Seventh-day Adventists choose to eat primarily a lacto-ovo-vegetarian diet, although a range of red meat and poultry intake is seen. Snowden and Phillips investigated whether the intake of meat in this group increased the risk of diabetes142. They studied self-reported, cross-sectional prevalence and 21-year mortality using linked death certificates in 25 698 California Adventists.
Among males, there was a significant excess mortality from diabetes among red meat eaters (RR = 1.9) (1.2-3.1) adjusted for BMI, age, physical activity and other variables, but this was not seen in women (RR = 1.1) (CI 0.8-1.6). Meat consumption was also more common among persons with prevalent diabetes at baseline. Given the inaccuracies of death certificates, and only a single measure of diet 21 years prior to follow-up, it is surprising that these results, including a dose-response for increased meat consumption, were seen. It is likely that the dietary patterns were consistent and existed over long periods, so that a single measure gave a reasonable index of habits.
No other studies of diet have explored this issue of meat consumption (or conversely, the complex dietary differences between lacto-ovo-vegetarians and non-vegetarians) given the small numbers of such individuals in the larger cohorts studied by others. This study also provides no useful information about specific dietary constituents, since the diet questionnaire was relatively simple given the large number of subjects seen in 1960.
Lundgren completed a 12-year follow-up of 1242 Swedish women and explored baseline dietary differences in a nested case-control analysis143. No association was seen with total calories, dietary protein, carbohydrate or fat intake, though analyses were largely univariate in nature. Fiber was not investigated in detail, though there was no association with bread, fruit or potato consumption. In a companion paper reporting the role of obesity-related variables, he found that incidence of diabetes was strongly related to baseline BMI, sum of skin-folds and waist-hip ratio144. In addition, increases in BMI and skin-folds were also highly related to increases of serum glucose in normoglycemic women.
Consistent with the findings of Medalie140, Feskens et al. reported no associations between dietary factors and clinical diabetes incidence in the 25-year follow-up of 841 middle-aged men in the Zutphen Study, which was the Netherlands portion of the Seven Countries Study145. This analysis used men enrolled from 1960 to 1985 with clinical diabetes as the end-point. In a later publication which used a portion of this same cohort, Feskens and colleagues analyzed 20-year follow-up data from the same Dutch cohort and added the Finnish cohorts of the Seven Countries Study146. In this later analysis, an OGTT was conducted at the 30-year follow-up visit, but diet data were used from the 1969-1970 visit, so only 20 years of follow-up were available. The authors now excluded people with clinical diabetes and used 2 h glucose among people not known previously to have diabetes as the primary end-point. This included 8% with newly diagnosed diabetes and 21% with IGT at follow-up.
Thorough multiple regression analysis revealed that prior intake of fat (especially saturated fat) was significantly associated with 2 h glucose, adjusted for BMI and age. There was a trend toward higher total and saturated fat intake (as percentage of energy) among people with newly diagnosed diabetes, but this was not statistically significant. They also noted an inverse association with 2 h glucose and vitamin C intake, also with increases in fish, potatoes, vegetables and legume intake between the surveys.
This report shows the differences that may result from different end-point definition within the same study. Significant associations with dietary fat were seen with 2 h glucose levels, but no association was seen using previously undiagnosed diabetes on the OGTT, which may have been due to smaller numbers of categorical end-points. This study has the advantage of very long-term follow-up, but suffers from limited numbers of dietary assessments and relatively small numbers of outcomes. It is also difficult to compare to the previously reported negative 25 year follow-up study using clinical diabetes as an end-point145. These two reports highlight the difficulties of these types of dietary studies. Here, the same dietary methods were used in all subjects, but the length of follow-up, proportion of the original cohort surviving, and the outcome all differed, as did the results between the reports.
Positive findings were also reported by Feskens and colleagues147,148 in two publications from a general practice study of elderly people in Rotterdam, the Netherlands. Among people aged 64-87 years at baseline who were followed for four years with annual OGTTs, higher intake of carbohydrates and pastries and lower intake of legumes high in water-soluble fiber were associated with increased risk of glucose intolerance (IGT plus diabetes)147. In a parallel analysis, they identified a lower incidence of glucose intolerance with increased fish intake (OR = 0.47) (CI 0.23-0.93) adjusted for age, gender, past MI, BMI, energy intake/kg body weight and carbohydrate intake148. No relationship was seen with total or saturated fat intake. They noted that increased habitual fish intake would increase omega-3 fatty acids, which have been shown in animal studies to improve insulin resistance.
The Zutphen studies and these general practice studies used identical measures of dietary intake (the cross-check method). However, the length of follow-up and definition of outcomes differed substantially. Whether the use of OGTT defined glucose intolerance (including IGT)146 compared with clinical diabetes145 is primarily responsible for the different findings is not clear.
Stern explored dietary intake and behavior in the San Antonio Heart Study5. He found no relationship of any carbohydrate intake variable (total, starch, sucrose) after eight years of follow-up among 1093 Mexican-American and non-Hispanic whites, using a repeat OGTT or hypoglycemic therapy to define an incident case of diabetes. Total calories per kilogram of body weight showed an inverse association, such that people who remained non-diabetic reported higher calorie intake per unit of body weight.
Stern and colleagues noted that this seemingly counterintuitive finding could be due to under-reporting by obese people who were at higher risk, had lower physical activity or a more efficient metabolic state, such that people at risk actually ate fewer calories but were more likely to store them as fat. They found no significant associations with total or saturated fat intake as percentage of calories, although people who developed diabetes did report higher fat intake. People who remained free of diabetes were significantly more likely to report avoiding fat intake, using a fat avoidance scale developed locally. These results provide modest support for the role of dietary fat. However, they are limited by the lack of interim dietary measures and the use of a single test for diabetes at eight years.
In a two-year follow-up of 134 Hispanic and non-Hispanic white people with IGT, Marshall and colleagues explored dietary relationships with the development of diabetes diagnosed by OGTT149. An estimated increase of 40 g/day in total dietary fat was associated with an increase in diabetes (OR = 7.4) (1.3- 40.6), adjusted for energy intake, gender, age, BMI, waist-hip ratio, centrality index, and fasting and 1 h insulin fasting glucose at baseline. These are some of the strongest results supporting the role of dietary fat in a prospective study.
A recent analysis of longitudinal data from the San Luis Valley Diabetes Study was also reported by Marshall et a/.15a There were 1069 non-diabetic subjects who were seen for a minimum of one and up to three visits over nine years (average 4.3 years). Complex longitudinal analyses were used which allowed multiple dietary intake data and fasting insulin levels to be examined allowing for both intra- and inter-individual variability over time. This technique, based on the Laird-Ware longitudinal models151, improves power substantially with repeated continuous measures152.
Marshall found that higher intake of total and saturated fat were strongly associated with higher fasting insulin levels, adjusted for BMI, waist circumference, physical activity, age, gender and ethnicity. Dietary fiber and starch intake were inversely associated with fasting insulin concentrations. No associations with fasting insulin concentrations were observed for monounsatur-ated fat, polyunsaturated fat, omega-3 fatty acids, sucrose, glucose and fructose intake. These results support animal studies and a limited number of human population studies that have suggested that increased saturated and total fat intake and decreased fiber and starch intake increase fasting insulin concentrations and may also increase insulin resistance.
Leonetti and colleagues studied 124 Japanese-American men in Seattle for five years123. They explored both nutrient content of the diet and reported energy expenditure. Among people with IGT, the percentage of calories from total dietary fat and animal fat were associated with the development of diabetes (higher fat) or reversion to normal glucose tolerance (lower fat). There was no overall effect of energy expenditure. A small effect was only seen in subgroups with a positive family history of diabetes or with IGT at baseline. Exploration of the interaction of diet and physical activity revealed that the percentage of calories from animal fat was also related to 2 h glucose levels after multiple adjustment, but only in those with less than the highest levels of energy expenditure. There was no effect of refined or complex carbohydrates. The relatively small numbers of subjects made it impossible to detect changes in categories of glucose tolerance (e.g. development of diabetes diagnosed by OGTT) with any power. The careful analysis of interactions which are biologically plausible (e.g. dietary fat effect strongest in sedentary people) provides a useful example for others to emulate in larger studies.
Hara and colleagues followed Japanese-Americans from the mid-1970s through to the 1980s for an average of 6.3 years in Los Angeles or Hawaii114. Subjects were ethnically pure Japanese emigrants from Hiroshima, Japan with an average age of 61 years. While the primary focus of the analysis was on physical activity (see above) and biochemical predictors, the authors note in the discussion that they found no associations between any dietary intake variable and the incidence of diabetes, although, in the abstract of the report, they suggest dietary fat intake is important. Detailed dietary intake data were available153 so it is unclear whether or not this represents a thorough analysis of these data.
Several analyses of the large health professionals cohorts in the USA (Nurses Health Study154-157, Health Professions Follow-up Study158) have explored dietary factors and risk of type 2 diabetes prospectively. Among 84 360 nurses followed for six years (1980-1986) for the development of clinically diagnosed diabetes (no OGTT was conducted) no excess risk was seen for high intakes of animal fat. Lower diabetes risk with higher intake of vegetable fat was seen154.
In the second report from the Nurses Health Study, women who had not yet developed diabetes and who completed a more detailed diet questionnaire in 1986 were analyzed155. Follow-up every two years for a subsequent six years (1986-1992) occurred on the remaining 65173 eligible women. Women with diabetes in the first analysis were excluded. Again, no relationship was seen for higher intake of animal or saturated fat, and the trend for an inverse association with vegetable fat seen in the first study was present, but was not significant. Foods with a high glycemic index, and low cereal fiber intake were found to independently increase diabetes risk. There was evidence of interaction; that is, diets lowest in cereals and highest in glycemic index carried the highest risk (RR = 2.5). Foods with a high glycemic index increase insulin demand and hyperinsulinemia, which may lead to a high demand for endogenous insulin production, stressing pancreatic reserve.
Similar findings were also observed among US male dentists, veterinarians, pharmacists, optometrists, osteopaths and podiatrists158, although the trend for high glycemic foods was less clear until simultaneous adjustment for cereal fiber intake occurred. The primary effect in this analysis appeared to be restricted to men in the category with both the highest glycemic index and lowest cereal intakes, and little dose-response existed for men in groups with intermediate intake.
In a follow-up analysis of the Nurses Health Study, Liu et al. explored whole-grain intake on risk of type 2 diabetes in women156. In 10-year follow-up data, women who consumed wholegrain products (e.g. dark bread, wholegrain breakfast cereal, cooked oatmeal, popcorn) were 27% less likely to develop clinical diabetes (RR = 0.73, CI 0.63-0.89 adjusted for multiple factors) and 31% more likely to develop diabetes with a higher intake of refined grain products (RR = 1.31, CI 1.12-1.53). Analysis exploring potential interactions with BMI, saturated fat intake and other variables did not significantly alter these relationships.
Among nurses154,155 and other health professionals158, lower magnesium intake was also noted to increase risk of diabetes in a dose-response fashion, with the relative risk steadily dropping as intake increased to the highest quartile (nurses RR = 0.62, CI 0.50-0.78)155; male health professionals RR = 0.72, CI 0.54-0.96)158. Among nurses, lower intakes of calcium and potassium were also associated with increased diabetes risk154, though this was not noted in the later analysis155. Similar findings for magnesium have also been reported from the Iowa Women's Health Study159.
The most recent analysis of the Nurses Health Study cohort explored specific types of dietary fat on diabetes incidence157. Using the 14-year follow-up data on 84204 women aged 34-59 years at baseline, Salmerón et al. again reported no effect of increased total dietary fat (RR = 0.98, CI 0.94-1.02 for a 5% increase in intake). Saturated and monounsaturated fatty acids were not found to be related, however, for every 5% increase in energy intake from polyunsaturated fatty acids, there was a reduction in diabetes risk (RR = 0.63, CI 0.53-0.76), and for a 2% increase in trans-fatty acids, there was excess risk (RR = 1.39, CI 1.15-1.67). An accompanying editorial notes that this novel hypothesis and finding are substantially limited by the methods used, and should be investigated in more detailed studies160.
Meyer et al. have also conducted detailed dietary analyses on the Iowa Women's Health Study cohort161. After 11 years of follow-up with four interim diet assessments similar to those used in the Nurses Health Study, 1890 self-reported cases of diabetes developed among the 35 988 women aged 55-69 years at baseline. No relationship was seen with total dietary fat. Incidence was negatively associated with dietary polyunsaturated fatty acids, vegetable fat and trans-fatty acids, whereas trans-fatty acids were positively associated with diabetes risk in the Nurses Health Study157. With detailed adjustment for magnesium and cereal fiber intake, both significantly related to diabetes incidence in this cohort159, as well as subsitution of other dietary fat components, the multiply-adjusted relative risks declined across quintiles of vegetable fat intake from 1.00, 0.90, 0.87, 0.84 and 0.82 (p = 0.02), although polyunsaturated fatty acids and trans-fatty acids no longer remained significant. Subsitution of polyunsaturated fatty acids for saturated fatty acids, and vegetable fat for animal fat, gave similar results. No effect modification by BMI, activity, alcohol or vitamin E intake was seen. Given the very high correlation with dietary fat intake components, some degree of attenuation between them is likely, making identification of specific fat subtypes difficult.
Ford and Mokdad reported on fruit and vegetable consumption in the 20-year follow-up of the NHANES-I Epidemiologic Follow-up Study (NHFES) cohort162 using clinically reported diabetes, hospitalizations or death certificates as end-points. No OGTTs were conducted. Persons eating five or more servings of fruits and vegetables per day, adjusted for demographics, smoking, medications, activity, BMI and alcohol use, were 27% less likely to develop incident diabetes (HR = 0.73, CI 0.54-0.98; ARR ^ 1.9/1000 person-years). This effect was primarily seen in women, and was attenuated when education was added to the model, but was not changed when percentage of calories from fat, total energy intake and use of vitamins were added. Given the details available about diet, incomplete adjustment for type of fat intake and levels of activity is likely, and no measures of glycemic load were available. Thus, it is not clear if fruit and vegetable intake is itself associated with lower incidence of diabetes, or whether it marks healthy behaviors in general. It appears consistent with results of Feskens146.
The results of the several prospective studies presented in Table 6.5 present a somewhat inconsistent picture. This may be due in part to the different type and quality of nutrient intake data, which included 24-hour recalls, food frequencies of varying detail, and simple scales. Each of these allowed different levels of nutrient aggregation, some with more detailed nutrient databases underlying them than in other studies. Outcomes also varied, from use of 2 h glucose levels or insulin levels, to clinically diagnosed diabetes, to diabetes detected by OGTT, and including mortality. Each of these end-points may identify different portions of the phenotypic and etiological spectrum and may at least partially explain divergent results. In addition, the amount of interim data between baseline and follow-up varied, as did the analytic complexity, with earlier studies often being largely univariate, while later studies used time-to-event methods or longitudinal models. It is likely that each of these differences accounts for some of the between-study variability, and highlights the fact that there are very few studies with methods in common.
Findings for total energy intake varied, and dietary fiber had an inconsistent association with diabetes as an outcome, though higher intakes were associated with lower levels of glucose146 or insulin150. The only prospective studies that identified a role for higher levels of dietary fat used 2 h glucose as an outcome123,146, or fasting insulin150, or OGTT diabetes among people with IGT149. The study by Stern5 also found that persons 'avoiding fat' (on a scale no one else has replicated) had lower risk of developing clinical or OGTT diabetes.
The largest prospective studies of clinical diabetes among non-diabetics at baseline did not find a role for total dietary fat154,155,157,158,161. Consistently, these studies differed from those that were positive by not having any glucose tolerance testing either at baseline or at follow-up. It is unclear whether this could explain the reported differences. In addition, few of the studies explored composition interaction (e.g. dietary fat intake in persons with low versus high fiber intake) with the exception of the Health Professionals studies that reported glycemic index-fiber, and wholegrain-dietary fat interactions. Whether such analyses would clarify the picture remains unclear.
At present, the evidence still does not support a consistently replicated role for specific nutrients in diabetes incidence, though dietary fat subtypes and a role for cereal fiber and wholegrains is accumulating. A diet that combines limited energy intake for weight maintenance, low saturated fat content, high fruit and vegetable and wholegrain intake appears prudent for the reduction of cardiovascular disease, some cancers and diabetes. Use of such a diet in both the DPS69 and the DPP72 was one of the lifestyle changes that resulted in reduction of diabetes incidence.
Was this article helpful?