Organization chart of participant flow through the study. Diet 1, high-carbohydrate (CHO)/high–glycemic index (GI); diet 2, high-CHO/low-GI; diet 3, high-protein/high-GI; and diet 4, high-protein/low-GI. ITT indicates intention-to-treat.
Changes in weight (A) and fat mass (B) over 12 weeks in 129 overweight young adults randomized to 4 diets of varying glycemic load. Diet 1, high-carbohydrate (CHO)/high–glycemic index (GI); diet 2, high-CHO/low-GI; diet 3, high-protein/high-GI; and diet 4, high-protein/low-GI. All values are given as means with error bars indicating standard errors.
Changes in blood lipid fractions from 0 to 12 weeks in 129 overweight young adults randomized to 4 diets of varying glycemic load. Diet 1, high-carbohydrate (CHO)/high–glycemic index (GI); diet 2, high-CHO/low-GI; diet 3, high-protein/high-GI; and diet 4, high-protein/low-GI. Total C indicates total cholesterol; LDL, low-density lipoprotein; HDL, high-density lipoprotein; and TG, triglycerides. All values are given as means with error bars indicating ±SE. To convert cholesterol values to milligrams per deciliter, divide by 0.0259.
Box plot of glycemic load (glycemic index [GI] × carbohydrate [CHO] content) of the 4 diets based on food diaries completed during weeks 4 and 8 of the intervention. Diet 1, high-CHO/high-GI; diet 2, high-CHO/low-GI; diet 3, high-protein/high-GI; and diet 4, high-protein/low-GI.
Mean plasma glucose (A) and insulin (B) profiles over 10 hours (h) in overweight female subjects after consumption of representative mixed meals for each of the 4 diets. Breakfast was eaten at time 0 (0800 h), the morning snack at 180 minutes (1100 h), lunch at 300 minutes (1300 h), and dinner at 540 minutes (1700 h). Diet 1, high-carbohydrate (CHO)/high–glycemic index (GI); diet 2, high-CHO/low-GI; diet 3, high-protein/high-GI; and diet 4, high-protein/low-GI. Eleven subjects took part in the profile day, with each one consuming each of the 4 diets on 4 separate days. AUC indicates area under the curve. All values are given as means with error bars indicating standard errors. To convert glucose values to milligrams per deciliter, divide by 0.0555; to convert insulin values to microunits per milliliter, divide by 6.945.
Customize your JAMA Network experience by selecting one or more topics from the list below.
McMillan-Price J, Petocz P, Atkinson F, et al. Comparison of 4 Diets of Varying Glycemic Load on Weight Loss and Cardiovascular Risk Reduction in Overweight and Obese Young Adults: A Randomized Controlled Trial. Arch Intern Med. 2006;166(14):1466–1475. doi:10.1001/archinte.166.14.1466
Despite the popularity of low–glycemic index (GI) and high-protein diets, to our knowledge no randomized, controlled trials have systematically compared their relative effects on weight loss and cardiovascular risk.
A total of 129 overweight or obese young adults (body mass index, ≥25 [calculated as weight in kilograms divided by the square of height in meters]) were assigned to 1 of 4 reduced-fat, high-fiber diets for 12 weeks. Diets 1 and 2 were high carbohydrate (55% of total energy intake), with high and low GIs, respectively; diets 3 and 4 were high protein (25% of total energy intake), with high and low GIs, respectively. The glycemic load was highest in diet 1 and lowest in diet 4. Changes in weight, body composition, and blood chemistry profile were studied.
While all groups lost a similar mean ± SE percentage of weight (diet 1, −4.2% ± 0.6%; diet 2, −5.5% ± 0.5%; diet 3, −6.2% ± 0.4%; and diet 4, −4.8% ± 0.7%; P = .09), the proportion of subjects in each group who lost 5% or more of body weight varied significantly by diet (diet 1, 31%; diet 2, 56%; diet 3, 66%; and diet 4, 33%; P = .01). Women on diets 2 and 3 lost approximately 80% more fat mass (−4.5 ± 0.5 [mean ± SE] kg and −4.6 ± 0.5 kg) than those on diet 1 (−2.5 ± 0.5 kg; P = .007). Mean ± SE low-density-lipoprotein cholesterol levels declined significantly in the diet 2 group (−6.6 ± 3.9 mg/dL [−0.17 ± 0.10 mmol/L]) but increased in the diet 3 group (+10.0 ± 3.9 mg/dL [+0.26 ± 0.10 mmol/L]; P = .02). Goals for energy distribution were not achieved exactly: both carbohydrate groups ate less fat, and the diet 2 group ate more fiber.
Both high-protein and low-GI regimens increase body fat loss, but cardiovascular risk reduction is optimized by a high-carbohydrate, low-GI diet.
clinicaltrials.gov Identifier: NCT00254215
The quantity and quality of carbohydrates (CHOs) are at the center of debate surrounding the optimal diet for weight loss. High-CHO intake, particularly refined CHOs, can exaggerate postprandial glycemia and has accompanied increases in obesity, type 2 diabetes mellitus, and the metabolic syndrome.1-3 While a low-fat, high-CHO diet remains the “best practice,” several studies suggest that lower-CHO, higher-protein, or low–glycemic load (GL) diets provide benefits for weight loss and cardiovascular risk reduction.4-11 Compared with the standard diet, restriction or modification of CHO intake can result in beneficial effects on energy expenditure,6 triglyceride (TG) concentrations, high-density lipoprotein (HDL) cholesterol levels, and glucose homeostasis.12
A unifying hypothesis is that a high-dietary GL (the contribution to postprandial glycemia of all foods in a diet) increases the difficulty of weight control because rapidly digestible CHOs can cause marked fluctuations in blood glucose and insulin levels, in turn stimulating hunger13 and inhibiting fat oxidation.14 Both low–glycemic index (GI) and high-protein diets have caught the public's attention, but clinicians and health professionals remain skeptical, calling for greater scientific evidence on which to base advice to patients. One concern isthat many sources of protein, such as meat and dairy products, can also be high in saturated fat. The accelerating obesity epidemic, the widespread popularity of alternate diets, and animal studies showing β-cell disruption on high-GI diets15 make the need for resolution of the debate more pressing.
We conducted a 12-week parallel, randomized, controlled trial of 4 weight loss diets of defined GL, varying in CHO and protein content and in GI. All diets aimed for the same fat content (30% of total energy intake [E]), type of fat (saturated or unsaturated), and total dietary fiber. The hypothesis was that diets of reduced GL would increase the rate of fat loss, with minimal loss of lean mass, and improve cardiovascular disease risk factors.
Volunteers were recruited locally using notice boards and newspaper advertisements. Young adults, 18 to 40 years of age, with a body mass index (BMI) of 25 or more (calculated as weight in kilograms divided by the square of height in meters), a body weight of less than 150 kg, and weight fluctuations of less than 5 kg in the previous 2 months, who were willing to eat red meat and maintain current physical activity were included. Exclusion criteria were chronic illness, regular medication other than birth control pills, eating disorders, special diets, pregnancy, food allergy, and insufficient command of the English language. In total, 148 individuals were screened between July 2002 and July 2004, and 129 of them (98 women and 31 men) met the inclusion criteria. Of those excluded, 2 were taking medications, 5 were older than 40 years, 3 were non–red meat eaters, 5 weighed more than 150 kg, and 4 were accepted but withdrew before randomization.
Subjects were stratified according to weight (<80 kg, 80-100 kg, and >100 kg) and sex and then randomly assigned to 1 of the 4 diets, which resulted in 4 well-matched groups with no significant differences in baseline characteristics (Table 1). Of the 129 enrolled subjects, 13 dropped out (all female): 1 became pregnant, 1 failed to complete the final analysis, 2 moved interstate, and 9 cited disappointment with the rate of weight loss.Figure 1 shows the flow of participants through the study.
All 4 diets were designed as reduced-energy, reduced-fat (30% E), moderate-fiber (30 g/d) eating plans with differences in the quantity and quality of available CHOs. Diet 1 was a high-CHO (55% E) and average-protein (15% E) diet based on high-GI whole grains, including fiber-rich breakfast cereals and breads. Diet 2 had the same macronutrient proportions but was based on previously verified low-GI foods.17 Diet 3 was a higher-protein (25% E), CHO-reduced (45% E) diet based on lean red meat and high-GI CHO whole grains. Diet 4 had the same macronutrient proportions as diet 3 but specified low-GI CHO choices. The target GL (GI × CHO content), calculated as the sum of foods in sample menus, was highest in diet 1 and lowest in diet 4 (Table 2).
Subjects were given eating plans that were devised to help them lose weight (providing approximately 1400 kcal [6000 kJ] for women and 1900 kcal [8000 kJ] for men) and achieve the desired macronutrient distribution. These plans specified the number of servings from each food group and suitable choices within each group. Additional lists of appropriate meals and snacks were also provided. Dairy intake was held constant to minimize confounding from this source. To allow satiety and appetite factors to function, subjects were told to “eat to appetite” but were not required to consume all specified servings and, if hungry, were permitted to increase the number of servings proportionately from each food group.
To maximize compliance, all key CHO and protein foods and some preprepared meals were provided. A color-coded “shop” system (different color for each diet) with bar code reader was used, and subjects collected foods of low or high GI (eg, bread, breakfast cereal, crackers, snack bars, oats, legumes, pasta, and basmati or jasmine rice), red meat portions, and specially prepared frozen meals each week. On the same day, they met with a dietitian who encouraged compliance and answered queries.
To confirm that the diets produced differences in day-long glycemia and insulinemia, mixed meals representative of each diet (Table 3) were fed over a 10-hour period to 11 weight-stable volunteers (mean ± SE age, 26.5 ± 4.4 years; mean ± SE BMI, 30.0 ± 4.3) who had completed the weight-loss intervention. The 4 menus were given in random order on separate days, 3 to 7 days apart. Finger-prick capillary blood samples were collected at 30- to 60-minute intervals, and responses were quantified as the incremental areas under the curve (AUC [mean ± SE]).
At baseline and on completion of the 12-week intervention, dual-energy x-ray absorptiometry (Lunar Prodigy; GE Healthcare, Giles, England) was performed to assess changes in body composition (such as fat mass and lean mass). Weight was recorded weekly on electronic scales (Tanita Corp, Tokyo, Japan). At baseline and at weeks 6 and 12, fasting venous blood samples were obtained for measurement of levels of glucose (glucose hexokinase enzyme assay); insulin and leptin (microparticle enzyme immunometric assay on AxSym; Abbott Diagnostics, Abbott Park, Ill); total cholesterol, HDL cholesterol, and TG by standard automated methods18; free fatty acids by a commercially available enzymatic colorimetric test kit (NEFAC; Wako Pure Chemical Industries Ltd, Osaka, Japan); and C-reactive protein by near-infrared immunonephelometry (Beckman Coulter, Sydney, Australia). Low-density lipoprotein (LDL) cholesterol levels were calculated using the Friedewald calculation. We defined hyperinsulinemia as fasting insulin levels higher than 16 μIU/mL (110 pmol/L)19 and hypertriglyceridemia as fasting TG levels greater than or equal to 133 mg/dL (1.5 mmol/L).11 The homeostasis model assessment (HOMA) was used to assess β-cell function and insulin sensitivity from fasting glucose and insulin concentrations.We used both the original model, HOMA1 [(fasting glucose × fasting insulin)/22.5], and the updated nonlinear computer model, HOMA2, as described by Wallace et al.16 At baseline and during weeks 4 and 8, subjects were asked to keep a 3-day food diary, including 2 weekdays and 1 weekend day, to assess dietary compliance and to estimate food intake. A dietitian entered the data into a customized database that incorporated the Australian food composition tables and published GI values17 using the glucose equals 100 scale (FoodWorks Professional 2005; Xyris Software, Brisbane, Australia). Additional GI data were obtained from an online database (www.glycemicindex.com). The study was approved by the human ethics committee of the University of Sydney, Sydney, Australia, and subjects gave written, informed consent.
Power calculations indicated that 120 subjects (30 in each arm) provided 90% power to detect a 2-kg difference in body weight change among groups using significance equals 5%. The primary end points were mean absolute change from baseline in body weight and fat mass at week 12. Pearson χ2 analysis was used to compare the proportion of subjects in each group who achieved 5% or more weight loss. Univariate and repeated-measures analyses of variance were used to assess the changes in weight, body composition, and blood parameters. Changes were assessed with and without adjustment for baseline differences. Missing data were replaced with the last known value for the primary intention-to-treat analysis and excluded in the secondary analysis. A commercially available software package (SPSS Version 12.0; SPSS Inc, Chicago, Ill) was used for all statistical analyses.
In the primary intention-to-treat analysis, all 4 diets resulted in weight reduction over 12 weeks (mean change, 4.2% to 6.2% of body weight). Reductions in weight, fat mass, and waist circumference were significant within each group (in each case P<.001) but not among the 4 diets (Table 4 and Figure 2). There were significant differences in the proportion of individuals who lost 5% or more of initial body weight: 31% of subjects on diet 1, 56% on diet 2, 66% on diet 3, and 33% on diet 4 (P = .01). Sex influenced fat mass changes, with a significant interaction with diet (P = .008). In a subanalysis of women (n = 98), there were highly significant mean ± SE differences in fat mass change (−2.5 ± 0.5 kg, −4.5 ± 0.5 kg, −4.6 ± 0.5 kg, and −2.9 ± 0.5 kg for diets 1, 2, 3, and 4, respectively; P = .007) (Table 4). The GI had a significantly different effect (P = .02) in the high-CHO diets (lowering the GI doubled the fat loss from 2.8 kg to 4.5 kg) than in the high-protein diets (lowering the GI did not increase fat loss). This effect was more pronounced in women (P = .001). Declines in lean body mass reached significance in the 2 high-GI groups (P = .03 for diet 1 and P = .02 for diet 3), but overall there were no significant differences among the 4 diets (P = .75).
The pattern of findings (changes in weight, waist circumference, fat mass, and lean mass) was unchanged in the secondary sensitivity analysis, from which subjects who had not completed the study were excluded (data not shown). The proportion of subjects with a weight loss of 5% or more (n = 116) remained significant (37%, 60%, 68%, and 32% for diets 1, 2, 3, and 4, respectively; P = .02). The findings were also similar when the analysis was confined to those with high fasting insulin levels (≥16 μIU/mL [≥110 pmol/L], n = 37) or high fasting TG levels (≥133 mg/dL [≥1.5 mmol/L], n = 38). Among the latter, the subjects on the diet with the lowest GL (diet 4) achieved greater fat loss (−2.0 ± 0.8 [mean ± SE] kg, −4.9 ± 0.8 kg, −4.4 ± 0.9 kg, and −5.6 ± 1.0 kg for diets 1, 2, 3, and 4, respectively; P = .03).
In both the primary intention-to-treat analysis (Table 5 and Figure 3) and the secondary sensitivity analysis (data not shown), there was no differential effect of diet composition on the levels of HDL cholesterol, TG, free fatty acids, and C-reactive protein; total HDL cholesterol ratio; or glucose homeostasis. However, there were divergent effects on total and LDL cholesterol levels (P = .04 and P = .02, respectively). Both parameters increased in the diet 3 group (+5% and +8%, respectively) in contrast to a reduction in the diet 2 group (−4% and −6%, respectively; P = .03 and P = .01 for pair-wise comparisons). The increase in LDL cholesterol levels in the diet 3 group was greater in women (+10%) and significantly different from the decrease seen in the diet 2 group (−9%; P = .001). Overall, the GI, but not the protein content, had a significant effect on change in total cholesterol levels (P = .02) and LDL cholesterol levels (P = .009).
Diet had significant effects on changes in leptin levels (P<.006), which decreased more in the diet 2 group, with a significant interaction between GI and CHO content (P = .003; Table 5). Hence, lowering the GI resulted in a larger decrease in leptin levels when the CHO intake was high, whereas the reverse was true when the CHO intake was lower. However, on an individual basis, the absolute decrease in leptin levels correlated significantly with change in fat mass (r = 0.27; P = .003), with no additional effect of GI or CHO content. Changes in fat mass were also correlated with changes in fasting insulin concentration (r = 0.19; P = .03) and changes in insulin sensitivity as measured by HOMA2 (r = 0.20; P = .02).
The analysis of the food diaries showed that all 4 groups achieved their intended CHO and protein distributions (Table 2). There was no difference in apparent energy intake (P = .41; Table 6). In pair-wise comparisons, there was a 2-fold difference in GL between the 2 extreme diets (129 ± 8 [mean ± SE] g/d vs 59 ± 4 g/d, diet 1 vs diet 4, respectively; P<.001), but the intermediate diets were not different from each other (89 ± 5 g/d vs 75 ± 3 g/d, diet 2 vs diet 3, respectively; P = .25) (Figure 4). Mean dietary fiber intake was 25 g/d across all groups, but diet 2 was the only group to achieve the target intake (30 g/d; P<.001). All groups reduced fat intake, but the high-CHO groups ate less fat than the high-protein groups (−10 g/d; P<.001). Accordingly, the high-CHO groups ate less of each type of fat, although the ratio of saturated to unsaturated fatty acids (approximately 0.6) remained constant (Table 6). The cholesterol intake was relatively low in all groups, but the high-protein groups consumed more cholesterol than the high-CHO groups (293 ± 18 [mean ± SE] mg/d, 239 ± 18 mg/d, 125 ± 10 mg/d, and 119 ± 14 mg/d on diets 1, 2, 3, and 4, respectively; P<.001).
Postprandial glucose and insulin concentrations fluctuated throughout the day as predicted by the calculated GI and GL of the meals (Figure 5). The incremental glucose AUC over 10 hours was highest in the diet 1 group (315 ± 36 mmol/L · min) and lowest in the diet 4 group (196 ± 30 mmol/L · min; P = .02 for all groups). Similarly, the insulin AUC was highest in the diet 1 group (7940 ± 1160 pmol/L · min) and lowest in the diet 4 group (4340 ± 800 pmol/L · min; P = .005 for all groups). The GL was significantly correlated with the blood glucose AUC (r = 0.35; P = .022) and the insulin AUC (r = 0.35; P = .02). Varying the GI had a stronger effect (P = .03) on glycemia than varying the protein content (P = .05), but protein content had a greater effect on insulinemia (P = .002).
The findings of this study suggest that postprandial glycemia and dietary GL may be unrecognized determinants of the effectiveness of energy-restricted diets. The conventional diet (diet 1) was associated with the highest level of postprandial glycemia as well as with the slowest rate of weight loss. Subjects instructed to follow a high-CHO, low-GI (diet 2) or a high-protein, high-GI diet (diet 3) were twice as likely to achieve the clinical goal of a weight loss of 5% or more. In women in particular (n = 98), these 2 reduced-GL diets were associated with 80% greater fat loss compared with the conventional low-fat diet (diet 1; P<.007), without compromising lean mass. The low-GI, high-CHO diet (diet 2), however, produced the best clinical outcomes, reducing both fat mass and LDL cholesterol levels. The high-protein, high-GI diet (diet 3) produced an increase in total and LDL cholesterol concentrations (+8% overall, +10% in women), which contrasted with the decrease in LDL cholesterol concentrations that was seen with both low-GI diets (P = .01). The favorable result for diet 2 could not simply be the result of its low-fat content (actual 22% E), because the other high-CHO group also achieved a very low-fat intake (actual 19% E) yet the slowest rate of fat loss. Similarly, the increase in total and LDL cholesterol levels with diet 3 could not be the result of higher fat (actual 27% E) or meat intake, because the subjects on diet 4 ate similar amounts (actual fat 29% E) but showed no unfavorable effects. The changes in cholesterol levels were associated statistically with differences in the GI and not with the protein content (P<.009). The diet 2 group achieved a higher fiber intake than the other groups (approximately 7 g more), which may partly account for the favorable result. Soluble fibers intrinsic to legumes and low-GI whole grains (but not to high-GI whole grains) bind dietary cholesterol20 and may be critical in the context of a high-protein diet.
Our study suggests that dietary GL may be more relevant to women than to men. Women generally lose weight more slowly and display differences in postprandial glucose and fat oxidation, which might influence the rate of fat loss.21 The GL might also be more important in individuals with hypertriglyceridemia.11 In this subgroup, the diet with the lowest GL (diet 4) produced the greatest fat loss and improvement in the total HDL cholesterol ratio. In the total group, however, there was no support for the hypothesis that the lowest GL (diet 4) would be associated with the best outcomes. One explanation might be that more “extreme” diets require greater discipline, which often declines over time. Alternately, genetic or metabolic predisposition may need to be taken into account to determine the effectiveness of one diet over another.19
The strengths of the study include the study design, number of diets compared simultaneously, large sample size, high continuation rate, high compliance rate, provision of key foods, and detailed and repeated ascertainment of dietary measurements. A particular strength was extensive knowledge of the GI of individual Australian foods.17 Importantly, the day-long profiles confirmed that the 4 diets produced differential postprandial responses, as predicted by their calculated GL, CHO content, and GI. A further strength was that the subjects were free-living young adults who represent an important target for early intervention.
The study has limitations. Diet goals for energy distribution were not met exactly. We cannot discount the effect of these differences on study outcomes. One interpretation could be that the most successful diet was not only low GI but also very low fat (22% E) and higher in fiber (30 g/d). A 12-week period provides no information about the sustainability or long-term effects of the diets on cardiovascular function, exercise tolerance, renal function, or bone health. However, this is a common time frame for a “diet-only” intervention22; insufficient weight loss might dictate pharmacological approaches with both expense and adverse effects.
In conclusion, at least in the short term, our findings suggest that dietary GL, and not just overall energy intake, influences weight loss and postprandial glycemia. Moderate reductions in GL appear to increase the rate of body fat loss, particularly in women. Diets based on low-GI whole grain products (in lieu of whole grains with a high GI) maximize cardiovascular risk reduction, particularly if protein intake is high. Reassuringly, this advice can optimize clinical outcomes within current nutrition guidelines, without the concerns that apply to low-CHO diets. Multicenter studies to evaluate weight reduction, weight maintenance, and long-term outcomes, particularly in individuals with established risk factors, are clearly warranted.
Correspondence: Jennie Brand-Miller, PhD, Human Nutrition Unit (G08), University of Sydney, Sydney, New South Wales, Australia 2006 (email@example.com).
Accepted for Publication: April 12, 2006.
Author Contributions: Ms McMillan-Price and Drs Petocz and Brand-Miller had full access to all the data and take responsibility for the integrity and accuracy of the data analysis.
Financial Disclosure: Ms McMillan-Price and Dr Brand-Miller are coauthors of The Low GI Diet Revolution (Marlowe & Co., New York, NY, 2005). Dr Brand-Miller is a coauthor of The New Glucose Revolution book series (Hodder and Stoughton, London, England; Marlowe & Co; and Hodder Headline, Sydney, and elsewhere).
Funding/Support: This study was supported in part by the National Heart Foundation of Australia and Meat and Livestock Australia.
Role of the Sponsors: The funding bodies did not participate in the study design; the collection, analysis, or interpretation of the data; the writing of the manuscript; or the decision to submit the article for publication.
Acknowledgment: We thank Stephan Jacob, MD, Abdullah Omari, MD, Paul Nestel, MD, and John Miller, PhD, for comments on the manuscript and Zia Ahmed, MAppSc, for technical assistance.
Create a personal account or sign in to: