Low-Fat Dietary Pattern and Risk of Treated Diabetes Mellitus in Postmenopausal Women: The Women's Health Initiative Randomized Controlled Dietary Modification Trial

Background  Decreased fat intake with weight loss and increased exercise may reduce the risk of diabetes mellitus in persons with impaired glucose tolerance. This study was undertaken to assess the effects of a low-fat dietary pattern on incidence of treated diabetes among generally healthy postmenopausal women.

Methods  A randomized controlled trial was conducted at 40 US clinical centers from 1993 to 2005, including 48 835 postmenopausal women aged 50 to 79 years. Women were randomly assigned to a usual-diet comparison group (n = 29 294 [60.0%]) or an intervention group with a 20% low-fat dietary pattern with increased vegetables, fruits, and grains (n = 19 541 [40.0%]). Self-reported incident diabetes treated with oral agents or insulin was assessed.

Results  Incident treated diabetes was reported by 1303 intervention participants (7.1%) and 2039 comparison participants (7.4%) (hazard ratio, 0.96; 95% confidence interval, 0.90-1.03; P = .25). Weight loss occurred in the intervention group, with a difference between intervention and comparison groups of 1.9 kg after 7.5 years (P < .001). Subgroup analysis suggested that greater decreases in percentage of energy from total fat reduced diabetes risk (P for trend = .04), which was not statistically significant after adjusting for weight loss.

Conclusions  A low-fat dietary pattern among generally healthy postmenopausal women showed no evidence of reducing diabetes risk after 8.1 years. Trends toward reduced incidence were greater with greater decreases in total fat intake and weight loss. Weight loss, rather than macronutrient composition, may be the dominant predictor of reduced risk of diabetes.

Trial Registration  clinicaltrials.gov Identifier: NCT00000611

The incidence and burden of diabetes continue to rise, particularly in developed countries. The prevalence among US adults 20 years and older is expected to increase to 11.2% by the year 2030.¹ In women, the prevalence of diabetes rises sharply with each decade of age after menopause and surpasses that in men after the seventh decade of life.^2,3 Women with diabetes who are 50 years and older live fewer years than women without diabetes,⁴ and comorbidity with cardiovascular disease is common.^4,5 A further toll is extracted in health care costs⁶ and decreased work productivity.⁷

Lifestyle interventions of weight loss and increased physical activity coupled with a low-fat dietary pattern have been shown to reduce the risk of diabetes mellitus in persons with impaired glucose tolerance.^8-11 When investigated alone, an ad libitum low-fat dietary pattern was found to improve the glucose profile among persons with impaired glucose tolerance; however, the intervention group also lost weight and adopted a higher physical activity level than did the control group.¹² Even without changes in physical activity, adoption of an ad libitum low-fat dietary pattern often results in decreased energy intake,^13-16 which may be beneficial in the prevention of diabetes. The low-fat diet results are encouraging because lowering the proportion of energy from fat typically increases the proportion of energy from carbohydrate and the glycemic load, which may increase the risk of developing diabetes.^17-19

The Women's Health Initiative (WHI) Dietary Modification Trial (DMT) was a randomized controlled trial designed to measure the effect of an ad libitum low-fat diet on the incidence of diabetes. The objectives of this report are to address the hypothesis that an ad libitum low-fat dietary pattern alone, without goals of exercise or weight loss, reduces the incidence of diabetes in generally healthy postmenopausal women.

The design and baseline description of the WHI DMT and recruitment methods have been published.^20-22 Briefly, 48 835 generally healthy postmenopausal women aged 50 to 79 years were recruited and enrolled at 40 clinical centers across the United States. Major exclusion criteria included having a history of breast cancer or colorectal cancer, any cancer within the previous 10 years except nonmelanoma skin cancer, medical conditions with a predicted survival of less than 3 years, adherence or retention concerns, current dietary intake of less than 32% of energy from fat, and type 1 diabetes mellitus. Mean follow-up time for the WHI DMT was 8.1 years, and the trial progressed to its planned conclusion date of March 31, 2005. Approximately 14% of participants in the WHI DMT also participated in 1 of the WHI hormone trials,^23,24 and nearly 52% participated in the WHI calcium and vitamin D trial.²⁵ The WHI protocol and consent forms were approved by the institutional review board for each participating institution and the clinical coordinating center. All participants provided informed consent.

A randomized permuted block algorithm, stratified by clinical center site and age (50-54, 55-59, 60-69, or 70-79 years) was used, with participants randomized to either a low-fat dietary pattern intervention group (40.0%) or to a usual-diet comparison group (60.0%).²⁶ The WHI DMT dietary intervention group received intensive nutritional and behavioral modification training consisting of 18 group sessions in the first year followed by quarterly sessions throughout the trial. Each participant received an individualized dietary fat gram goal estimating 20% of energy from fat during the intervention and a common dietary goal of 5 or more servings daily of combined vegetables and fruits and 6 or more servings daily of grains. Self-monitoring techniques and group session attendance were emphasized. The diet was not intended to promote reduced energy intake, and there was neither a weight loss goal nor a physical activity goal. Participants in the usual-diet comparison group were given a copy of the Dietary Guidelines for Americans^27,28 but were not asked to make dietary changes. Details of the intervention program have been published.²⁹

Potential participants for the WHI DMT completed 3 prerandomization screening visits during which baseline information was obtained.²⁶ Follow-up for clinical events was scheduled to occur every 6 months. Participants were instructed to bring all current prescription and nonprescription medications in their original containers, and medication was documented in a pharmacy database (Master Drug Database; Medi-Span, Indianapolis, Indiana).

Dietary intake was monitored primarily by a food frequency questionnaire (FFQ) designed for the WHI.³⁰ The FFQ was administered to all DMT participants during screening (considered the baseline measure), 1 year after randomization, and thereafter annually to one-third of the participants on a rotating basis. The methods for assigning the glycemic index and glycemic load values used in the WHI FFQ have been reported elsewhere.³¹

All participants had blood drawn after an overnight 8-hour fast at baseline and at the first annual visit.²⁶ A total of 2816 (5.8%) of the DMT participants had serum analyzed at baseline and years 1, 3, and 6 for diet-related biomarkers including fasting glucose and insulin levels. Insulin resistance was calculated from fasting glucose and insulin according to the homeostasis model assessment, insulin resistance (HOMA-IR) model: insulin/{22.5 × exp[−ln(glucose/18)]}, where the units for insulin are microunits per milliliter and the units for glucose are millimoles per liter.³²

Diabetes at baseline was documented by self-report by asking each woman if she had ever been told by a physician that she had “sugar diabetes” when not pregnant. Participants responding positively were considered to have prevalent diabetes and were removed from the analyses. During the semiannual follow-up contacts, documentation was limited to treated diabetes when participants were asked, “Since the date given on the front of this form, has a doctor prescribed any of the following pills or treatments?” Choices included “pills for diabetes” and “insulin shots for diabetes.” It may be assumed that participants were regularly seen by a health care provider because 92.9% reported at baseline having a regular health care provider, and 79.1% of those participants reported having seen their provider within the past year. During the average 8.1 years of follow-up, participants reported having had an average of 8.5 (comparison group) or 8.4 (intervention group) physical examinations. Fasting blood glucose levels were not used to identify diabetes because only 5.8% of WHI DMT participants had blood samples analyzed. Self-report of treated diabetes among the postmenopausal women of the WHI DMT was likely type 2 diabetes mellitus, although diagnosis was not confirmed by medical record audit. Data from a WHI diabetes confirmation study showed that, in the WHI DMT, 80% of participants with prevalent self-reported treated diabetes brought an oral agent or insulin to the baseline drug inventory, and 77% with incident self-reported treated diabetes during the first year of the trial brought a diabetes medication to the year 1 drug inventory. Furthermore, among the 525 WHI participants who had a fasting blood glucose measurement at baseline and self-reported ever having been told by a physician they had “sugar diabetes,” 73% of the WHI DMT participants had a fasting glucose level greater than 125 mg/dL (to convert glucose to millimoles per liter, multiply by 0.0555).³³

Descriptive statistics for baseline demographic and health information, laboratory measures related to diabetes, and dietary intakes were computed and compared by randomization assignment. Means and standard deviations were computed for dietary FFQ data and diabetes risk factor data at baseline, year 1, and year 6. The 6-year time point for dietary data included years 5, 6, and 7 to provide a sampling from all active WHI DMT participants. Differences between intervention and comparison for baseline, year 1, and year 6 were computed and tested for significance at P < .001 by means of a 2-sample t test or appropriate nonparametric techniques. Multiple comparisons were tested by Bonferroni t tests or similar tests. Bivariate comparisons of categorical variables were performed with the χ² test or Fisher exact test as appropriate.

Primary outcome comparisons of treated diabetes were analyzed as time-to-event hazard ratios and 95% confidence intervals from Cox proportional hazards analyses. Outcomes were those available from the September 2005 database. Significance was analyzed by log-rank tests. Analyses were stratified by, and thus adjusted for, the following: (1) randomization assignment, (2) age at baseline (50-59, 60-69, or 70-79 years), (3) randomization to the Hormone Therapy Trial (not randomized, randomized to active estrogen alone or placebo for women without a uterus, or randomized to active estrogen plus progestin or placebo for women with a uterus). A Kaplan-Meier curve was drawn to depict cumulative hazard estimates over time. Annualized rates of incident diabetes were assessed for absolute disease rate comparisons.

A sensitivity (adherence) analysis was done by inverse probability weighting^25,34 for session attendance for DMT intervention participants plus clinic visit and additional factors related to adherence, such as WHI hormone trials randomization, demographics, lifestyle habits, and psychosocial factors, for all participants. Because participants in the DMT intervention group lost a modest amount of weight compared with the DMT comparison group (a difference of 1.9 kg over 7.5 years; P < .001),¹⁶ a proportional hazards model with weight change as a time-dependent covariate was used to explore possible effects of postrandomization weight change on diabetes risk.

Subgroup analyses for interactions of randomization assignment with the risk of developing treated diabetes were assessed for baseline demographics, health history, dietary factors, and diabetes risk factors. Additional subgroup analyses compared the risk of developing diabetes with dietary intakes at year 1 of percentage of energy from fat and number of servings of vegetable, fruit, and grain, which were the 3 DMT dietary intervention goals. Dietary intakes at year 1 were chosen to examine the potential interactions in relation to the year when maximum change in dietary intake had been reported by the intervention group. Quartiles of intakes reported by the DMT intervention group were compared with the overall intake reported by the comparison group.

The year 1 subgroup data were analyzed as 1 unadjusted model and 2 adjusted models. Of the 2 adjusted models, one adjusted for baseline factors of age, race/ethnicity, education, smoking status, alcohol use, physical activity, baseline body mass index, functional and psychosocial factors influencing adherence (physical functioning, mental health, optimism, hostility, negative emotional expressiveness),^35,36 and dietary intake for the pertinent dietary variable. For example, year 1 quartiles of percentage of energy from total fat were adjusted for baseline intake of percentage of energy from total fat. The other model adjusted for the baseline factors listed for the first model plus change in body weight from baseline to year 1. Diabetes cases reported before year 1 were excluded from the analyses to decrease the possibility of undiagnosed diabetes at baseline. Participants with missing data for any of the adjustment factors were excluded from the 3 year 1 subgroup analyses. Hazard ratios and P values for interaction were computed for the subgroup analyses.

Among the 48 835 WHI DMT participants, 45 887 are represented in these analyses (see Figure 1 for participant flow). Excluded were the 2948 participants (6.0%) who self-reported prevalent diabetes at baseline. Blood analyses were available for 2577 of the participants (5.6%). Participants with prevalent diabetes were older, heavier (higher body mass index), and less physically active and had higher blood pressure than participants not reporting diabetes. Baseline dietary intakes for participants with diabetes were higher for total fat, combined vegetables and fruits, grains, and fiber, whereas intake was lower for saturated fat.

On average, there were no significant differences (P < .05) in baseline characteristics between the intervention and comparison groups of participants without prevalent diabetes (Table 1). There was a borderline statistically significant difference in systolic blood pressure of 0.3 mm Hg between groups, which is not considered clinically different. Within tertiles of serum glucose level among the 5.8% subsample analyzed, slightly more of the intervention participants had blood glucose levels greater than 98 mg/dL, although mean glucose level did not differ between the intervention and comparison groups. Participants were 62 years of age on average; most were white, not randomized to the hormone therapy trials, not exposed to exogenous estrogen or estrogen plus progestin, and nonsmokers or past smokers.

As shown in Table 2, at baseline, the mean percentage of energy from total fat was 37.7% (SD, 5.0%). No differences in dietary intake at baseline were seen between the intervention and comparison groups. At year 1, the intervention group reported lower total dietary fat as a percentage of energy and higher intakes of vegetables, fruits, grains, and dietary glycemic index and glycemic load than did the comparison group. The differences between intervention and comparison groups at year 6 were somewhat smaller than those at year 1, but they remained significant, except dietary glycemic index at year 6.

Table 3 shows that, on average, participants at baseline were overweight and had systolic blood pressure in the prehypertensive range, although blood glucose levels were in the normal range with no differences between groups. One year after randomization, weight, body mass index, waist circumference, systolic and diastolic blood pressure, and serum glucose level were lower in the intervention group (P < .001). Six years after randomization, weight and body mass index were lower in the intervention group (P < .001).

Overall, 1303 participants (7.1%; annual incidence, 0.88%) in the WHI DMT intervention group and 2039 participants (7.4%; annual incidence, 0.91%) in the comparison group self-reported a new (incident) diagnosis of treated diabetes during follow-up. The estimated hazard ratio was 0.96 (95% confidence interval, 0.90-1.03; Figure 2), with essentially the same results if only treated diabetes was excluded at baseline (data not shown). The adherence sensitivity analysis, adjusting for participation in the intervention and DMT protocol, resulted in a hazard ratio of 0.93 (95% confidence interval, 0.81-1.07). The time-dependent weight change proportional hazards model resulted in a hazard ratio of 0.96 (95% confidence interval, 0.89-1.02), nearly identical to the primary hazard ratio.

In baseline subgroup analyses, there were no significant interactions between randomization assignment and dietary intakes (Figure 3). A statistically significant interaction was seen among tertiles of serum insulin levels and HOMA-IR, but no one tertile differed significantly from unity (Figure 4).

For the year 1 subgroup analyses, data from an additional 12 767 participants were excluded because of missing data (n = 10 535) or diabetes reported between baseline and year 1 (n = 2232). Adjusted subgroup analyses between year 1 dietary factors and randomization assignment showed a statistically significant effect for percentage of energy from total fat (P = .04). After additional adjustment for weight change between baseline and year 1, an effect of percentage of energy from dietary fat on incidence of treated diabetes was no longer statistically significant (Table 4). On adjustment for baseline factors and weight change, no apparent effect was seen for changes in intake of vegetables, fruits, or grains.

Lifestyle interventions combining a low-fat dietary pattern, weight loss, and exercise have been found to lower risk of developing type 2 diabetes mellitus.^8-11 In the WHI DMT, implemented for an average of 8.1 years among generally healthy postmenopausal women, intervention with a low-fat dietary pattern alone did not result in a statistically significant reduction in treated diabetes incidence compared with participants consuming their usual diet. Neither, however, did the low-fat dietary pattern result in an increased risk of diabetes.

A number of reasons may account for the differences between our findings and those of previous trials. The participants in the Diabetes Prevention Program (DPP),¹⁰ the Finnish Diabetes Prevention Study (DPS),^8,9 and the Da Qing Diabetes IGT [Impaired Glucose Tolerance] and Diabetes Study¹¹ were at higher risk of developing diabetes at baseline when compared with the participants in the WHI DMT. However, most important, the DPS, DPP, and Da Qing studies included physical activity and weight loss as primary objectives, whereas the goals of the WHI DMT were to analyze the effects of changes in diet composition. In the DPP, weight loss was the strongest predictor of diabetes risk reduction, although physical activity, even without weight loss, also predicted risk reduction.³⁷ The effect of a low-fat dietary pattern on diabetes risk reduction in the DPP and Finnish DPS was no longer statistically significant after adjusting for weight loss and physical activity.^37,38

In the WHI DMT, the ad libitum low-fat dietary pattern assigned to the intervention group was intended to be body weight neutral, although participants in the intervention group reported consuming less energy and lost a modest amount of weight (a mean of 2.2 kg [2.8%]) after the first year of the study,¹⁶ which is often experienced with ad libitum low-fat dietary patterns.^13,15,39 After 7 years of follow-up, participants in the WHI DMT intervention group maintained a mean weight loss of 0.8 kg from baseline, whereas there was an average of a 0.1-kg weight loss in the comparison group.¹⁶ By contrast to the WHI DMT, the average weight reductions in the lifestyle intervention groups of the diabetes studies were 5.6 kg in the DPP,¹⁰ 4.2 kg in the DPS,⁸ and 3.3 kg in the Da Qing study.¹¹

Dietary total fat, on its own, has overall shown little evidence of being a risk factor for diabetes. Although research by Marshall and colleagues⁴⁰ suggested that dietary total fat was a risk factor for developing diabetes, much of the prospective research since then has not corroborated such an association.^41-44 The findings from the WHI are consistent with the prospective studies in not showing an effect of dietary total fat alone.

The impact of early dietary changes can be partially examined through subgroup analyses based on dietary change at year 1. While the statistically significant interaction (P = .04) between intervention and comparison groups after adjustment for baseline factors suggested a reduced risk of diabetes with greater fat reduction, the significance disappeared on further adjustment for weight change. This suggests that weight loss accounts for the protective effect. Thus, in a generally healthy population of postmenopausal women, the role of an ad libitum low-fat dietary pattern in diabetes prevention may be as a mediator of weight loss. Subgroup analyses showed a statistically significant interaction among tertiles of baseline serum insulin level and HOMA-IR, suggesting the possibility of differential effects of a low-fat dietary pattern. However, this should be interpreted cautiously because of the number of subgroup analyses and the small number of cases in the lowest tertile of insulin level and HOMA-IR.

Vegetable and fruit intake in the WHI DMT did not reduce the risk of diabetes after adjustment for baseline factors and weight change. These results are consistent with observational research from the Iowa Women's Health Study and the Nurses' Health Study.^45,46 However, among overweight women from the Nurses' Health Study, higher intakes of vegetables and fruits were associated with reduced risk of diabetes. Results from the Finnish Mobile Clinic Health Examination Survey indicated that higher intakes of green vegetables, fruits, and berries were associated with reduced risk of diabetes.⁴⁷ In the US National Health and Nutrition Examination Survey, a protective effect of vegetables and fruits was seen, although the effect was ameliorated when education level was considered.⁴⁸

Some studies have shown that a higher dietary glycemic index and glycemic load may predict elevated risk of diabetes.^17-19 However, other studies have not shown an effect of glycemic index or glycemic load on diabetes incidence.^42,49,50 In the WHI DMT, the glycemic index did not differ to a biologically significant degree between the DMT comparison and intervention groups at baseline, year 1, or year 6. Dietary glycemic load was higher in the intervention group at years 1 and 6 but was not found to be associated with risk of diabetes.

Limitations of our study warrant consideration. In the WHI DMT, incident diabetes was limited to self-report of diabetes treated by insulin or oral agents, although the medication inventory agreement with self-report was high³³ and the number of physical examinations suggests that health was routinely being monitored. Second, dietary intake was captured by self-report and may present a source of bias.⁵¹ Third, the WHI FFQ was designed to be sensitive to sources of dietary fat, changes in fat intake, and vegetables and fruits, not glycemic index or glycemic load. Fourth, in the sensitivity analysis, there was not a method for fully comparing adherence between the 2 groups given that only the intervention group was asked to attend dietary change sessions. Fifth, it is possible that there was an increased awareness of diabetes risk, screening, and hence diagnosis among the intervention group, leading to an underestimation of the association between a low-fat diet and incident diabetes. However, this is unlikely because the intervention and comparison groups reported having received the same number of physical examinations during the follow-up period. Finally, subgroup analyses must be viewed with caution because of the large number of comparisons, the potential for residual confounding, and missing data.

Overall, the WHI DMT ad libitum low-fat dietary intervention was not associated with a reduced or increased risk of treated diabetes in generally healthy postmenopausal women. However, there were suggestions that, among participants with greater reductions in dietary fat, reductions in energy intake resulted in weight loss. Thus, any reduction in diabetes risk appears to be related to weight loss rather than to the macronutrient composition of the diet.

Correspondence: Lesley F. Tinker, PhD, Women's Health Initiative, Fred Hutchinson Cancer Research Center, PO Box 19024, Mail Stop M3-A410, Seattle, WA 98109 (ltinker@whi.org).

Accepted for Publication: December 20, 2007.

Author Contributions: Dr Tinker had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Tinker, Margolis, Manson, and Wenger. Acquisition of data: Bonds, Manson, Howard, Perri, Beresford, Rodríguez, Safford, Stevens, and Parker. Analysis and interpretation of data: Tinker, Bonds, Margolis, Manson, Howard, Larson, Perri, Robinson, Wenger, and Stevens. Drafting of the manuscript: Tinker, Bonds, Larson, and Safford. Critical revision of the manuscript for important intellectual content: Tinker, Bonds, Margolis, Manson, Howard, Larson, Perri, Beresford, Robinson, Rodríguez, Wenger, Stevens, and Parker. Statistical analysis: Larson and Safford. Obtained funding: Manson, Howard, Beresford, Rodríguez, and Stevens. Administrative, technical, and material support: Manson, Perri, Stevens, and Parker. Study supervision: Tinker, Manson, Howard, Perri, Wenger, and Stevens.

Financial Disclosure: Dr Robinson has received grants from Abbott, Andrx Labs, Astra-Zeneca, Atherogenics Inc, Bristol-Myers Squibb, GlaxoSmithKline, Hoffman La Roche, Merck, Pfizer, Procter & Gamble, Sankyo, Schering-Plough, Takeda, and Wyeth Ayerst; speaker honoraria for educational programs from Bristol-Myers Squibb, Merck, and Pfizer; and honoraria from Reliant; and serves on the Consultant/Advisory Board for Bristol-Myers Squibb, Merck, Pfizer, Proliant, Wellmark, and the American Emu Association. Dr Howard is a consultant for Merck and the Egg Nutrition Council; has received research support from Pfizer, Merck, and Schering-Plough; and has lectured for Schering-Plough.

Funding/Support: The WHI program was funded by the National Heart, Lung, and Blood Institute of the National Institutes of Health, US Department of Health and Human Services.

Role of the Sponsor: The funding organization had representation on the steering committee, which governed the design and conduct of the study, the interpretation of the data, and the preparation and approval of the manuscript. The National Heart, Lung, and Blood Institute Program Office reviewed the manuscript before publication.

Independent Statistical Analysis: All statistical analyses for the study were performed by statisticians at the Fred Hutchinson Cancer Research Center, Division of Public Health Sciences, under the direction of the principal investigator of the Coordinating Center, Ross Prentice, PhD.