ABSTRACT

Background Trials comparing the effectiveness and safety of weight-loss diets are frequently limited by short follow-up times and high dropout rates.

Methods In this 2-year trial, we randomly assigned 322 moderately obese subjects (mean age, 52 years; mean body-mass index [the weight in kilograms divided by the square of the height in meters], 31; male sex, 86%) to one of three diets: low-fat, restricted-calorie; Mediterranean, restricted-calorie; or low-carbohydrate, non–restricted-calorie.

Results The rate of adherence to a study diet was 95.4% at 1 year and 84.6% at 2 years. The Mediterranean-diet group consumed the largest amounts of dietary fiber and had the highest ratio of monounsaturated to saturated fat (P<0.05 for all comparisons among treatment groups). The low-carbohydrate group consumed the smallest amount of carbohydrates and the largest amounts of fat, protein, and cholesterol and had the highest percentage of participants with detectable urinary ketones (P<0.05 for all comparisons among treatment groups). The mean weight loss was 2.9 kg for the low-fat group, 4.4 kg for the Mediterranean-diet group, and 4.7 kg for the low-carbohydrate group (P<0.001 for the interaction between diet group and time); among the 272 participants who completed the intervention, the mean weight losses were 3.3 kg, 4.6 kg, and 5.5 kg, respectively. The relative reduction in the ratio of total cholesterol to high-density lipoprotein cholesterol was 20% in the low-carbohydrate group and 12% in the low-fat group (P=0.01). Among the 36 subjects with diabetes, changes in fasting plasma glucose and insulin levels were more favorable among those assigned to the Mediterranean diet than among those assigned to the low-fat diet (P<0.001 for the interaction among diabetes and Mediterranean diet and time with respect to fasting glucose levels).

Conclusions Mediterranean and low-carbohydrate diets may be effective alternatives to low-fat diets. The more favorable effects on lipids (with the low-carbohydrate diet) and on glycemic control (with the Mediterranean diet) suggest that personal preferences and metabolic considerations might inform individualized tailoring of dietary interventions. (ClinicalTrials.gov number, NCT00160108 [ClinicalTrials.gov] .)

The dramatic increase in obesity worldwide remains challenging and underscores the urgent need to test the effectiveness and safety of several widely used weight-loss diets.^1,2,3 Low-carbohydrate, high-protein, high-fat diets (referred to as low-carbohydrate diets) have been compared with low-fat, energy-restricted diets.^4,5,6,7,8,9 A meta-analysis of five trials with 447 participants¹⁰ and a recent 1-year trial involving 311 obese women⁴ suggested that a low-carbohydrate diet is a feasible alternative to a low-fat diet for producing weight loss and may have favorable metabolic effects. However, longer-term studies are lacking.^4,10 A Mediterranean diet with a moderate amount of fat and a high proportion of monounsaturated fat provides cardiovascular benefits.¹¹ A recent review citing several trials¹² included a few that suggested that the Mediterranean diet was beneficial for weight loss.^13,14 However, this positive effect has not been conclusively demonstrated.¹⁵

Common limitations of dietary trials include high attrition rates (15 to 50% within a year), small size, short duration, lack of assessment of adherence, and unequal intensity of intervention.^{10,12,15,16,17} We conducted the 2-year Dietary Intervention Randomized Controlled Trial (DIRECT) to compare the effectiveness and safety of three nutritional protocols: a low-fat, restricted-calorie diet; a Mediterranean, restricted-calorie diet; and a low-carbohydrate, non–restricted-calorie diet.

Methods

Eligibility and Study Design

We conducted the trial between July 2005 and June 2007 in Dimona, Israel, in a workplace at a research center with an on-site medical clinic. Recruitment began in December 2004. The criteria for eligibility were an age of 40 to 65 years and a body-mass index (BMI, the weight in kilograms divided by the square of the height in meters) of at least 27, or the presence of type 2 diabetes (according to the American Diabetes Association criteria¹⁸) or coronary heart disease, regardless of age and BMI. Persons were excluded if they were pregnant or lactating, had a serum creatinine level of 2 mg per deciliter (177 µmol per liter) or more, had liver dysfunction (an increase by a factor of at least 2 above the upper limit of normal in alanine aminotransferase and aspartate aminotransferase levels), had gastrointestinal problems that would prevent them from following any of the test diets, had active cancer, or were participating in another diet trial.

The participants were randomly assigned within strata of sex, age (below or above the median), BMI (below or above the median), history of coronary heart disease (yes or no), history of type 2 diabetes (yes or no), and current use of statins (none, <1 year, or 1 year) with the use of Monte Carlo simulations. The participants received no financial compensation or gifts. The study was approved and monitored by the human subjects committee of Soroka Medical Center and Ben-Gurion University. Each participant provided written informed consent.

The members of each of the three diet groups were assigned to subgroups of 17 to 19 participants, with six subgroups for each group. Each diet group was assigned a registered dietitian who led all six subgroups of that group. The dietitians met with their groups in weeks 1, 3, 5, and 7 and thereafter at 6-week intervals, for a total of 18 sessions of 90 minutes each. We adapted the Israeli version (developed by the Maccabi Health Maintenance Organization) of the diabetes-prevention program¹⁹ and developed additional themes for each diet group (see Supplementary Appendix 1, available with the full text of this article at www.nejm.org). In order to maintain equal intensity of treatment, the workshop format and the quality of the materials were similar among the three diet groups, except for instructions and materials specific to each diet strategy. Six times during the 2-year intervention, another dietitian conducted 10-to-15-minute motivational telephone calls with participants who were having difficulty adhering to the diets and gave a summary of each call to the group dietitian. In addition, a group of spouses received education to strengthen their support of the participants (data not shown).

Low-Fat Diet

The low-fat, restricted-calorie diet was based on American Heart Association²⁰ guidelines. We aimed at an energy intake of 1500 kcal per day for women and 1800 kcal per day for men, with 30% of calories from fat, 10% of calories from saturated fat, and an intake of 300 mg of cholesterol per day. The participants were counseled to consume low-fat grains, vegetables, fruits, and legumes and to limit their consumption of additional fats, sweets, and high-fat snacks.

Mediterranean Diet

The moderate-fat, restricted-calorie, Mediterranean diet was rich in vegetables and low in red meat, with poultry and fish replacing beef and lamb. We restricted energy intake to 1500 kcal per day for women and 1800 kcal per day for men, with a goal of no more than 35% of calories from fat; the main sources of added fat were 30 to 45 g of olive oil and a handful of nuts (five to seven nuts, <20 g) per day. The diet is based on the recommendations of Willett and Skerrett.²¹

Low-Carbohydrate Diet

The low-carbohydrate, non–restricted-calorie diet aimed to provide 20 g of carbohydrates per day for the 2-month induction phase and immediately after religious holidays, with a gradual increase to a maximum of 120 g per day to maintain the weight loss. The intakes of total calories, protein, and fat were not limited. However, the participants were counseled to choose vegetarian sources of fat and protein and to avoid trans fat. The diet was based on the Atkins diet (see Supplementary Appendix 2).²²

Nutritional and Color Labeling of Food in the Cafeteria

Lunch is typically the main meal in Israel. The self-service cafeteria in the workplace provided a varied menu and was the exclusive source of lunch for the participants. A dietitian worked closely with the kitchen staff to adjust specific food items to specific diet groups. Each food item was provided with a label showing the number of calories and the number of grams of carbohydrates, fat, and saturated fat, according to an analysis based on the Israeli nutritional database. Each food item was also labeled with a full circle (indicating "feel free to consume") or a half circle (indicating "consume in moderation"). The labels were color-coded according to diet group and were updated daily (see Supplementary Appendix 2).²³

Electronic Questionnaires at Baseline and Follow-up

Adherence to the diets was evaluated by a validated food-frequency questionnaire²⁴ that included 127 food items and three portion-size pictures for 17 items.²⁵ A subgroup of participants completed two repeated 24-hour dietary recalls to verify absolute intake (data not shown). We used a validated questionnaire to assess physical activity.²⁶ At baseline and at 6, 12, and 24 months of follow-up, the questionnaires were self-administered electronically through the workplace intranet. The 15% of participants who requested aid in completing the questionnaires were assisted by the study nurse. The electronic questionnaire helped to ensure completeness of the data by prompting the participant when a question was not answered, and it permitted rapid automated reporting by the group dietitians.

Outcomes

The participants were weighed without shoes to the nearest 0.1 kg every month. With the use of a wall-mounted stadiometer, height was measured to the nearest millimeter at baseline for determination of BMI. Waist circumference was measured halfway between the last rib and the iliac crest. Blood pressure was measured every 3 months with the use of an automated system (Datascop Acutor 4) after 5 minutes of rest.

Blood samples were obtained by venipuncture at 8 a.m. after a 12-hour fast at baseline and at 6, 12, and 24 months and were stored at –80°C until an assay for lipids, inflammatory biomarkers, and insulin could be performed. Levels of fasting plasma glucose, glycated hemoglobin, and liver enzymes were measured in fresh samples. The level of glycated hemoglobin was determined with the use of Cobas Integra reagents and equipment. Serum levels of total cholesterol, high-density-lipoprotein (HDL) cholesterol, low-density-lipoprotein (LDL) cholesterol, and triglycerides were determined enzymatically with a Wako R-30 automatic analyzer, with coefficients of variation of 1.3% for cholesterol and 2.1% for triglycerides. Plasma insulin levels were measured with the use of an enzyme immunometric assay (Immulite automated analyzer, Diagnostic Products), with a coefficient of variation of 2.5%. Plasma levels of high-molecular-weight adiponectin were measured by an enzyme-linked immunosorbent assay (ELISA) (AdipoGen or Axxora), with a coefficient of variation of 4.8%. Plasma leptin levels were assessed by ELISA (Mediagnost), with a coefficient of variation of 2.4%. Plasma levels of high-sensitivity C-reactive protein were measured by ELISA (DiaMed), with a coefficient of variation of 1.9%. The clinic and laboratory staff members were unaware of the treatment assignments, and the study coordinators were unaware of all outcome data until the end of the intervention.

Statistical Analysis

For weight loss, the prespecified primary aim was the change in weight from baseline to 24 months. We used the Israeli food database²³ in the analysis of the results of the dietary questionnaires. We analyzed the dietary-composition data and biomarkers with the use of raw unadjusted means, without imputation of missing data. We compared the dietary-intake values between groups at each time point with the use of an analysis of variance in which all pairwise comparisons among the three diet groups were performed with the use of Tukey!bnny&ears Studentized range test. We transformed physical-activity scores into metabolic equivalents per week²⁷ according to the amount of time spent in various forms of exercise per week, with each activity weighted in terms of its level of intensity. For intention-to-treat analyses, we included all 322 participants and used the most recent values for weight and blood pressure. To evaluate the repeated measurements over time, we used generalized estimating equations for panel data analysis, also known as cross-sectional time-series analysis, with the use of the Stata software XTGEE command; this allowed us to account for the nonindependence of repeated measurements of the same bioindicator in the same participant over time. We used age, sex, time point, and diet group as explanatory variables in our models. To study changes over time and the effects of sex or the presence or absence of diabetes, we added appropriate interaction terms. We assessed the within-person changes from baseline in each diet group with the use of pairwise comparisons. We calculated the homeostasis model assessment of insulin resistance (HOMA-IR) according to the following equation²⁸: insulin (U/ml) x fasting glucose (mmol/liter) ÷ 22.5. For a mean (±SD) difference between groups of at least 2±10 kg of weight loss, with 100 participants per group and a type I error of 5%, the power to detect significant differences in weight loss is greater than 90%. We used SPSS software, version 15, and Stata software, version 9, for the statistical analysis.

Results

Characteristics of the Participants

The baseline characteristics of the participants are shown in Table 1. The mean age was 52 years and the mean BMI was 31. Most participants (86%) were men. The overall rate of adherence (Figure 1) was 95.4% at 12 months and 84.6% at 24 months; the 24-month adherence rates were 90.4% in the low-fat group, 85.3% in the Mediterranean-diet group, and 78.0% in the low-carbohydrate group (P=0.04 for the comparison among diet groups). During the study, there was little change in usage of medications, and there were no significant differences among groups in the amount of change; four participants initiated and three stopped cholesterol-lowering therapy. Twenty participants initiated blood-pressure treatment, five initiated medications for glycemic control, and one reduced the dosage of medications for glycemic control.