, to empirically identify dietary patterns reflecting actual eating behaviors within each of the study regions [¹⁹,²⁰].

Table 2 presents the characteristics of the India Health Study participants by region. Delhi had the greatest proportion of participants with higher education and income. Based on BMI alone, approximately half of the participants across the regions were overweight or obese, while 80-90% met the criteria for abdominal adiposity. By gender (data not shown), 4% of women and 6% of men had a BMI < 18.5, while 20% of women and 6% of men had a BMI ≥ 30 [²⁵]. Across the regions, the prevalence of hypertension ranged from 46 to 51% and the prevalence of diabetes ranged from 10 to 28%. In Delhi, participants had the lowest physical activity levels and more than half were dyslipidemic. In Trivandrum, participants appeared to consume a lower proportion of fat and higher proportion of carbohydrate than participants in Delhi or Mumbai. However, the distribution of macronutrients within each region did not vary substantially across tertiles of dietary patterns (Additional file 1: Appendix). Across all regions, concordance with dietary patterns across tertiles of mean factor scores did appear to differ quite markably with levels of physical activity, income, and education (Additional file 1: Appendix).

Table 3 shows the main results of the factor analysis and presents for each region the two retained factors and top loading food items defining each dietary pattern. Table 4 presents correlations between nutrient intakes, broader food groups, and dietary patterns within each region. In Delhi, the primary pattern, which we called "fruit-dairy," included fruit, fruit juice, and mixed dishes likely to contain cheese, yogurt, or other types of dairy. It was most strongly correlated with total fruit (r = 0.55), total dairy (r = 0.51), calcium, and cholesterol intake (r > 0.20). The secondary pattern in Delhi, "vegetables-pulses," was positively correlated with total vegetables (r = 0.46), total pulses (r = 0.32), total fat (r = 0.22), and retinol intake (r = 0.46), but inversely correlated with rice, meat, and protein intake (r < -0.20). In Trivandrum the "pulses-rice" pattern was defined by top loading food items or mixed dishes composed mainly of pulses and fermented rice (Table 3). This pattern was also inversely correlated with intakes of iron, calcium, retinol, and total fish (Table 4). The top loading items for the secondary "sweets-snacks" pattern in Trivandrum were mainly sweet and fried savory snacks (Table 3). This pattern was also positively correlated with fat, cholesterol, and retinol intake (r≥0.37), as shown in Table 4. In Mumbai, the primary "fruit-vegetables" pattern was the most strongly correlated with total vegetables and total fruit (r = 0.72), as well as calcium, iron, and retinol intake. The "snacks-meat" pattern in Mumbai was strongly correlated with intake of cholesterol (r = 0.57), meat (r = 0.46), and sweets (r = 0.32).

Multivariate-adjusted associations for dietary patterns with abdominal adiposity, hypertension, and diabetes within each region are presented in Table 5. In Delhi, the fruit and dairy pattern was positively associated with both abdominal adiposity and hypertension, but not associated with diabetes or dyslipidemia (not shown). No associations were found for the secondary pattern in Delhi. In Trivandrum, the pulses and rice pattern was inversely associated with diabetes (IFG ≥126 mg/dl) and/or pre-diabetes [IFG≥100 mg/dl; OR and 95% CI for highest versus lowest tertile, 0.69 (0.52-0.92); P trend = 0.01; data presented in text only]. The sweets and snacks pattern was positively associated with abdominal adiposity [OR and 95% CI: 2.05 (1.34-3.14); P trend = 0.03] and pre-diabetes [IFG≥100 mg/dl; OR and 95% CI for highest versus lowest tertile, 1.20 (0.94-1.53); P trend = 0.16; data presented in text only], but not associated with diabetes (IFG ≥ 126) or hypertension. In Mumbai, the fruit and vegetables pattern was inversely associated with hypertension, while the snack and meat pattern appeared to be positively associated with abdominal adiposity [OR and 95% CI: 1.61 (0.96-2.71); P trend = 0.09].

In the largest study region, Trivandrum, we conducted a sensitivity analysis (data presented in text only) comparing dietary patterns associations in participants with normal blood glucose (<100 mg/dl; n = 661; referent group) to each of the following: prediabetics (100-125 mg/dl; n = 407), untreated diabetics (≥126 mg/dl; n = 232), and treated diabetics (valid prescription medication for managing blood sugar; n = 252). The pulses and rice pattern was most strongly inversely associated with treated diabetes [multivariate-adjusted OR and 95% CI, highest to lowest tertile: 0.54 (0.33-0.89); P trend = 0.02]. Conversely, the snack and meat pattern was positively associated with treated diabetes [(1.00 (ref), 1.69 (1.11-2.56), 1.76 (1.12-2.75); P trend = 0.02].

In each of the study regions two dietary patterns emerged and varied associations with cardio-metabolic risk factors were observed. In cross-sectional analyses adjusted for key demographic and lifestyle confounders, diets across all regions characterized by dairy, fried snacks, and sweets appeared to be positively asssociated with abdominal adiposity. Conversly, dietary patterns in Trivandrum and Mumbai, characterized by intake of vegetables and pulses, were inversely related to diabetes and hypertension.

South Asians in India, and throughout the world, are an important population to study due to their greatly elevated risk of diabetes and cardiovascular disease [³⁸,³⁹]. However, compared to the breadth of studies on Western-style dietary patterns and chronic disease etiology [¹⁴,⁴⁰], few studies have examined food patterns in the high-risk Indian population [¹³,¹⁶,¹⁸,^41-43] or collected biological samples. Compared to Caucasian populations, South Asians typically develop metabolic syndromes at lower BMIs, and are known to have increased visceral fat and insulin resistance [³,⁴⁴]. Abdominal adiposity, impaired glucose and lipid levels, as well as hypertension in a high-proportion of study participants corresponds with the high-risk "Asian Indian phenotype" that may be a product of genetic adaptations to food insecurity, fetal or early childhood malnutrition, as well as more recent environmental exposures including adult diet [²,¹⁸,^44-49].

High fat intake (~40% of total energy intake), particularly in Delhi and Mumbai, was one indication that components of the Indian diet may be contributing to a high-risk diet and health profile. However, we found little evidence of Westernization of the diet [⁵⁰] through food items, such as red meat, sweetened beverages, and processed or fast foods, but did observe other characteristics such as sugary and high-fat foods perhaps reflecting a transitional diet arising from access to cheap oils and sweeteners [⁸]. Top loading food items for the dietary patterns included fruit, vegetables, chutneys, and tea, as well as traditional Indian fried snacks and desserts. Across all regions, dietary patterns frequently contained traditional mixed dishes composed mainly of vegetables, pulses, cereals and/or potato, but there were regional differences in the types of cereals (e.g., fermented rice, plain rice, wheat products) and potential protein sources such as pulses, dairy, and eggs. A study conducted in the 1970's among Indian physicians [⁵¹,⁵²], reported a very low-fat, rice and pulse-based diet in the South. This contrasted with a high-fat, wheat-based diet in the North with frequent consumption of vegetables cooked in ghee, milk, and yogurt. More recently, dietary data collected in India from women in the National Family Health Survey (NFHS-2)[¹⁷] also found that intake of animal foods (eggs, dairy, fish, and meat) varied across the different regions [¹⁶]. Compared to China and other Asian countries, high amounts of sugar are consumed in India [⁹]. Our analysis suggests a continued preference for the traditional Indian sweets, as opposed to Western desserts (cake, pies, candy, etc.) which may be more easily recognized as unhealthy [⁵³]. In addition to sugar, many of these Indian sweets are often prepared with a substantial amount of saturated fat from ghee or coconut components.

Previous studies have reported that cardiovascular disease risk in India is likely to be inversely associated with intake of fruits, vegetables, and mustard oil [¹²,⁴³]; and positively associated with intake of refined carbohydrates and unhealthy fats (reviewed in [¹⁸,⁵⁴]). In Trivandrum we found that the more traditional pulses and rice pattern was inversely associated with diabetes. Similarly, the more prudent-appearing fruit and vegetable pattern in Mumbai was inversely associated with hypertension. Secondary patterns in both of these regions, characterized by intake of fried snacks and sweets, appeared to be positively associated with abdominal adiposity. In Delhi, the northern-most region, where dietary patterns appeared to reflect access to a greater variety of foods, the predominant fruit and dairy pattern was positively associated with both abdominal adiposity and hypertension. Although one may not typically consider fruit part of a high-risk diet, higher concordance with this pattern was more common among participants of higher socio-economic status and lower physical activity and may also reflect dietary choices or substitutions to manage a chronic condition. Unlike Trivandrum and Mumbai, the vegetables and pulses pattern in Delhi, was not associated with any chronic conditions and we found no associations for dietary patterns and dyslipidemia.

The relationship between diet and chronic disease risk is certainly complex and we are likely to encounter unfamiliar challenges in the Indian diet. The type, as well as the cooking, of vegetables (green leafy, starchy, stir-fried, stewed, boiled, etc.) in traditional Indian mixed dishes may alter some of their preventive properties [¹²,^55-57] and may also contribute substantially to added fat [⁵¹,⁵⁸]. The contrasting correlations we observed between intakes of iron, calcium, and retinol and the more traditional patterns in Trivandrum versus Mumbai, suggest that the nutrient quality of these regional diets vary considerably. Other cross-sectional studies in South Asian populations, living in India and abroad, have suggested that characteristically high fruit and vegetable intake may be associated with lower LDL and total cholesterol [⁴³], but that high-carbohydrate diets, overall, may be associated with higher triglycerides and lower HDL cholesterol [⁵⁹,⁶⁰], as well as hyperinsulinemia [⁶¹] in South Asians. Unraveling the nutrient composition of specific Indian foods [⁴¹], particularly the fatty acid, sodium, and glycemic composition [⁵⁸], as well as the complex relationship between diet, socio-economic status, obesity, and chronic disease risk [⁶²] is likely to be of great relevance for India.

Major strengths of our study were the use of interviewer-administered questionnaires developed specifically for the study, collection of fasting blood, and measurement of anthropometry and blood pressure by trained medical staff. Fasting blood and blood pressure measurements, along with medication use, were invaluable in classifying diabetes and hypertension status, as self-reports were clearly an underestimate of the prevalence in this population. However, the external validity of our findings may be limited in this lower to upper middle class sample with very high rates of abdominal adiposity and obesity. As with many developing countries, a moderate improvement in socio-economic status is likely to increase access to both over-nutrition and a sedentary lifestyle; thus, affluence may serve as a key risk factor for obesity, diabetes, and other related conditions [⁶³]. Other large cross-sectional studies have observed a double burden of under- and over-weight [¹⁷,^64-66] and throughout India, nutritional status and chronic disease risk profiles vary substantially by intra-and inter-state socio-economic extremes [^67-69]. The prevalence of cardio-metabolic risk factors we observed more closely resembled studies of middle-aged adults residing in metropolitan areas of India [⁷⁰,⁷¹] and that of South Asians residing in the U.S. [⁷²]. However, it also plausible that recruitment of IHS participants from households may have resulted in a clustering of risk factors [⁷³].

We recognize the limitations of the cross-sectional study design to make strong conclusions regarding causality and chronic disease etiology. Dietary patterns identified by factor analysis are intended to represent the actual eating patterns that arose from the study sample, but do not necessarily capture optimal diets or unhealthy extremes that one may want to specifically target for intervention or prevention. Factor analysis also involves some level of subjectivity in selecting and grouping food items. However, in our case, a trained nutritionist in India grouped all the food items collected in the DH into meaningful subgroups prior to the patterns analysis. In addition to the number and composition of the food items, various statistical options and methods for factor analysis, as well as the selection of the number of factors to retain in the final analysis, may also affect the overall explained factor variance [⁷⁴]. Although we used standard methods and a priori grouped food items, the detailed DH and large number of food items collected [⁷⁵] may have limited variation explained by the dietary patterns (8-12% within each region), as well as overall interpretability. Although our variance estimates exceeded those from analyses in similarly underserved populations [⁷⁶], some larger U.S. cohort studies achieved total explained variance levels as high as 20 to 30% [³⁵,⁷⁷]. Dietary patterns analysis conducted in some Western cohorts may also have benefitted from a larger sample size, as well as a more cohesive population with regard to ethnicity, education, and economic access to a variety of foods to meet nutritional needs and preferences.

Participants in all three regions primarily consumed food items that did not appear largely "unhealthy" by Western standards. More traditional dietary patterns in the southern regions, characterized by intake of vegetables and pulses, were inversely associated with the prevalence of diabetes and hypertension. However, overall anthropometric and biological measurements spanning across the majority of the study population characterized high-risk levels not usually seen in studies within the U.S. and Europe. Compelling associations between regional Indian diets and highly prevalent cardio-metabolic risk factors, such as abdominal adiposity and hypertension, suggest that many of the unique regional components of Indian diets, such as high-fat dairy, sweets and fried snacks, may be characterizing a high-risk diet. Continued investigation of diet, evaluating both quality and quantity, as well as other environmental and biological factors, will be needed to better understand the risk profile in this population and potential means of prevention.

BMI: body mass index; CI: confidence interval; DH: diet history; FFQ: Food Frequency Questionnaire; IHS: India Health Study; HDL: high density lipoprotein; IFG: impaired fasting glucose; LDL: low density lipoprotein; MET: metabolic equivalents of task; OR: odds ratio; TC: total cholesterol; TG: triglycerides; WHR: waist-to-hip ratio.

The authors declare that they have no competing interests.

The authors' responsibilities were as follows--CRD conducted the statistical analysis, interpretation of results, and drafting of the manuscript; DP, BIG, LMF, and RS contributed to the interpretation of results and drafting of the manuscript; KK served as the study nutritionist and with RS developed the diet assessment; BIG provided statistical and methodological support; ND and PSG recruited study participants and acquired data; LR conducted laboratory analysis; HS and AM provided site-specific dietary assessment support; YS provided data management and study coordination; RS, DP, AM, PCG, NC, TR, KR conceived of the study, and participated in its design and coordination. All authors read and approved the final manuscript.

This research was supported by the Intramural Research Program of the NIH, National Cancer Institute. We are indebted to the participants of the IHS as well as Sriram Hariharan, Puneet Chadda, Mary McAdams, Eric Berger, and Sujata Dixit-Joshi.