The flow of studies in our meta-analysis is depicted in Figure . From 785 potentially relevant references, 503 records remained after duplicates between databases were eliminated, and 406 articles were eliminated after reviewing titles and abstracts. A total of 97 full-text articles were reviewed for eligibility. Of those, 23 studies met all of the eligibility criteria and were included in the meta-analysis (5
). These studies included data from 2,788 trial participants (1,392 on low-carbohydrate diets and 1,396 on low-fat diets).
Flow diagram of systematic review (from January 1, 1966 to June 20, 2011).
The characteristics of these 23 randomized controlled trials are presented in Table . All trials were parallel except for 1 trial that had a factorial design (13
). Trial participants were usually not blinded to their assignment because of the nature of the intervention—most interventions provided dietary instruction, leaving food buying and/or preparation to the participants. Study duration ranged from 6 to 24 months, and 16 studies had an intervention duration of 12 months or longer (6
). Most trials were conducted among obese or overweight patients without cardiovascular diseases or diabetes mellitus. However, 4 studies were conducted in patients with diabetes (23
), 1 was conducted in patients with prediabetes (31
), and 1 was conducted in patients with coronary heart disease (23
). The goal dietary nutritional composition varied across the studies, with carbohydrate consumption ranging from 4% to 45% (weighted mean, 23%) of energy intake in the low-carbohydrate group and fat ranging from 10% to 30% (weighted mean, 26%) of energy intake in the low-fat group. Self-reported mean energy intake weighted by trial sample sizes was similar for both diets at approximately 2,000 kcal.
Characteristics of 23 Randomized Controlled Clinical Trials From Multiple Countries Comparing Low-Carbohydrate Diets With Low-Fat Diets, 1966–2011
Table shows baseline characteristics of trial participants in the included studies. The mean age ranged from 27 to 60 years. Approximately 40% of participants were male. Baseline levels of body weight and metabolic risk factors were similar between the 2 diets in each study but varied among studies.
Baseline Characteristics of Study Participants in 23 Randomized Controlled Trials From Multiple Countries Comparing Low-Carbohydrate Diets With Low-Fat Dietsa, 1966–2011
Pooled mean net changes and 95% confidence intervals for metabolic risk factors are presented in Web Figure 1 (available at http://aje.oxfordjournals.org/
). The weighted mean changes in outcomes were −6.1 versus −5.0 kg for body weight, −6.2 versus −6.0 cm for waist circumference, −4.6 versus −10.1 mg/dL for total cholesterol, −2.1 versus −6.0 mg/dL for LDL cholesterol, 4.5 versus 1.6 mg/dL for HDL cholesterol, −0.7 versus −0.5 for ratio of total to HDL cholesterol, −30.4 versus −17.1 mg/dL for triglycerides, −3.5 versus −3.0 mm Hg for systolic blood pressure, and −10.4 versus −10.1 mg/dL for fasting blood glucose for low-carbohydrate versus low-fat diets, respectively. Pooled mean net changes and 95% confidence intervals representing differences between the diets in body weight (−1.0 kg, 95% confidence interval (CI): −2.2, 0.2) and waist circumference (−0.1 cm, 95% CI: −0.6, 0.4) reductions were not statistically significant. Compared with participants on low-fat diets, those on low-carbohydrate diets experienced slightly but statistically significantly less reduction in total cholesterol (pooled mean net change, 2.7 mg/dL, 95% CI: 0.8, 4.6) and LDL cholesterol (pooled mean net change, 3.7 mg/dL, 95% CI: 1.0, 6.4) but a greater increase in HDL cholesterol (pooled mean net change, 3.3 mg/dL, 95% CI: 1.9, 4.7) and a greater decrease in triglycerides (pooled mean net change, −14.0 mg/dL, 95% CI: −19.4, −8.7). These differences remained statistically significant after correction for multiple comparisons. Pooled mean net changes in systolic blood pressure (−1.0 mm Hg, 95% CI: −3.5, 1.5), ratio of total to HDL cholesterol (−0.1, 95% CI: −0.3, 0.1), and fasting blood glucose (−0.3 mg/dL, 95% CI: −1.9, 1.3) were not significantly different between the 2 diets. The pooled mean net changes in diastolic blood pressures and serum insulin were also not significant (data not shown).
There was no significant heterogeneity in the net changes in total cholesterol (I2 = 0.2%, P = 0.45), triglycerides (I2 = 55.6%, P = 0.07), waist circumference (I2 = 12.5%, P = 0.33), fasting blood glucose (I2 = 41.2%, P = 0.06), or serum insulin (I2 = 7.8%, P = 0.29) among these trials. However, statistically significant heterogeneity was detected for body weight (I2 = 85.7%, P < 0.001), systolic blood pressure (I2 = 91.7%, P < 0.001), diastolic blood pressure (I2 = 40.8%, P = 0.04), LDL cholesterol (I2 = 50.0%, P = 0.01), HDL cholesterol (I2 = 78.6%, P < 0.001), and ratio of total to HDL cholesterol (I2 = 75.0%, P = 0.003) among trials.
We examined the potential for publication bias by plotting sample sizes against mean net changes in each metabolic risk factor (Web Figure 2
). Possible publication bias was detected for triglycerides (P
= 0.02) using Begg's rank correlation and for body weight (P
= 0.03), total cholesterol (P
= 0.03), LDL cholesterol (P
= 0.001), HDL cholesterol (P
< 0.001), ratio of total to HDL cholesterol (P
= 0.03), and insulin (P
= 0.03) using Egger's linear regression tests. We used the trim-and-fill method to estimate the potential effect of publication bias on our results. When corrected for the effects of possible publication bias, pooled net change estimates for total cholesterol, LDL cholesterol, and HDL cholesterol became nonsignificant, but the pooled mean net change in body weight was significant at −3.2 kg (95% CI: −4.5 to −2.0), favoring low-carbohydrate diets.
In sensitivity analyses, the exclusion of any one study from the analysis did not significantly alter the net changes in metabolic risk factors. In addition, after removing studies with fewer than 20 participants per group or studies among patients with breast cancer, polycystic ovarian syndrome, or coronary heart disease or who had undergone gastric bypass surgery, body weight reduction was significantly greater on low-carbohydrate diets, with pooled mean net changes in body weight of −1.3 kg (95% CI: −2.5 to −0.1, n = 19 studies) and −1.4 kg (95% CI: −2.6 to −0.2, n = 18 studies), respectively (Table ).
Main Results and Results From Sensitivity Analyses, 1966–2011
Subgroup analyses by gender, diabetic status, level of carbohydrate restriction, and study duration did not identify statistically significant differences in the majority of metabolic risk factor reductions between low-carbohydrate and low-fat diets (Web Tables 1–5
). However, there was a significantly greater reduction in body weight among participants who were on low-carbohydrate diets for more than 1 year (−0.9 kg, 95% CI: −1.6, −0.3) and those with a high level of carbohydrate restriction (−2.0 kg, 95% CI: −3.4 to −0.6) when compared with persons on low-fat diets. In addition, in the younger age group, low-carbohydrate diets resulted in significantly greater reductions in systolic blood pressure (−2.7 mm Hg, 95% CI: −4.5 to −0.9), diastolic blood pressure (−1.5 mm Hg, −2.7 to −0.2), and serum insulin (−0.9 μIU/mL, −1.8 to −0.1) as compared with results in the older age groups. However, these findings were not statistically significant after adjustment for multiple testing.