点击显示 收起
1 From the Departments of Food Science and Human Nutrition (JWK, HSS, and BL-H) and of Statistics (MJD), University of Florida, Gainesville, FL.
2 Supported by the primary investigator (JK). Funding did not come from any outside source. 3 Reprints not available. Address correspondence to J Krieger, University of Florida, Department of Food Science and Human Nutrition, PO Box 110370, Gainesville, FL 32611-0370. E-mail: jkrieger{at}ufl.edu.
ABSTRACT
Background: It is unclear whether low-carbohydrate, high-protein, weight-loss diets benefit body mass and composition beyond energy restriction alone.
Objective: The objective was to use meta-regression to determine the effects of variations in protein and carbohydrate intakes on body mass and composition during energy restriction.
Design: English-language studies with a dietary intervention of 4200 kJ/d (1000 kcal/d), with a duration of 4 wk, and conducted in subjects aged 19 y were considered eligible for inclusion. A self-reported intake in conjunction with a biological marker of macronutrient intake was required as a minimum level of dietary control. A total of 87 studies comprising 165 intervention groups met the inclusion criteria.
Results: After control for energy intake, diets consisting of 35–41.4% energy from carbohydrate were associated with a 1.74 kg greater loss of body mass, a 0.69 kg greater loss of fat-free mass, a 1.29% greater loss in percentage body fat, and a 2.05 kg greater loss of fat mass than were diets with a higher percentage of energy from carbohydrate. In studies that were conducted for >12 wk, these differences increased to 6.56 kg, 1.74 kg, 3.55%, and 5.57 kg, respectively. Protein intakes of >1.05 g/kg were associated with 0.60 kg additional fat-free mass retention compared with diets with protein intakes 1.05 g/kg. In studies conducted for >12 wk, this difference increased to 1.21 kg. No significant effects of protein intake on loss of either body mass or fat mass were observed.
Conclusion: Low-carbohydrate, high-protein diets favorably affect body mass and composition independent of energy intake, which in part supports the proposed metabolic advantage of these diets.
Key Words: Meta-analysis • body composition • high-protein diet • low-carbohydrate diet • weight loss
INTRODUCTION
Low-carbohydrate diets have become popular in recent years and contain less carbohydrate than that found in the Acceptable Macronutrient Distribution Range of 45–65% of energy (1). Samaha et al (2) reported that a low-carbohydrate diet resulted in greater weight loss than did a low-fat diet over a period of 6 mo; however, this was confounded by a greater, but not significant, reduction in energy intake in the low-carbohydrate group. In other studies, low-carbohydrate diets have resulted in greater weight loss than have low-fat diets, despite similar energy intakes between groups (3–5). However, this was not observed in all studies (6). Thus, whether a reduction in carbohydrate intake offers any benefit beyond energy restriction alone is unclear.
Low-carbohydrate diets typically contain more protein than the Recommended Dietary Allowance (RDA) of 0.8 g protein/kg body mass (7). The protein RDA is established by using data from subjects in energy balance (7). Because energy restriction can decrease nitrogen balance (8), the RDA may not be optimal for fat-free mass (FFM) retention during energy restriction. The effects of replacing carbohydrate with protein during energy restriction have been the focus of some recent investigations (9–12), but results have been inconsistent, with some studies showing an increased fat loss or FFM preservation in women but not men (9–11) and one study showing no effect (12).
These inconsistencies may relate to either differences in the study designs or small trials with low statistical power. Thus, it may be advantageous to combine the results of dietary intervention trials with meta-regression and to use study-level and group-level characteristics to predict changes in body mass and composition. Bravata et al (13) performed a meta-analysis of 94 dietary intervention trials and observed that carbohydrate content was not associated with the degree of weight loss (P = 0.90). However, they did not present data on body composition. It is also possible that they did not detect an effect of carbohydrate intake because of the high heterogeneity between the studies. In support for this possibility, they reported a near-trend (P = 0.10) of carbohydrate intake on weight loss when only a subset of homogeneous trials was examined. They excluded highly controlled interventions in which subjects were confined to a hospital or research center. Because self-reported energy intake is unreliable (14), a meta-regression of more highly controlled dietary interventions is needed. The purpose of this meta-regression was to determine the effects of variations in protein and carbohydrate intake on body mass and body composition measurements during energy restriction.
SUBJECTS AND METHODS
Study selection
Searches for English language studies that were published between 1 January 1950 and 18 September 2005 were performed in PubMed, the Cochrane Central Register of Controlled Trials, the Cumulative Index of the Nursing and Allied Health Literature, and SportDiscus (Table 1). Relevant studies published before 1950 were identified via bibliography searches of the retrieved articles. Studies involving a dietary intervention, subjects aged 19 y, and pre- and postdietary measurements of body mass or body composition constituted the initial criteria for eligibility. Sufficient data to determine energy intake, baseline body mass, macronutrient composition, and the mean change of the outcome measures were also required. The exclusion criteria are shown in Table 1. Studies were selected to meet a minimum level of dietary control. When dietary intake was self-reported, a biological marker measurement (ie, urinary or serum ketones, urinary nitrogen excretion, blood urea nitrogen, or plasma fatty acids) was required as an objective measure of compliance. In the absence of a biological marker, the investigators had to supply 60% of the subjects' energy intake as a requirement for eligibility in the meta-regression. Studies in which the authors reported that subjects were not in full compliance with the dietary intervention were excluded. All studies were performed in accordance with ethical guidelines.
View this table:
TABLE 1. Results of literature search
Data abstraction
Data were tabulated onto a spreadsheet with the use of MICROSOFT EXCEL (Microsoft Corp, Redmond, WA). Treatment arms that did not meet the inclusion criteria were excluded from the analysis. For crossover designs, separate rows were coded for each intervention. Periods of energy balance or insufficient dietary control were excluded.
Variables abstracted from each study were the following: study design, n, age, sex, baseline body mass (kg), quality of dietary control (moderate or high), duration of treatment (wk), exercise intervention (yes or no), method to measure body composition (field or laboratory), protein intake (percentage of energy, total g, and g/kg body mass), carbohydrate intake (percentage of energy and total g), fat intake (percentage of energy and total g), total energy intake (kJ), body mass change (kg), fat-free mass change (kg), percentage change in body fat (BF), and percentage change in fat mass (FM). The study design, age, sex, baseline body mass, quality of dietary control, duration of treatment, exercise intervention, method to measure body composition, percentage energy from carbohydrate (categorized into quartiles), and protein intake (g/kg body mass; categorized into quartiles) were included as predictors in the statistical models. If means (±SEMs) were not reported, the values were calculated from the individual subject data (when available). Data from subjects who did not meet the inclusion criteria were not included in the calculations. In some studies, there were multiple treatment arms, but the mean age was only provided for the entire study participant population. In those cases, the mean age for the entire participant population was included.
A dietary control quality classification was assigned to each group. The control was classified as moderate when the dietary control consisted of food records and a biological marker. The control was also classified as moderate when only part of the subjects' energy intake was supplied to them. The control was classified as high when all subjects' energy intake was supplied. The mean duration of the study was used when the duration of diet varied. The method of measuring body composition was classified as either a laboratory measure (ie, dual-energy X-ray absorptiometry, air densitometry, or hydrodensitometry) or a field measure (ie, skinfold thicknesses, bioelectric impedance analysis, or total-body electrical conductivity) (15). Carbohydrate intake (percentage energy) and protein intake (g/kg body mass) were classified into quartiles. Additional analyses were carried out with carbohydrate intake separated into low (1st quartile) and high (quartiles 2–4) intakes and protein intake separated into low (less than or equal to the median) and high (more than the median) intakes.
An independent investigator recoded 10 randomly selected studies to test the reliability of the abstraction process. Per case agreement was determined by dividing the variables coded the same by the total number of variables. A mean agreement of 0.96 was reached, which indicated that the abstraction process was reliable.
Missing values
In many studies, there was insufficient data to abstract all variables. When a value was missing for a dependent variable, the intervention group was excluded from the analysis for that outcome. Missing values for covariates were calculated from available data when possible. Any remaining missing covariates and within-group variances were replaced by using multiple imputation (16). Ten imputed data sets were created and analyzed for each outcome, and the results were combined for statistical inferences.
Statistical analyses
The variance within each intervention group was calculated as the squared SEM of the difference between pre- and postdiet outcomes. If the SEM of the difference was not reported, the SEM of the difference was calculated by using the P value or CI (when available). Otherwise, an upper bound on the SEM was calculated by using the following formula (17):
RESULTS
Body mass change
The analysis of changes in the subjects' body mass comprised 165 treatment groups from 87 studies (Table 2). The mean change in body mass between these studies was –5.99 kg (CI: –6.71, –5.26 kg). The Bayesian Information Criterion decreased from 735.0 in the full model to 726.9 in the reduced model. The reduced model was not significantly different from the full model (P = 0.31).
View this table:
TABLE 2. Studies included in the analysis
Predictors in the reduced models are shown in Table 3. Diets with carbohydrate intake in the lowest quartile (
View this table:
TABLE 3. Final reduced models for body mass1
Fat-free mass change
The analysis of changes in FFM included 102 treatment groups from 51 studies (Table 3). The mean change was –1.20 kg (CI: –1.51, –0.87 kg). The reduced model was not significantly different from the full model (P = 0.83).
Predictors in the reduced models are shown in Table 4. The amount of FFM retained tended to increase with each successive quartile of protein intake, with a significant difference existing between the upper 2 quartiles (>1.05 g/kg) and the first quartile (0.70 g/kg). Specifically, the third quartile (>1.05 and 1.20 g/kg) was associated with 0.78 kg additional FFM retention (CI: 0.02, 1.54 kg) and the fourth quartile (>1.20 g/kg) was associated with 0.96 kg additional FFM retention (CI: 0.16, 1.77 kg). When protein intake was categorized as high (>1.05 g/kg) or low (1.05 g/kg), a significant effect remained, although the amount of FFM retained in the high-protein intake group decreased to 0.60 kg (CI: 0.16, 1.05 kg). In studies conducted for 12 wk, the additional FFM retained by the high-protein intake group decreased to 0.34 kg and was no longer significant (CI: –0.14, 0.82 kg). In studies conducted for >12 wk, high-protein diets were associated with an additional 1.21 kg FFM retention (CI: 0.49, 1.93 kg).
View this table:
TABLE 4. Final reduced models for fat-free mass1
Compared with carbohydrate intake in the lowest quartile (41.4% of energy), the carbohydrate intake in the highest quartile (>56.9% of energy) was associated with 0.98 kg greater FFM retention (CI: 0.25, 1.70 kg). Carbohydrate intake in the second (>41.4%, 46.4%) and third (>46.4%, 56.9%) quartiles tended to be associated with 0.62–0.65 kg more FFM retention (P = 0.06). When carbohydrate intake was classified as either low (41.4%) or high (>41.4%), low-carbohydrate diets were associated with a greater loss of FFM (0.69 kg; CI: –0.16, –1.22 kg) than were high-carbohydrate diets. In studies conducted for 12 wk, the magnitude of this effect decreased to 0.31 kg (CI: –0.90, 0.27 kg) and was no longer significant. In studies conducted for >12 wk, low-carbohydrate diets were associated with a greater loss of FFM (1.74 kg; CI: 0.01, 3.47 kg) than were high-carbohydrate diets. A sensitivity analysis did not uncover any influential studies and there was no evidence of a publication bias (P = 0.10).
Percentage changes in body fat
The analysis of percentage changes in BF was composed of 98 treatment groups from 49 studies (Table 3). The mean change was –3.00% (CI: –3.53%, –2.46%). The reduced model was not significantly different from the full model (P = 0.75).
Predictors in the reduced model are shown in Table 5. Protein intake in the third quartile (>1.06 g/kg and 1.20 g/kg) was associated with a greater loss of percentage BF (1.32%; CI: 0.11%, 2.53%) than was the first quartile (0.73 g/kg). When protein intake was classified as high (>1.06 g/kg) or low (1.06 g/kg), there was a trend (P = 0.09) toward a 0.64% (CI: –0.09%, 1.38%) greater loss of percentage BF with the higher protein intake. In studies conducted for 12 wk, the loss in percentage BF in the high-protein group compared with the low-protein group decreased to 0.45% and the trend no longer existed (P = 0.38; CI: –0.56%, 1.46%). In studies conducted for >12 wk, the loss in percentage BF increased to 0.96% in the high-protein group compared to the low-protein group, but the difference was not significant (P = 0.21; CI: –0.76%, 2.67%).
View this table:
TABLE 5. Final reduced models for percentage body fat1
Diets with a carbohydrate intake in the lowest quartile (41.4% energy) were associated with a 1.32–1.48% greater decrease in percentage BF than were diets with carbohydrate intake in the highest 3 quartiles. When carbohydrate intake was categorized as low (41.4% energy) or high (>41.4% energy), the significantly greater decrease in percentage BF in the low-carbohydrate intake group remained (1.29%; CI: 0.46%, 2.12%). In studies conducted for 12 wk, the greater loss in percentage BF in the lowest carbohydrate intake quartile tended toward significance (1.00%; CI: –0.06%, 2.06%; P = 0.06). In studies conducted for >12 wk, low-carbohydrate diets were associated with a greater decrease in percentage BF (3.55%; CI: 1.62%, 5.49%) than were high-carbohydrate diets. A sensitivity analysis did not uncover any influential studies and there was no evidence of publication bias (P = 0.27).
Fat mass changes
The analysis of changes in FM included 108 treatment groups from 52 studies (Table 3). The mean change was –4.71 kg (CI: –5.41, –4.00 kg). The reduced model was not significantly different from the full model (P = 0.48).
Predictors in the reduced model are shown in Table 6. Protein intake in the third quartile (>1.06 and 1.18 g/kg) was associated with a greater loss of FM (1.68 kg; CI: 0.01, 3.35 kg) than was the first quartile of protein intake (0.73 g/kg). When protein intake was classified as high (>1.06 g/kg) or low (1.06 g/kg), there was no significant effect of protein intake on FM loss (P = 0.19).
View this table:
TABLE 6. Final reduced models for fat mass1
Diets with carbohydrate intake in the lowest quartile (
DISCUSSION
The purpose of this meta-analysis was to determine whether low-carbohydrate, high-protein diets have effects on body composition independent of energy intake. By including carbohydrate and protein intake as predictors in multilevel regression models, one can determine the independent contributions of these variables to the variation in body mass and composition changes across a large number of studies while simultaneously adjusting for the differences in energy intake and other covariates between the studies and treatment groups.
Protein intake was a significant predictor of FFM retention. A daily protein intake of >1.05 g/kg (
Compared with higher carbohydrate intakes, low-carbohydrate diets (35–41.4% energy) increased the loss of body mass, BF, and percentage BF, even after control for energy intake as a covariate in the regression analyses. The mean total carbohydrate intake in the low-carbohydrate studies ranged from 79–97 g, depending on the analysis. Typically, a carbohydrate intake of <100 g will cause ketosis (1). These results support the apparent metabolic advantage of low-carbohydrate, ketogenic diets (104). The additional body mass change is not likely due to water loss, because the duration of the diet periods (6–24 wk) was too protracted (5, 75, 92) and estimations of total body water tend to be similar between low-carbohydrate and low-fat diets after 2 wk (5). The similar results of the analyses on body mass and BF also supports the concept that the effect on body mass of low-carbohydrate diets is an effect on FM rather than on body water. Feinman and Fine (104) argued that low-carbohydrate diets increase the demands on protein and amino acid turnover for gluconeogenesis. Because this process has a high energy cost, it would increase the energy deficit for a given energy intake, thereby supporting the theory of a metabolic advantage of low-carbohydrate diets. In contrast, Buchholz and Schoeller (105) averaged the results of 10 studies and reported no effect of low-carbohydrate diets on 24-h energy expenditure. However, none of the studies they cited involved ketogenic diets, and most of the studies were conducted with subjects in energy balance. A hypocaloric, ketogenic diet would be expected to increase the demand for gluconeogenesis because of the low energy and carbohydrate availability. In contrast to this hypothesis, Brehm et al (4) reported no differences in total energy expenditure when a low-carbohydrate diet was compared with a low-fat diet. However, total energy expenditure was estimated rather than directly measured with the use of a whole-body calorimeter or doubly labeled water. Future research should focus on the effects of low-carbohydrate diets on energy expenditure with the use of these measurement tools.
Alternatively, the higher loss in BF observed with low-carbohydrate diets than that observed with low-fat diets may relate to changes in insulin concentrations, because less insulin promotes free-fatty acid mobilization from BF storage (106). Volek et al (75) reported a significant positive correlation between decreases in insulin concentration and reductions in FM (R2 = 0.67) and percentage BF (R2 = 0.70) when subjects were placed on a diet of 8% energy from carbohydrate (46 g/d) for 6 wk. The additional fat loss may also be related to the excretion of ketones in the urine and breath; however, this would only account for a maximum of 420 kJ/d (107), which would only amount to 1 kg of additional BF loss over a 3-mo dieting period. This is only one-half of the greater loss of FM observed with the low-carbohydrate diets the current analysis.
It is also possible that subjects on low-fat diets systematically underreport energy intake compared with subjects on low-carbohydrate diets. In support for this hypothesis, Brehm et al (4) observed that actual weight loss closely matched the predicted weight loss in the low-carbohydrate group, but actual weight loss was less than the predicted weight loss in the low-fat group. In the current analysis, high quality studies (ie, those in which food was prepared for the subjects) resulted in greater weight and FM loss than did lower quality studies (ie, those that generally involved self-reported measurements in conjunction with a biological marker of macronutrient intake). However, the effects of carbohydrate intake were independent of study quality, which indicates that carbohydrate intake had an effect whether the subjects self-reported food intake or consumed food that was prepared for them. Thus, our analyses do not support the idea of a systematic bias in the reporting of energy intake.
Low-carbohydrate diets were associated with a greater FFM loss than were low-fat diets. The additional FFM loss may reflect an additional loss of body water, because body water is a component of FFM and ketosis may cause water excretion (108). The additional FFM loss may also be caused by lower insulin concentrations, because insulin inhibits proteolysis (109).
Sensitivity analyses indicated that the results were quite robust to the removal of individual studies. Thus, no studies had a large effect on the estimates produced by the regression models. Also, with the exception of FM, there was no evidence of a publication bias. The slope of the funnel plot regression for FM was quite low (0.10), which indicated a weak relation between sample size and weight. This relation was in a positive direction, which indicated that larger population studies had a greater effect on the analysis than did smaller studies. This is expected, because larger sample sizes tend to reduce the variation in within-treatment groups. Thus, the significant slope observed for FM likely does not represent a publication bias.
In conclusion, low-carbohydrate diets may increase the loss of body mass, FFM, FM, and percentage BF during weight reduction compared with traditional diets. The RDA for protein may be insufficient for optimal FFM retention during weight loss; high protein intakes (>1.05 g/kg) may improve FFM retention.
ACKNOWLEDGMENTS
We thank Jan Hudgens for independently recoding 10 randomly selected studies.
JWK collected, analyzed, and interpreted the data, was involved in the design of the study, and was the primary writer of the manuscript. HSS, MJD, and BL-H were involved in the design of the study, data interpretation, and writing of the manuscript. None of the authors had any financial or personal conflicts of interest.
REFERENCES