Content uploaded by Terezinha da Rocha Ataíde
Author content
All content in this area was uploaded by Terezinha da Rocha Ataíde on Aug 05, 2014
Content may be subject to copyright.
Systematic Review with Meta-analysis
Very-low-carbohydrate ketogenic diet
v.
low-fat diet for long-term
weight loss: a meta-analysis of randomised controlled trials
Nassib Bezerra Bueno*, Ingrid Sofia Vieira de Melo, Suzana Lima de Oliveira
and Terezinha da Rocha Ataide
Laborato
´rio de Nutric¸a˜o Experimental, Faculdade de Nutric¸a˜o, Universidade Federal de Alagoas, Campus A. C. Simo
˜es,
BR 104 Norte, Km 97, 57.072-970 Tabuleiro do Martins, Maceio
´, AL, Brazil
(Submitted 7 November 2012 – Final revision received 24 January 2013 – Accepted 28 January 2013 – First published online 7 May 2013)
Abstract
The role of very-low-carbohydrate ketogenic diets (VLCKD) in the long-term management of obesity is not well established. The present
meta-analysis aimed to investigate whether individuals assigned to a VLCKD (i.e. a diet with no more than 50 g carbohydrates/d) achieve
better long-term body weight and cardiovascular risk factor management when compared with individuals assigned to a conventional low-
fat diet (LFD; i.e. a restricted-energy diet with less than 30 % of energy from fat). Through August 2012, MEDLINE, CENTRAL, ScienceDirect,
Scopus, LILACS, SciELO, ClinicalTrials.gov and grey literature databases were searched, using no date or language restrictions, for random-
ised controlled trials that assigned adults to a VLCKD or a LFD, with 12 months or more of follow-up. The primary outcome was body
weight. The secondary outcomes were TAG, HDL-cholesterol (HDL-C), LDL-cholesterol (LDL-C), systolic and diastolic blood pressure,
glucose, insulin, HbA
1c
and C-reactive protein levels. A total of thirteen studies met the inclusion/exclusion criteria. In the overall analysis,
five outcomes revealed significant results. Individuals assigned to a VLCKD showed decreased body weight (weighted mean difference
20·91 (95 % CI 21·65, 20·17) kg, 1415 patients), TAG (weighted mean difference 20·18 (95 % CI 20·27, 20·08) mmol/l, 1258 patients)
and diastolic blood pressure (weighted mean difference 21·43 (95 % CI 22·49, 20·37) mmHg, 1298 patients) while increased HDL-C
(weighted mean difference 0·09 (95 % CI 0·06, 0·12) mmol/l, 1257 patients) and LDL-C (weighted mean difference 0·12 (95 % CI 0·04,
0·2) mmol/l, 1255 patients). Individuals assigned to a VLCKD achieve a greater weight loss than those assigned to a LFD in the long
term; hence, a VLCKD may be an alternative tool against obesity.
Key words: Cardiovascular risk factors: Low-carbohydrate diets: Meta-analysis: Obesity: Weight loss
Obesity continues to be a major worldwide health problem,
despite the efforts of the medical community. At least 2·8
million adults die from obesity-related causes each year, and
65 % of the worldwide population lives in countries where
obesity causes more deaths than underweight
(1)
. Although it
is a difficult task, intensive lifestyle interventions can achieve
weight loss that is sustained over the long term, as shown
by the findings of a recent large clinical trial
(2)
.
Diet is a cornerstone of any lifestyle intervention programme.
The dietary plan that restricts energy and fat is the most common
strategy, and based on it, several other dietary strategies have
been proposed
(3 – 5)
. The very-low-carbohydrate ketogenic
diet (VLCKD) differs from these approaches. According to
Accurso et al.
(6)
, in the early phases of this therapy, individuals
must have approximately 50 g carbohydrates/d or 10 % of
energy from a nominal 8400 kJ (approximately 2000 kcal) diet,
unlike low-carbohydrate diets, which may have up to 130 g
carbohydrates/d or 26 % of energy from a nominal diet.
A major concern regarding the prescription of the VLCKD is
the adherence of the individuals assigned to it, since it promotes
important lifestyle changes
(7)
.
Given the importance of dietary counselling in weight
loss, it is useful to investigate the effectiveness of different
dietary therapies. A recent large randomised clinical trial,
which assigned individuals to diets ranging from 35 to 65 %
of dietary carbohydrate content, showed that, at this level of
carbohydrate intake, there is no difference in weight loss
between interventions
(8)
. Nonetheless, evidence suggests
*Corresponding author: N. B. Bueno, email nassibbb@hotmail.com
Abbreviations: DBP, diastolic blood pressure; HDL-C, HDL-cholesterol; LDL-C, LDL-cholesterol; LFD, low-fat diet; SBP, systolic blood pressure; VLCKD,
very-low-carbohydrate ketogenic diet; WMD, weighted mean differences.
British Journal of Nutrition (2013), 110, 1178–1187 doi:10.1017/S0007114513000548
qThe Authors 2013
British Journal of Nutrition
that greater dietary carbohydrate restrictions lead to greater
weight loss
(9)
. Indeed, previous meta-analyses have shown
that carbohydrate-restricted diets promote greater weight
loss than conventional energy-restricted low-fat diets
(LFD)
(10,11)
. However, these analyses did not exclusively
focus on VLCKD studies
(10)
, or included mostly trials with
6 months of follow-up
(11)
; hence, these analyses do not
guarantee the long-term effectiveness of the VLCKD.
A recent meta-analysis by Santos et al.
(12)
reported that
low-carbohydrate diets lead to significantly favourable changes
in body weight and major cardiovascular risk factors. Never-
theless, this analysis was based only on the individuals
who had adopted a low-carbohydrate diet, comparing final
values against baseline values. Although it was an important
investigation, the question of whether an abrupt change to an
individual’s lifestyle, such as the adoption of a VLCKD, leads to
relevant long-term clinical improvements remains unanswered.
Thus, the present meta-analysis evaluated randomised
controlled trials to determine whether overweight and obese
individuals assigned to a VLCKD achieve greater weight loss
and manage cardiovascular risk factors more effectively than
those assigned to a LFD over the long term (defined as
12 months or more post-intervention).
Methods
The present meta-analysis is reported in accordance with the
Preferred Reporting Items for Systematic Reviews and Meta-
Analyses (PRISMA) Statement
(13)
. The protocol was previously
published in the PROSPERO database (http://www.crd.york.
ac.uk/PROSPERO), under registration no. CRD42012002408.
Search strategy
The following databases were searched until August 2012:
MEDLINE, CENTRAL, ScienceDirect, Scopus, LILACS, SciELO
and ClinicalTrials.gov. In addition, the following grey litera-
ture databases were searched: OpenGrey.eu, DissOnline.de,
NYAM.org and ClinicalEvidence.com. There was no manual
search of the included articles, and no specialists in the field
were contacted to avoid the risk of citation bias
(14)
. The search
strategy included terms related to the intervention (VLCKD),
the primary outcome (weight loss) and the secondary outcomes
(cardiovascular risk factors), as well as related terms designed
to improve the sensitivity of a search for randomised controlled
trials
(15)
. The search was not restricted to any particular years
of publication or languages. The complete search strategy is
shown in the Supplementary material (available online).
Eligibility criteria
Only randomised controlled trials that met the following
criteria were included: (1) the study participants were indivi-
duals older than 18 years old who were assigned to a LFD
(i.e. a restricted-energy diet with less than 30 % of energy
from fat) or to a VLCKD (i.e. a diet with no more than 50 g
carbohydrates/d or 10 % of daily energy from carbohydrates);
(2) the follow-up period was 12 months or more; (3) the
participants had a mean BMI greater than 27·5 kg/m
2
. The
third criterion allowed the inclusion of studies of populations
who are already at high risk beyond this BMI threshold
(16)
.
The present analysis aimed to evaluate the differences
in the outcomes of the prescribed diets, without addressing
individual adherence to the diets. There were no restrictions
based on sex, race or co-morbidities. At a minimum, the
studies must have assessed weight loss as an outcome and
must have reported mean values or the differences between
the mean values. The exclusion criteria were as follows:
(1) studies with a concomitant pharmacological intervention
and (2) duplicate publications of the included trials.
Data extraction
The titles and abstracts of the retrieved articles were evaluated
independently by two investigators who were not blinded to
the authors or the journal titles. The full-text versions of poten-
tially eligible articles were retrieved for further evaluation.
The primary outcome sought in the studies was the
mean change between the baseline body weight and the
final body weight (in kg), with the associated measure of
dispersion. The secondary outcomes were the mean changes
between the baseline and final values (with the associated
measures of dispersion) for TAG (in mg/dl (to convert to
mmol/l, multiply by 0·0113)), HDL-cholesterol (HDL-C) and
LDL-cholesterol (LDL-C) (in mg/dl (to convert to mmol/l, mul-
tiply by 0·0259)), fasting blood glucose (in mg/dl (to convert
to mmol/l, multiply by 0·0555)), insulin (in mU/ml (to convert
to pmol/l, multiply by 6·945)), C-reactive protein (in mg/l
(to convert to nmol/l, multiply by 9·524)), HbA
1c
(percentage),
and systolic and diastolic blood pressure (SBP and DBP,
respectively, in mmHg).
All the necessary information was extracted from the pub-
lished articles, protocols and commentaries related to each
study, and when necessary, the authors were contacted to
obtain additional information. For the studies that had more
than two experimental groups, the most suitable one was
chosen. Any disagreements were resolved by consensus.
A standard form for storing data was created based on the
Cochrane Collaboration model
(17)
.
Assessment of risk of bias
Risk of bias was evaluated according to the Cochrane
Handbook recommendations
(18)
, at the primary outcome
level. The quality of the studies were assessed by two investi-
gators independently in five categories: adequate sequence
generation; allocation concealment; blinding of the outcome
assessors; handling of missing data (intention-to-treat or per-
protocol analysis); selective outcome reporting. The nature
of the trials required an open intervention with no blinding
of the trial participants or the investigators.
Data analysis
The absolute changes for each outcome, reported as the
differences between the final and baseline mean values,
Carbohydrate restriction for weight loss 1179
British Journal of Nutrition
were analysed. The treatment effects across the trials were
pooled, and weighted mean differences (WMD) for the out-
come measures were calculated. The study weights were
assigned by using the inverse variance method
(19)
, and the
calculations were performed using a random-effects model
(20)
.
An
a
value of 0·05 was considered to be statistically signi-
ficant. When it was not possible to retrieve adequate data,
imputations were performed
(21)
. These imputations are
shown in the Supplementary material (available online).
Statistical heterogeneity among the studies was tested using
the Cochran Qtest, and inconsistency was tested using the
I
2
test. A Pvalue less than 0·10 was considered to be statis-
tically significant. Whenever a result showed heterogeneity,
it was explored in three different ways. First, each analysis
was repeated, removing each study one at a time in order to
assess whether a particular study explained the heterogeneity.
Second, univariate meta-regressions were performed to ana-
lyse whether methodological covariates were influencing
the results
(22)
. The covariates included the risk of bias in the
study, adequate nutritional counselling of the individuals
(studies that included individual or group meetings with a
dietitian at least bimonthly until the end of the follow-up
period were considered as adequate), the use of an inten-
tion-to-treat analysis, the study follow-up length in months
and the presence of co-morbidities in the inclusion criteria
for the participants in each study. Thereafter, it was planned
to perform a multivariate meta-regression including all covari-
ates that had a Pvalue less than 0·10 in the univariate analysis.
Finally, subgroup analyses were performed on studies that
shared certain methodological features, including studies
with a low risk of bias, studies using an intention-to-treat
analysis and studies with 24 months of follow-up. Subgroup
analyses were conducted regardless of heterogeneity.
Contour-enhanced funnel plots
(23)
were created and
Egger’s test
(24)
was performed to evaluate publication bias;
Pvalues less than 0·10 were considered to be statistically signi-
ficant. All analyses were conducted using Stata software 9.0
(StataCorp). Graphs were plotted using RevMan 5.4 (Cochrane
Collaboration).
Results
Included studies
From 3123 potentially relevant records identified by searching
the databases, twenty-five full-text publications met the
inclusion criteria and were retrieved for further assessment.
From these, eleven were excluded after the full-text analysis,
leaving fourteen full texts included in the qualitative and
quantitative analysis (Table 1). The flow diagram illustrating
the search and selection of studies is shown in Fig. 1. Reasons
for exclusion are shown in the Supplementary material
(available online).
From the fourteen full-text articles included, the report by
Vetter et al.
(25)
had characteristics that were unexpected and
not mentioned in the inclusion or exclusion criteria for the
review. This report describes a body weight analysis of the
individuals included in the study by Stern et al.
(26)
, conducted
36 months after randomisation. Nevertheless, follow-up
ceased after 12 months; thus, it was not possible to assess
whether the individuals continued with the intervention in
the period after follow-up, so the data from this full-text article
were included in a sensitivity analysis.
In total, thirteen studies were included in the quantitative
analysis, with a total of 1577 individuals randomised to a
condition (787 to a LFD group and 790 to a VLCKD group).
From these, six studies had more than two intervention
groups, and it was determined by consensus which groups
fit best in the analysis. Intervention groups of all studies are
shown in the Supplementary material (available online).
Assessment of risk of bias
The risk of bias in the studies at the primary outcome level is
shown in Table 2. In the final result, nine from the thirteen
included studies were assessed as having a low risk of bias.
Of these nine studies, two did not report the sequence
generation method used, while seven did not report using
any measure to conceal the allocation. All the nine studies
did not report blinding of the outcome assessors, but as all
Table 1. Characteristics of the included studies
Source
Duration
(months)
Dietary
counselling
Dropouts
(n/N)
Females
(%) Country Risk factor
Mean age
(years)
Mean BMI
(kg/m
2
)
CHO intake/d
(VLCKD)*
Brinkworth et al.
(28)
12 Adequate 38/107 70 Australia CV risk factor 50·6 33·6 36 g
Dansinger et al.
(50)
12 Inadequate 41/80 47 USA CV risk factor 47 35 190 g
Davis et al.
(51)
12 Adequate 14/105 78 USA T2D 53·5 35·9 33 %
Dyson et al.
(52)
24 Inadequate 4/26 73 UK T2D 52 35·1 Unreported
Foster et al.
(53)
12 Inadequate 37/63 68 USA None 44·9 34·1 Unreported
Foster et al.
(27)
24 Adequate 113/307 68 USA None 45·5 36·1 Unreported
Gardner et al.
(30)
12 Inadequate 26/153 100 USA None 42 32 34 %
Iqbal et al.
(29)
24 Adequate 76/144 10 USA T2D 60 37·4 47 %
Lim et al.
(54)
15 Inadequate 25/60 80 Australia CV risk factor 48·4 31·4 36 %
McAuley et al.
(55)
12 Inadequate 15/63 100 NZ None 45 36·1 33 %
Shai et al.
(49)
24 Adequate 44/213 16 Israel CV risk factor 51·5 30·7 40%
Stern et al.
(26)
12 Adequate 45/132 17 USA None 53·5 42·9 120 g
Truby et al.
(56)
12† Inadequate 98/116 72 UK None 39·8 32 Unreported
CHO, carbohydrate; VLCKD, very-low-carbohydrate ketogenic diet; CV, cardiovascular; T2D, type 2 diabetes mellitus; NZ, New Zealand.
* Mean carbohydrate intake in the VLCKD group at the end of the follow-up, measured by dietary assessment, shown as g/d or percentage of energy from carbohydrates per d.
† Truby et al.
(56)
assessed only the body weight at 12 months.
N. B. Bueno et al.1180
British Journal of Nutrition
the outcomes are objective, it is unlikely that this domain
affected the results of the trials. Regarding the handling of
missing data, five studies were categorised as having a high
risk of bias because they utilised a per-protocol analysis.
There was no evidence of selective outcome reporting.
Data analysis
Body weight. All the thirteen included studies (1415
patients) were assessed (Fig. 2(a)). The individuals assigned
to a VLCKD achieved a significantly greater weight loss com-
pared with the individuals assigned to a LFD (WMD 20·91
(95 % CI 21·65, 20·17) kg, P¼0·02; I
2
¼0%, P¼0·47). This
result was consistent across all subgroup analyses, except
for the subgroup of studies with 24 months of follow-up
(data not shown). The substitution of the data from Stern
et al.
(26)
for the data from Vetter et al.
(25)
changed the results
(WMD 20·73 (95 % CI 21·52, 0·06) kg, P¼0·07; I
2
¼5%,
P¼0·39). There was no evidence of publication bias
(P¼0·34). The contour-enhanced funnel plots for body
weight and all other outcomes are shown in the Supplemen-
tary material (available online).
TAG . In total, twelve studies (1258 patients) were assessed
(Fig. 2(b)). The individuals assigned to a VLCKD showed
a significantly greater reduction in TAG than the indivi-
duals assigned to a LFD (WMD 20·18 (95 % CI 20·27,
20·08) mmol/l, P,0·001; I
2
¼12 %, P¼0·33). This result
was consistent across all subgroup analyses, except for the
subgroup of studies with 24 months of follow-up (data not
shown). Heterogeneity was reversed when the study by
Foster et al.
(27)
was excluded, and also when the study
by Stern et al.
(26)
was excluded, but there were no statistically
significant changes in the results. The evidence of publication
bias (P¼0·04) was also reversed with the exclusion of
both aforementioned studies. The meta-regression analysis
showed that the covariate ‘study follow-up length’ affected
the results significantly (r
2
87·19 %, P¼0·09; Table 3).
HDL-cholesterol. Overall, twelve studies (1257 patients)
were assessed (Fig. 2(c)). The individuals assigned to a
VLCKD achieved a significantly greater increase in their
Records identified through
database searching
(
n
3378)
IdentificationScreeningEligibilityIncluded
Additional records identified
through other sources
(
n
0)
Records after duplicates
removed
(
n
3123)
Records excluded (
n
3098)
Not a randomised controlled
trial (
n
1389)
Not a dietary intervention for
weight loss (
n
1013)
Inadequate carbohydrate
content (
n
580)
Participants younger than 18
years old (
n
11)
Less than 12 months of
follow-up (
n
59)
Did not report the necessary
results (
n
46)
Full-text articles excluded
(
n
11)
Did not report the necessary
results (
n
3)
Less than 12 months of
follow-up (
n
3)
Inadequate carbohydrate
content (
n
3)
Concomitant drug
intervention (
n
1)
Repeated results (
n
1)
Records screened
(
n
3123)
Full-text articles assessed
for eligibility
(
n
25)
Studies included in
qualitative synthesis
(
n
13)
Studies included in
quantitative synthesis
(meta-analysis)
(
n
13)
Fig. 1. Flow diagram of the study selection.
Carbohydrate restriction for weight loss 1181
British Journal of Nutrition
HDL-C levels compared with the individuals assigned to a LFD
(WMD 0·09 (95 % CI 0·06, 0·12) mmol/l, P,0·001; I
2
¼9%,
P¼0·36). All the subgroups showed the same result (data
not shown). The study by Brinkworth et al.
(28)
and the study
by Iqbal et al.
(29)
were each individually responsible for the
heterogeneity in the overall analysis, and the stepwise exclu-
sion of both studies did not change the main result (data
not shown). In the meta-regression analysis, only the covariate
‘study follow-up length’ significantly affected the results
(r
2
¼100 %, P¼0·03; Table 3). There was no evidence of pub-
lication bias (P¼0·53).
LDL-cholesterol. A total of twelve studies (1255 patients)
were assessed (Fig. 2(d)). The individuals assigned to a
VLCKD achieved a significantly greater increase in their
LDL-C levels compared with the individuals assigned to a
LFD (WMD 0·12 (95 % CI 0·04, 0·2) mmol/l, P¼0·002;
I
2
¼0%, P¼0·7). The subgroup of studies with 24 months
of follow-up was the only subgroup that showed different
results (data not shown). There was no evidence of publi-
cation bias (P¼0·42).
Systolic and diastolic blood pressure. Overall, eleven
studies (1298 patients) were included in the SBP (Fig. 3(A))
and DBP analyses (Fig. 3(B)). There were no differences in
SBP between the groups (WMD in favour of the VLCKD
21·47 (95 % CI 23·44, 0·50) mmHg, P¼0·14; I
2
¼33 %,
P¼0·13), a result that held in the subgroup analyses. However,
individuals assigned to a VLCKD had a significantly greater
reduction in DBP than the individuals assigned to a LFD
(WMD 21·43 (95 % CI 22·49, 20·37) mmHg, P¼0·008;
I
2
¼3%, P¼0·41).
The sensitivity analysis for SBP showed that the study by
Gardner et al.
(30)
was responsible for the heterogeneity, and
its exclusion did not change the results (data not shown).
The covariate ‘adequate nutritional counselling’ significantly
affected the SBP results (r
2
¼79·7 %, P¼0·05; Table 3).
Due to the extremely low heterogeneity, neither a sensitivity
analysis nor a meta-regression analysis was undertaken for
DBP, and only the subgroup of studies with 24 months of
follow-up showed different results (data not shown). There
was no evidence of publication bias for SBP (P¼0·79), but
the DBP analysis showed statistically significant publication
bias (P¼0·04), which was not reversed by the exclusion of
any study.
Fasting blood glucose, insulin, HbA
1c
and C-reactive
protein. These analyses were performed in less than ten
studies; thus, no sensitivity, subgroup, meta-regression and
publication bias analyses were conducted. None of these
analyses showed statistically significant results. The forest
plots for these analyses are shown in the Supplementary
material (available online). For the fasting blood glucose anal-
ysis, eight studies (770 patients) were assessed (WMD in
favour of the VLCKD 20·08 (95 % CI 20·18, 0·02) mmol/l,
P¼0·11; I
2
¼0%,P¼0·88). For the insulin analysis, six studies
(584 patients) were assessed (WMD in favour of the VLCKD
25·52 (95 % CI 213·62, 2·57) pmol/l, P¼0·18; I
2
¼26 %,
P¼0·24). For the HbA
1c
analysis, four studies (319 patients)
were assessed (WMD in favour of the VLCKD 20·24 (95 %
CI 20·55, 0·06) %, P¼0·12; I
2
¼0%, P¼0·59). Finally, for
the C-reactive protein analysis, four studies (355 patients)
were also assessed (WMD in favour of the VLCKD 21·85
(95 % CI 26·66, 2·96) nmol/l, P¼0·45; I
2
¼0%, P¼0·55).
Discussion
The present meta-analysis showed that individuals assigned to
a VLCKD achieve greater reductions in body weight, TAG
and DBP, but they also demonstrate a greater increase in
LDL-C and HDL-C levels over a treatment follow-up period
of 12 months or more, compared with individuals assigned
to a LFD. Only the change in HDL-C levels retained statistical
significance in the subgroup analysis of studies with 24
months of follow-up; however, it is important to note that
this analysis included only four studies. Low risk of bias was
not unanimous, although this characteristic did not influence
any of the results, since potential bias was explored by con-
ducting subgroup and meta-regression analyses. Also, studies
that included individuals with co-morbidities were not sources
of heterogeneity. Furthermore, only the TAG and the DBP
analyses revealed evidence of publication bias.
With regard to the primary outcome, the present findings
are similar to the findings of previous meta-analyses
(10,11)
.
The supposed beneficial effect of a VLCKD on body weight
may be due to the modulation of resting energy expenditure.
Table 2. Risk of bias of the included studies
Source
Sequence
generation
Allocation
concealment Blinding
Missing
data
Selective
report Overall
Brinkworth et al.
(28)
Low High Unclear Low Low High
Dansinger et al.
(50)
Low Low Low Low Low Low
Davis et al.
(51)
Low High Unclear Low Low High
Dyson et al.
(52)
Low Low Low High Low Low
Foster et al.
(53)
Low Unclear Unclear Low Low Low
Foster et al.
(27)
Low Unclear Unclear Low Low Low
Gardner et al.
(30)
Low Low Low Low Low Low
Iqbal et al.
(29)
Low Unclear Unclear Low Low Low
Lim et al.
(54)
Unclear Unclear Unclear High Low High
McAuley et al.
(55)
Low Low Unclear High Low Low
Shai et al.
(49)
Low Unclear Low Low Low Low
Stern et al.
(26)
Low Unclear Unclear Low Low Low
Truby et al.
(56)
Low Unclear High High Low High
N. B. Bueno et al.1182
British Journal of Nutrition
(a)
(b)
(c)
(d)
712 703 –0·91(–1·65, –0·17)
Favours VLCKD Favours LFD
631 0·12 (0·04, 0·20)
–1 –0·5 0 0·5 1
Favours VLCKD Favours LFD
–0·2 –0·1 0 0·1 0·2
Favours VLCKDFavours LFD
–0·5 –0·25 0 0·25 0·5
Favours VLCKD
Favours LFD
625
LDL-C (mmol/l)
Study Mean, SD and total Mean, SD and total
VLCKD LFD
Weight (%)
Mean difference
(95 %CI)
Mean difference
IV, random, 95 %CI
Heterogeneity: t2= 0·00; c2=11·72, df= 12 (
P
= 0·47);
I
2=0 %
Test for overall effect:
Z
=2·42 (
P
=0·02)
Heterogeneity: t2= 0·00; c2=12·44, df= 11 (
P
=0·33);
I
2=12 %
Test for overall effect:
Z
=3·68 (
P
=00002)
Heterogeneity: t2= 0·00; c2=12·09, df= 11 (
P
=0·36);
I
2=9 %
Test for overall effect:
Z
=6·40 (
P
=0·00001)
Heterogeneity: t2= 0·00; c2=8·16, df= 11 (
P
= 0·70);
I
2=0 %
Test for overall effect:
Z
=3·02 (
P
= 0·002)
–13·1 11·86 55 –11·6 52 –1·50 (–5·93, 2·93)11·53
1·20 (–1·51, 3·91)7·3 40–3·3404·8–2·1
1·10 (–3·64, 5·84)3·97 11–0·8116·960·3
0·00 (–2·05, 2·05)5·8 50–3·1554·8–3·1
–2·80 (–6·53, 0·93)830–4·4337–7·2
–1·03 (–1·41·3·47)10·98 15 4–7·3715310·82–6·34
–1·30 (–3·99, 1·39)7·74 71–0·2678·36–1·5
–0·80 (–3·98, 2·38)4·7 18–2·1174·9–2·9
–1·00 (–8·02, 8·02)12·2 24–4·42412·6–5·4
–2·00 (–4·99, 0·99)8·4 64–3·1628·7–5·1
–1·70 (–3·16, 6·56)6·2 9–10·794·1–9
–2·10 (–4·13, 0·07)5·55 76–2·6777·16–4·7
–1·80 (–3·26, –0·34)4·2 104–2·91096·5–4·7
–0·2 0·57 10 0 11 –0·20 (–0·82, 0·42)0·87
–0·58 0·63 33 –0·22 36 –0·36 (–0·66, –0·06)0·66
–0·15 0·88 55 –0·01 50 –0·14 (–0·47, 0·19)0·86
–0·42 0·4 33 –0·02 30 –0·40 (–0·69, –0·11)0·7
–0·29 1·15 67 –0·15 71 –0·14 (–0·50, 0·22)1
–0·66 1·79 44 0·05 43 –0·71 (–1·31, –0·11)0·96
–0·47 0·69 24 –0·31 24 –0·16 (–0·52, 0·20)0·56
–0·25 0·96 81 –0·03 90 –0·22 (–0·50, 0·06)0·9
0·3 0·4 33 0·07 36 0·23 (0·05, 0·41)0·36
0·16 0·27 55 0·06 50 0·10 (0·01, 0·19)0·21
0·18 0·22 10 0·09 11 0·09 (–0·14, 0·32)0·32
0·22 0·31 33 0·04 30 0·18 (0·05, 0·31)0·21
0·02 0·28 67 0·02 71 0·0 0 (–0·09, 0·09)0·24
–0·18 0·62 40 –0·33 40 0·15 (–0·09, 0·39)0·49
–0·04 0·63 55 –0·13 50 0·14 (–0·11, 0·39)0·66
0 0·51 10 –0·4 11 0·40 (–0·15, 0·95)0·77
0·02 0·7 33 –0·18 30 0·20 (–0·10, 0·50)0·49
–0·21 1·29 67 –0·16 71 –0·05 (–0·47, 0·37)1·2
–0·3 0·7 17 –0·3 18 0·00 (–0·46, 0·46)0·7
–0·1 1·07 24 –0·1 24 0·00 (–0·48, 0·48)0·58
–0·08 1·02 81 0 90 –0·08 (–0·37, 0·21)0·89
0·18 0·91 42 –0·1 43 0·28 (–0·07, 0·63)0·75
0·1 0·3 17 0·1 18 0·00 (–0·20, 0·20)0·3
0·6 1·15 33 0·1 36 0·50 (0·06, 0·94)0·6
0·12 0·28 24 –0·02 24 0·14 (–0·00, 0·28)0·23
0·22 0·28 81 0·17 90 0·05 (–0·03, 0·13)0·24
–0·03 0·18 44 –0·13 43 0·10 (0·03, 0·17)0·14
0·09 0·18 40 –0·01 40 0·10 (0·02, 0·18)0·17
–0·01 0·95 40 0·069 40 –0·07 (–0·39, 0·25)0·41
–0·2 0·7 17 0·1 18 –0·30 (–0·83, 0·23)0·9
–0·14 0·76 153 –0·16 154 0·02 (–0·15, 0·19)0·8
–0·33 0·67 70 –0·17 64 –0·16 (–0·36, 0·04)0·52
0·2 0·29 153 0·12 154 0·08 (0·02, 0·14)0·24
–0·12 0·72 153 –0·21 154 0·09 (–0·05, 0·23)
0·56
0·13 0·24 70 0 64 0·13 (0·06, 0·20)0·16
0·02 0·59 70 –0·1 64 0·12 (–0·06, 0·30)0·49
627 631 –0·18(–0·27, –0·08)
627 631
100· 0
100· 0
2·8
7·4
2·4
13·0
3·9
9·2
7·5
5·4
1·1
6·1
2·3
13·3
25·5
2·2
8·4
7·2
9·4
6·2
2·4
6·4
9·8
2·4
8·5
1·4
4·5
9·3
10·1
9·9
2·0
6·9
3·5
2·8
2·6
7·3
4·8
2·0
3·1
3·6
11· 2
13·2
11· 7
7·7
3·0
20·7
16·7
17·5
29·0
14·2
18·0
100· 0
100·0 0·09 (0·06, 0·12)
HDL-C (mmol/l)
Study Mean, SD and total Mean, SD and total
VLCKD LFD
Weight (%)
Mean difference
(95 %CI)
Mean difference
IV, random, 95 %CI
TAG (mmol/l)
Study Mean, SD and total Mean, SD and total
VLCKD LFD
Weight (%)
Mean difference
(95 %CI)
Mean difference
IV, random, 95 %CI
Body weight (kg)
Study Mean, SD and total Mean, SD and total
VLCKD LFD
Weight (%)
Mean difference
(95 %CI)
Mean difference
IV, random, 95 %CI
–4 –2 0 2 4
Total (95 % CI)
Total (95 % CI)
Brinkworth
et al
. (28)
Dansinger
et al
. (50)
Dyson
et al
. (52)
Davis
et al
. (51)
Foster
et al
. (53)
Foster
et al
. (27)
Iqbal
et al
. (29)
Lim
et al
. (54)
McAuley
et al
. (55)
Stern
et al
. (26)
Truby
et al
. (56)
Lim
et al
. (30)
Shai
et al
. (49)
Dyson
et al
. (52)
Brinkworth
et al
. (28)
Davis
et al
. (51)
Foster
et al
. (53)
Iqbal
et al
. (29)
Stern
et al
. (26)
McAuley
et al
. (55)
Shai
et al
. (49)
Brinkworth
et al
. (28)
Davis
et al
. (51)
Dyson
et al
. (52)
Foster
et al
. (53)
Iqbal
et al
. (29)
Dansinger
et al
. (50)
Davis
et al
. (51)
Dyson
et al
. (52)
Foster
et al
. (53)
Iqbal
et al
. (29)
Lim
et al
. (54)
McAuley
et al
. (55)
Shai
et al
. (49)
Stern
et al
. (26)
Lim
et al
. (54)
Brinkworth
et al
. (28)
McAuley
et al
. (55)
Shai
et al
. (49)
Stern
et al
. (26)
Dansinger
et al
. (50)
Dansinger
et al
. (50)
Lim
et al
. (54)
Foster
et al
. (27)
Gardner
et al
. (30)
Foster
et al
. (27)
Foster
et al
. (27)
Gardner
et al
. (30)
Gardner
et al
. (30)
Total (95 % CI)
Total (95 % CI)
Fig. 2. Absolute changes in (a) body weight, (b) TAG, (c) HDL-cholesterol (HDL-C) and (d) LDL-cholesterol (LDL-C). VLCKD, very-low-carbohydrate ketogenic
diet; LFD, energy-restricted low-fat diet.
Carbohydrate restriction for weight loss 1183
British Journal of Nutrition
Under isoenergetic conditions, Ebbeling et al.
(31)
found that a
carbohydrate-restricted diet is better than a LFD for retaining
an individual’s BMR. In addition, Westman et al.
(32)
hypoth-
esised that a VLCKD reduces insulin levels, which would
explain the satietogenic effects of this diet. This hypoinsuli-
naemic effect of the VLCKD was not evidenced in this analysis.
TAG decreased significantly in individuals assigned to a
VLCKD. The heterogeneity in the analysis and the evidence
of publication bias were entirely attributable to the study by
Foster et al.
(27)
, which was the only study to present neutral
results in this analysis. On the other hand, individuals assigned
to a VLCKD showed significantly increased levels of both
LDL-C and HDL-C levels. As discussed by Volek et al.
(33)
, the
preservation of the circulating HDL-C and the hypotriacylgly-
cerolaemic effect of a VLCKD might be explained by the
reduction in the dieting individuals’ postprandial lipaemia.
Conversely, the increase in LDL-C concentration associated
with the VLCKD is an expected finding that is attributable to
the increase in saturated fat intake. However, this finding war-
rants further investigation. Krauss et al.
(34)
showed that high
fat intake, combined with carbohydrate restriction, raises the
levels of larger-sized LDL-C, which are known to be less
atherogenic than the small, dense LDL-C
(35)
.
There was also evidence that individuals assigned to a
VLCKD showed a significantly greater reduction in DBP.
Hession et al.
(10)
analysed five studies and found that
carbohydrate-restricted diets only influenced SBP. Usually,
hypertension is attributable to obesity and Na intake, but
Appel et al.
(36)
showed that substituting carbohydrates for
proteins and monounsaturated fats may decrease blood press-
ure beyond the decrease expected with Na restriction alone.
It is remarkable to note that although five outcomes
demonstrated statistical significance, these findings must be
carefully interpreted regarding its clinical significance
(37)
. For
example, a typical 1·70 m-tall adult with a BMI of 30 kg/m
2
weighs 87 kg; hence, a weight loss of 0·91 kg, as observed
here, would represent only 1·04 % of the initial body weight.
However, large randomised clinical trials with long-term diet-
ary interventions aiming weight loss showed that individuals
under intensive lifestyle interventions lose about 4·8 kg
(2,38)
.
Hence, the further reduction of 0·9 kg in the individuals
assigned to a VLCKD would represent almost 20 % of the
awaited weight loss achieved with long-term dietary inter-
ventions. Additionally, if we assume the cut-off points of
the metabolic syndrome
(39)
, similar percentages would be
found regarding the other outcomes. The extra reduction of
1·43 mmHg in DBP achieved by individuals assigned to a
VLCKD is similar to the reductions promoted by other dietary
interventions, such as Mg supplementation
(40)
or consumption
of flavonol-rich products
(41)
.
Undoubtedly, the present findings demonstrate that a
VLCKD has favourable effects on body weight and some cardi-
ovascular risk factors, as stated by Santos et al.
(12)
; however, in
the long term and when compared with conventional therapy,
the differences appear to be of little clinical significance,
although statistically significant. Healthcare professionals
should weigh the advantages and disadvantages of recom-
mending a VLCKD and consider their patients’ will power,
since this therapy prominently alters an individual’s daily habits.
The present meta-analysis has several limitations. First, it
used aggregated data from the studies instead of individual
patient data. Second, only blood risk factors were assessed,
neglecting important pathological markers such as hepatic
lipid infiltration
(42)
, endothelial function
(43)
, general cardio-
vascular events
(44)
and renal function
(45)
, which are important
in assessing the safety of dietary therapies. Third, the adher-
ence to the VLCKD in the included studies was low
(Table 1). At the end of the follow-up period in most studies,
carbohydrate intake was higher than the protocol allowed.
However, in most cases, there was good adherence in the
Table 3. Meta-regression analysis
(Coefficients and 95 % confidence intervals)
Covariates Coefficient 95 % CI Adj r
2
(%) P
Adequate nutritional counselling
TAG (mmol/l) 4·051 217·008, 25·110 29·67 0·677
HDL-C (mmol/l) 21·645 24·030, 0·740 100·00 0·155
Systolic blood pressure (mmHg) 3·582 20·167, 7·333 79·79 0·059
Co-morbidities
TAG (mmol/l) 22·498 223·740, 18·742 217·52 0·799
HDL-C (mmol/l) 21·542 23·852, 0·768 100·00 0·168
Systolic blood pressure (mmHg) 2·309 22·063, 6·682 28·18 0·263
Intention-to-treat analysis
TAG (mmol/l) 13·173 28·286, 34·632 34·23 0·201
HDL-C (mmol/l) 20·406 23·233, 2·421 2644·4 0·755
Systolic blood pressure (mmHg) 20·946 26·441, 4·548 213·25 0·706
Length of the follow-up
TAG (mmol/l) 1·259 20·261, 2·781 87·19 0·095
HDL-C (mmol/l) 20·208 20·404, 20·013 100·00 0·039
Systolic blood pressure (mmHg) 0·094 20·318, 0·506 21·55 0·619
Low risk of bias
TAG (mmol/l) 13·494 27·667, 34·656 33·87 0·186
HDL-C (mmol/l) 0·117 22·610, 2·845 2510·1 0·925
Systolic blood pressure (mmHg) 21·821 27·270, 3·627 26·98 0·469
Adj r
2
, adjusted r
2
; HDL-C, HDL-cholesterol.
N. B. Bueno et al.1184
British Journal of Nutrition
short term, which may explain why meta-analyses of 6-month
studies show more impressive results than meta-analyses of
longer-term studies, like the present analysis. Greenberg
et al.
(46)
found that among dieters, the initial weight reduction
in the first 6 months is the main predictor of both long-term
retention and success in weight loss, which may explain the
statistically significant differences observed here.
The Cochrane risk of bias tool was used in the present
meta-analysis. Despite being the most recommended tool to
assess the risk of bias in randomised controlled trials, it may
face some limitations when assessing behavioural or lifestyle
interventions, such as dietary ones
(47)
. These interventions
are usually complex, i.e. have multiple components, which
deem its fidelity (the extent to which the intervention has
been delivered as planned) an important issue to be
assessed
(48)
. Since the risk of bias tool does not directly
address fidelity, it may be difficult to distinguish between an
ineffective intervention and a failed implementation
(47)
.
Upcoming trials should focus on dietary adherence, imple-
menting measures to ensure that individuals adhere to the
protocol, as was done by some of the included studies
(28,49)
,
permitting better investigation of the long-term effects of a
VLCKD. Nevertheless, it is necessary to consider the feasibility
of such measures, like those applied by Shai et al.
(49)
, where
the investigators managed the lunches of all individuals, in a
real-life scenario.
In conclusion, the present meta-analysis demonstrates that
individuals assigned to a VLCKD achieve significantly greater
long-term reductions in body weight, diastolic blood pressure
and TAG, as well as greater LDL and HDL increases when
compared with individuals assigned to a LFD; hence, the
VLCKD may be an alternative tool against obesity. Investi-
gations beyond that of blood cardiovascular risk factors
merit further study.
Supplementary material
To view supplementary material for this article, please visit
http://dx.doi.org/10.1017/S0007114513000548
Acknowledgements
The present study was partially supported by Conselho Nacio-
nal de Pesquisa e Desenvolvimento Cientı
´fico e Tecnolo
´gico
(CNPq) grant 130639/2011-7, by means of a studentship to
N. B. B. The sponsors of the study had no role in the study
(a) SBP (mmHg)
Study
Brinkworth
et al.
(28)
Dansinger
et al.
(50)
Davis
et al.
(51)
Foster
et al.
(53)
Foster
et al.
(27)
Gardner
et al.
(30)
Iqbal
et al.
(29)
Lim
et al.
(54)
McAuley
et al.
(55)
Shai
et al.
(49)
Stern
et al.
(26)
Brinkworth
et al.
(28)
Dansinger
et al.
(50)
Davis
et al.
(51)
Foster
et al.
(53)
Foster
et al.
(27)
Gardner
et al.
(30)
Iqbal
et al.
(29)
Lim
et al.
(54)
McAuley
et al.
(55)
Shai
et al.
(49)
Stern
et al.
(26)
Total (95 % CI)
Test for overall effect:
Z
= 1·46 (
P
= 0·14)
Total (95 % CI)
Test for overall effect:
Z
= 2·64 (
P
= 0·008)
(b) DBP (mmHg)
–13·8
Mean, SD and total Mean, SD and total
14·36
12
15·6
15
15·17
11
30·28
10·6
15·9
12·8
19 44
109
24
17
67
77
153
33
55
40
33
44
652
109
24
17
67
77
153
33
55
40
3319·9–6·3
–1·4
–2·9
–4·6
–3·19
–4·4
–3·8
–6·6
–4
–0·8
3
7·5
9·4
12
9·24
8·4
19·64
12·1
9·81
8·7
15
LFDVLCKD
Mean, SD and total Mean, SD and total
LFDVLCKD
12 36 7·5
Weight (%)
Weight (%)
Mean difference
(95 % CI)
Mean difference
(95 % CI)
Mean difference
IV, random, 95 % CI
Mean difference
IV, random, 95 % CI
0·80 (–5·47, 7·07)
–0·30 (–4·72, 4·12)
3·80 (–3·70, 11·30)
–5·50 (–14·30 3·30)
–0·21 (–3·66, 3·24)
–5·70 (–8·71 –2·69)
–6·70 (–16·29 –2·89)
–4·60 (–12·55 3·35)
1·00 (–6·49, 8·49)
1·40 (–2·90, 3·70)
–1·00 (–8·18, 6·18)
–1·47 (–3·44, 0·50)
–10
Favours VLCKD Favours LFD
Favours VLCKD Favours LFD
–5
–4 –2 0 2 4
0105
1·60 (–5·88, 9·08)
1·60 (–5·33, 1·13)
1·70 (–5·76, 3·36)
1·60 (–5·60, 6·80)
–2· 69 (–4·88, –0·50)
–3·70(–6·01, –1·39)
0·50 (–5·62, 6 62)
0·90 (–6·12, 7 92)
–1·00 (–6·73, 4·73)
0·10 (–2·16, 2·36)
2·00 (–3·35, 7·35)
–1·43 (–2·49, –0·37)
12·0
5·6
4·3
15·8
17·8
3·7
5·1
5·6
16·4
6·0
100·0%
40
50
30
154
76
71
18
24
104
43
36 2·0
14·3
6·7
2·9
21·6
19·6
3·0
2·3
3·4
20·4
3·9
100·0
9·6–7·9
0·2
–2·2
–5·2
–0·5
–0·7
–4·3
–7·5
–3
–0·9
1
4·6
11·6
13
10·32
6
16·85
8·7
10·43
8·1
10
40
50
30
154
76
71
18
24
104
43
646
646
7·7
22·6
20
15·67
7·7
26·96
13·3
9·89
11·8
15
–14·6
0·5
–1·8
3·6
–2·59
–1·9
–4·5
–6
–6
–4·3
2
652
0·2
0·2
–1·9
–2·8
–7·6
–11·2
–10·6
–5
–3·9
1
Study
Heterogeneity: t2= 3·33; c2 = 14·94, df = 10 (
P
= 0·13);
l
2 = 33 %
Heterogeneity: t2= 0·11; c2 = 10·33, df = 10 (
P
= 0·41);
l
2=3 %
Fig. 3. Absolute changes in (a) systolic blood pressure (SBP) and (b) diastolic blood pressure (DBP). LFD, energy-restricted low-fat diet; VLCKD, very-low-carbo-
hydrate ketogenic diet.
Carbohydrate restriction for weight loss 1185
British Journal of Nutrition
design, data collection, data analysis, data interpretation or
writing of the report. N. B. B. and I. S. V. d. M. acquired, ana-
lysed and interpreted the data. All authors designed the study,
drafted and critically reviewed the manuscript. The authors
declare no conflicts of interest.
References
1. World Health Organization (2012) Obesity and over-
weight. http://www.who.int/mediacentre/factsheets/fs311/en/
(accessed 10 May 2012).
2. Go
¨gebakan O, Kohl A, Osterhoff MA, et al. (2011) Effects of
weight loss and long-term weight maintenance with diets
varying in protein and glycemic index on cardiovascular
risk factors: the diet, obesity, and genes (DiOGenes) study:
a randomized, controlled trial. Circulation 124, 2829–2838.
3. Ornish D (1997) Every Day Cooking With Dean Ornish. New
York: Harper Collins.
4. Sears B & Lawren W (1995) Enter the Zone. New York:
Harper Collins.
5. Weight Watchers Publishing Group (1997) Weight Watchers
New Complete Cookbook. New York: Macmill.
6. Accurso A, Bernstein R, Dahlqvist A, et al. (2008) Dietary
carbohydrate restriction in type 2 diabetes mellitus and
metabolic syndrome: time for a critical appraisal. Nutr
Metab (Lond) 5,9.
7. Alhassan S, Kim S, Bersamin A, et al. (2008) Dietary adher-
ence and weight loss success among overweight women:
results from the A to Z weight loss study. Int J Obes (Lond)
32, 985–991.
8. Sacks FM, Bray GA, Carey VJ, et al. (2009) Comparison of
weight-loss diets with different compositions of fat, protein,
and carbohydrates. N Engl J Med 360, 859–873.
9. Krieger JW, Sitren HS, Daniels MJ, et al. (2006) Effects
of variation in protein and carbohydrate intake on body
mass and composition during energy restriction: a meta-
regression. Am J Clin Nutr 83, 260–274.
10. Hession M, Rolland C, Kulkarni U, et al. (2009) Systematic
review of randomized controlled trials of low-carbohydrate
vs. low-fat/low-calorie diets in the management of obesity
and its comorbidities. Obes Rev 10, 36–50.
11. Nordmann AJ, Nordmann A, Briel M, et al. (2006) Effects
of low-carbohydrate vs low-fat diets on weight loss and
cardiovascular risk factors: a meta-analysis of randomized
controlled trials. Arch Intern Med 166, 285 – 293.
12. Santos FL, Esteves SS, da Costa Pereira A, et al. (2012)
Systematic review and meta-analysis of clinical trials of the
effects of low carbohydrate diets on cardiovascular risk
factors. Obes Rev 13, 1048–1066.
13. Moher D, Liberati A, Tetzlaff J, et al. (2009) Preferred
reporting items for systematic reviews and meta-analyses:
the PRISMA statement. Ann Intern Med 151, 264 –269.
14. Sterne JAC, Egger M & Moher D (2011) Addressing reporting
biases. In Higgins JPT, Green S, eds. Cochrane Handbook for
Systematic Reviews of Interventions version 5.1.0 [updated
March 2011]. http://www.cochrane-handbook.org (accessed
10 March 2012).
15. Robinson KA & Dickersin K (2002) Development of a highly
sensitive search strategy for the retrieval of reports of con-
trolled trials using PubMed. Int J Epidemiol 31, 150–153.
16. World Health Organization Expert Consultation (2004)
Appropriate body-mass index for Asian populations and its
implications for policy and intervention strategies. Lancet
363, 157–163.
17. Higgins JPT & Deeks JJ (2011) Selecting studies and collect-
ing data. In Higgins JPT, Green S, eds. Cochrane Handbook
for Systematic Reviews of Interventions version 5.1.0
[updated March 2011]. http://www.cochrane-handbook.org
(accessed 22 March 2012).
18. Higgins JPT, Altman DG & Sterne JAC (2011) Assessing risk
of bias in included studies. In Higgins JPT, Green S, eds.
Cochrane Handbook for Systematic Reviews of Interventions
version 5.1.0 [updated March 2011]. http://www.cochrane-
handbook.org (accessed 21 March 2012).
19. Deeks JJ, Higgins JPT & Altman DG (2011) Analysing data
and undertaking meta-analyses. In Higgins JPT, Green S,
eds. Cochrane Handbook for Systematic Reviews of Interven-
tions version 5.1.0 [updated March 2011]. http://www.
cochrane-handbook.org (accessed 01 April 2012).
20. DerSimonian R & Laird N (1986) Meta-analysis in clinical
trials. Control Clin Trials 7, 177 –188.
21. Higgins JPT, Deeks JJ & Altman DG (2011) Special topics in
statistics. In Higgins JPT, Green S, eds. Cochrane Handbook
for Systematic Reviews of Interventions version 5.1.0
[updated March 2011]. http://www.cochrane-handbook.org
(accessed 01 April 2012).
22. Harbord RM & Higgins JPT (2008) Meta-regression in stata.
Stata J 8, 493–519.
23. Peters JL, Sutton AJ, Jones DR, et al. (2008) Contour-
enhanced meta-analysis funnel plots help distinguish
publication bias from other causes of asymmetry. J Clin
Epidemiol 61, 991–996.
24. Egger M, Smith GD, Schneider M, et al. (1997) Bias in meta-
analysis detected by a simple, graphical test. BMJ 315,
629–634.
25. Vetter ML, Iqbal N, Volger S, et al. (2010) Long-term effects
of low-carbohydrate versus low-fat diets in obese persons.
Ann Int Med 152, 334–335.
26. Stern L, Iqbal N, Seshadri P, et al. (2004) The effects of
low-carbohydrate versus conventional weight loss diets in
severely obese adults: one-year follow-up of a randomized
trial. Ann Intern Med 140, 778 –785.
27. Foster GD, Wyatt HR, Hill JO, et al. (2010) Weight and
metabolic outcomes after 2 years on a low-carbohydrate
versus low-fat diet: a randomized trial. Ann Intern Med
153, 147–157.
28. Brinkworth GD, Noakes M, Buckley JD, et al. (2009)
Long-term effects of a very-low-carbohydrate weight loss
diet compared with an isocaloric low-fat diet after 12 mo.
Am J Clin Nutr 90, 23–32.
29. Iqbal N, Vetter ML, Moore RH, et al. (2010) Effects of a
low-intensity intervention that prescribed a low-carbohydrate
vs. a low-fat diet in obese, diabetic participants. Obesity (Silver
Spring) 18, 1733–1738.
30. Gardner CD, Kiazand A, Alhassan S, et al. (2007) Compari-
son of the Atkins, Zone, Ornish, and LEARN diets for
change in weight and related risk factors among overweight
premenopausal women: the A to Z Weight Loss Study:
a randomized trial. JAMA 297, 969–977.
31. Ebbeling CB, Swain JF, Feldman HA, et al. (2012) Effects of
dietary composition on energy expenditure during weight-
loss maintenance. JAMA 307, 2627–2634.
32. Westman EC, Feinman RD, Mavropoulos JC, et al. (2007)
Low-carbohydrate nutrition and metabolism. Am J Clin
Nutr 86, 276–284.
33. Volek JS, Sharman MJ & Forsythe CE (2005) Modification of
lipoproteins by very low-carbohydrate diets. J Nutr 135,
1339–1342.
34. Krauss RM, Blanche PJ, Rawlings RS, et al. (2006)
Separate effects of reduced carbohydrate intake and weight
N. B. Bueno et al.1186
British Journal of Nutrition
loss on atherogenic dyslipidemia. Am J Clin Nutr 83,
1025–1031.
35. Berneis KK & Krauss RM (2002) Metabolic origins and
clinical significance of LDL heterogeneity. J Lipid Res 43,
1363–1379.
36. Appel LJ, Sacks FM, Carey VJ, et al. (2005) Effects of protein,
monounsaturated fat, and carbohydrate intake on blood
pressure and serum lipids: results of the omniheart random-
ized trial. JAMA 294, 2455–2464.
37. Anonymous (2001) Primer on statistical significance and
Pvalues. Eff Clin Pract 4, 183–184.
38. Wadden TA, Volger S, Sarwer DB, et al. (2011) A two-year
randomized trial of obesity treatment in primary care prac-
tice. N Engl J Med 365, 1969–1979.
39. Alberti KG, Zimmet P, Shaw J, et al. (2005) The metabolic
syndrome – a new worldwide definition. Lancet 366,
1059–1062.
40. Kass L, Weekes J & Carpenter L (2012) Effect of magnesium
supplementation on blood pressure: a meta-analysis. Eur J
Clin Nutr 66, 411–418.
41. Ried K, Sullivan TR, Fakler P, et al. (2012) Effect of cocoa on
blood pressure. The Cochrane Database of Systematic
Reviews, issue 8 CD008893.
42. Haufe S, Engeli S, Kast P, et al. (2011) Randomized compari-
son of reduced fat and reduced carbohydrate hypocaloric
diets on intrahepatic fat in overweight and obese human
subjects. Hepatology 53, 1504–1514.
43. Phillips SA, Jurva JW, Syed AQ, et al. (2008) Benefit of
low-fat over low-carbohydrate diet on endothelial health
in obesity. Hypertension 51, 376–382.
44. Lagiou P, Sandin S, Lof M, et al. (2012) Low carbohydrate-
high protein diet and incidence of cardiovascular diseases
in Swedish women: prospective cohort study. BMJ 344,
e4026–e4037.
45. Friedman AN, Ogden LG, Foster GD, et al. (2012) Compara-
tive effects of low-carbohydrate high-protein versus low-fat
diets on the kidney. Clin J Am Soc Nephrol 7, 1103 –1111.
46. Greenberg I, Stampfer MJ, Schwarzfuchs D, et al. (2009)
Adherence and success in long-term weight loss diets: the
dietary intervention randomized controlled trial (DIRECT).
J Am Coll Nutr 28, 159–168.
47. Institute of Medicine (2011) Finding What Works in Health
Care: Standards for Systematic Reviews. Washington, DC:
The National Academy Press.
48. Centre for Reviews and Dissemination (2009) Syste-
matic Reviews: CRD’s Guidance for Undertaking Reviews
in Health Care. York: York Publishing Services.
49. Shai I, Schwarzfuchs D, Henkin Y, et al. (2008) Weight loss
with a low-carbohydrate, Mediterranean, or low-fat diet.
N Engl J Med 359, 229–241.
50. Dansinger ML, Gleason JA, Griffith JL, et al. (2005) Compari-
son of the atkins, ornish, weight watchers, and zone diets for
weight loss and heart disease risk reduction: a randomized
trial. JAMA 293, 43–53.
51. Davis NJ, Tomuta N, Schetcher C, et al. (2009) Comparative
study of the effects of a 1-year dietary intervention of a
low-carbohydrate diet versus a low-fat diet on weight and
glycemic control in type 2 diabetes. Diabetes Care 32,
1147–1152.
52. Dyson PA, Beatty S & Matthews DR (2010) An assessment of
low-carbohydrate or low-fat diets for weight loss at 2 year’s
follow-up. Diabet Med 27, 363–368.
53. Foster GD, Wyatt HR, Hill JO, et al. (2003) A randomized trial
of a low-carbohydrate diet for obesity. N Engl J Med 348,
2082–2090.
54. Lim SS, Noakes M, Keogh JB, et al. (2010) Long-term effects
of a low carbohydrate, low fat or high unsaturated fat diet
compared to a no-intervention control. Nutr Metab Cardio-
vasc Dis 20, 599–607.
55. McAuley KA, Smith KJ, Taylor RW, et al. (2006) Long-term
effects of popular dietary approaches on weight loss and fea-
tures of insulin resistance. Int J Obes (Lond) 30, 342–349.
56. Truby H, Baic S, deLooy A, et al. (2006) Randomised con-
trolled trial of four commercial weight loss programmes in
the UK: initial findings from the BBC “diet trials”. BMJ 332,
1309–1314.
Carbohydrate restriction for weight loss 1187
British Journal of Nutrition
