With an estimated prevalence of 350 million cases worldwide, diabetes represents one of the most serious and pressing current health problems. Type 2 diabetes accounts for approximately 90–95 % of cases with manifested diabetes( 1 ). Due to the detrimental consequences and diabetes-associated disorders (e.g. retinopathy, neuropathy and CVD), it is necessary to use every available tool to prevent the onset as well the progression of the disease. Again, type 2 diabetes is of prime importance, as its pathogenesis can be prevented by lifestyle modifications. Increased physical activity and improved nutritional habits in the form of healthy balanced diets are of particular importance in the deceleration of type 2 diabetes manifestations. In 2003, the American Diabetes Association classified an intermediate group of individuals who did not meet the criteria for symptomatic diabetes with respect to their glucose levels. However, due to impaired fasting glucose (FG) levels (values of fasting plasma glucose in the range of 5·6–6·9 mmol/l) and impaired glucose tolerance (2 h values of plasma glucose in the range of 7·8–11 mmol/l following an oral glucose tolerance test) or permanently increased glycosylated Hb levels (5·7–6·4 % or 38·7–46·4 mmol/mol), these individuals are considered to be at an increased risk of type 2 diabetes and termed pre-diabetics. Impaired FG levels and impaired glucose tolerance are usually associated with obesity (especially increased abdominal and visceral fat mass), dyslipidaemia, increased plasma TAG levels, decreased plasma HDL-cholesterol (HDL-C) levels and hypertension, all representing established type 2 diabetes risk factors( 2 ). Numerous long-term intervention studies have shown that lifestyle improvements exert beneficial effects on the onset and progression of type 2 diabetes. Thus, both weight reduction and physical exercise have been shown to be significantly associated with a decreased incidence of diabetes( Reference Lindstrom, Ilanne-Parikka and Peltonen 3 , Reference Li, Zhang and Wang 4 ).
To accomplish the objective of weight reduction, the American Diabetes Association recommended energy-reduced dietary protocols without giving any preference to macronutrient composition. According to its position statement, both a low-carbohydrate (LC) regimen and a low-fat (LF) regimen are considered to be effective means for weight management( Reference Bantle, Wylie-Rosett and Albright 5 ). In a meta-analysis published in 2009, Kodama et al. ( Reference Kodama, Saito and Tanaka 6 ) investigated the short-term effects of LF v. LC diet consumption in patients with type 2 diabetes. They could observe significantly more pronounced changes in fasting insulin (FI), TAG and HDL-C levels, all in favour of the LC dietary protocol. With respect to their macronutrient distribution, it seems legitimate to put high-fat (HF) diets on a level with LC diets, assuming that a reduced carbohydrate content will most probably be accompanied by an increased amount of fat in the total energy content (TEC). The aim of the present systematic review was to examine the long-term effects ( ≥ 12 months) of HF v. LF diet consumption on the parameters of glycaemic control in both pre-diabetics and patients with established type 2 diabetes. Furthermore, cardiovascular risk factors (e.g. TAG, LDL-cholesterol and HDL-C) were included as outcome parameters. As has been described previously, a HF diet was defined to provide >30 % of TEC in the form of fat, whereas ≤ 30 % of TEC was taken up in the form of fat in the LF counterparts( Reference Schwingshackl and Hoffmann 7 ). In this context, the data reported by Kodama et al. ( Reference Kodama, Saito and Tanaka 6 ) would argue against the equivalent usefulness of LF ( = high-carbohydrate) and LC ( = HF) diets emphasising the value of the latter. However, the validity of this conclusion has to be substantiated by the analysis of long-term studies including the objectives of primary prevention, i.e. investigating the effects of different dietary protocols on risk parameters in pre-diabetic individuals.
Methods
Data sources and searches
Literature search restricted to randomised controlled trials (RCT) without any restrictions on language and calendar date was carried out using the electronic databases MEDLINE (until November 2013), Embase (until November 2013) and the Cochrane Trial Register (until November 2013) with the following search terms: low-fat diet; high-fat diet; low-carbohydrate diet; high-carbohydrate diet. Full details of the electronic search strategy are given in the online supplementary material. Furthermore, the reference lists from the retrieved articles were checked to search for further relevant studies. This systematic review was planned, conducted and reported in adherence to the standards of quality for reporting meta-analyses( Reference Moher, Liberati and Tetzlaff 8 ). Literature search as well as article abstraction was conducted independently by both the authors, with disagreements being resolved by consensus.
Study selection
Studies were included in the meta-analysis if they met all the following criteria: (1) randomised controlled design; (2) minimum intervention period with a follow-up period of 12 months; (3) comparison of a HF dietary intervention (>30 % of TEC) with a LF dietary intervention ( ≤ 30 % of TEC), stated as the goal of intervention; (4) age of the subjects ≥ 18 years; (5) enrolment of subjects with either impaired FG levels ( ≥ 5·6 mmol/l) or insulin resistance( Reference McAuley, Williams and Mann 9 ) or type 2 diabetes( 2 ); (6) assessment of the ‘outcome of interest’ markers: body weight, total cholesterol (TC), LDL-cholesterol, HDL-C, TC:HDL-C ratio, TAG, systolic and diastolic blood pressure (SBP/DBP), C-reactive protein, FI, FG, glycosylated Hb, and adiponectin; (7) report of post-intervention mean or mean of values recorded at two time points with standard deviation (or basic data to calculate these parameters). If data of ongoing studies were published as updates, results of only those of the longest duration were included. Studies that enrolled patients with type 1 diabetes mellitus were excluded.
Data extraction and quality assessment – overall quality of evidence
The focus of this systematic review was the examination of the effects of LF v. HF diet consumption on the values of glycaemic control and cardiovascular risk factors in individuals at a high risk of type 2 diabetes or with manifested type 2 diabetes. The following types of dietary interventions were evaluated: in the primary analysis, LF diets were defined as those providing ≤30 % of TEC in the form of fat, approximately 15 % of TEC from protein and approximately 55 % of TEC from carbohydrates and HF diets were defined as those providing >30 % of TEC as fat. HF diets were further classified according to alternative macronutrient composition as follows:
-
(1) usual diet (total fat content >30 % of TEC and SFA content >10 % of TEC);
-
(2) LC diet ( < 50 g carbohydrates/daily);
-
(3) high-monounsaturated fat, MUFA diet (total fat content >30 % of TEC and total MUFA content >12 % of TEC).
The risk of bias assessment tool provided by the Cochrane Collaboration was applied specifying the following bias domains: selection bias (random sequence generation and allocation concealment); performance/detection bias (blinding of participants and personnel/blinding of outcome assessment); attrition bias (incomplete data outcome); reporting bias (selective reporting); other bias (see online supplementary Fig. S1)( Reference Higgins, Altman and Gotzsche 10 ).
The following data were extracted for each study: the first author's last name; year of publication; study duration; sex distribution and age; BMI; percentage of diabetics; sample size; outcomes; post-intervention mean values or differences in the mean of values recorded at two time points with the corresponding standard deviation. The quality of evidence was assessed according to the Grading of Recommendations Assessment, Development and Evaluation (GRADE) guidelines( Reference Guyatt, Oxman and Vist 11 ).
Data synthesis and analysis
For each outcome measure of interest, a meta-analysis was carried out to determine the pooled effect of the intervention in terms of weighted mean differences (MD) between the post-intervention (or differences in means) values of the HF group and those of the LF group. Combining both the post-intervention values and differences in means in one meta-analysis is a legitimate method described by the Cochrane Collaboration( Reference Higgins and Green 12 ). All data were analysed using the Review Manager 5.1 software, provided by the Cochrane Collaboration (http://ims.cochrane.org/revman). The random-effects model was used to estimate MD with 95 % CI. Forest plots were generated to illustrate the study-specific effect sizes along with 95 % CI. Heterogeneity between the trial results was tested with a standard χ2 test. The I 2 parameter was used to quantify any inconsistency:
$$\begin{eqnarray} I ^{2} = (( Q - df))/ Q \times 100\,\%\, \end{eqnarray}$$
where Q is the χ2 statistic. A cut-off point ≥ 50 % was chosen for I 2 to indicate substantial heterogeneity( Reference Higgins and Green 12 ). To evaluate substantial heterogeneity, several post hoc univariate random-effects meta-regressions were performed to examine the association between total fat, SFA, MUFA, PUFA, carbohydrate, protein, dietary cholesterol and fibre intake as independent variables and changes in FG, FI, TC and HDL-C levels (where substantial heterogeneity could be detected) as dependent variables. Furthermore, BMI, age, sex and study duration were used as covariates. The P values for differences in effects between the covariates were obtained using the metareg function of Stata 12.0 (Stata-Corporation). Two-sided P values < 0·05 were considered to be statistically significant. To increase the precision of the estimates of macronutrient intake (total fat, SFA, MUFA, PUFA, fibre, cholesterol, carbohydrate and protein (all in percentage of TEC)), data from FFQ, 24 h dietary recalls or 3–7 d dietary protocols (if available) were used instead of the theoretical macronutrient composition values of dietary intervention protocols. Funnel plots were sketched to indicate potential publication bias (e.g. the tendency for studies yielding statistically significant results to be more likely to be submitted and accepted for publication). To determine the presence of publication bias, the symmetry of the funnel plots in which MD were plotted against their corresponding standard errors was assessed. A primary analysis of all studies oriented towards the definition of HF and LF diets was carried out, followed by a subanalysis of the specific kind of dietary intervention as described in the selected studies. In addition, a sensitivity analysis including only subjects with type 2 diabetes as well as a sensitivity analysis to determine the risk of bias of the trials( Reference Liberati, Altman and Tetzlaff 13 , Reference Nordmann, Suter-Zimmermann and Bucher 14 ) was conducted.
Among the studies included in the review, two studies( Reference Elhayany, Lustman and Abel 15 , Reference Wolever, Gibbs and Mehling 16 ) used two types of LF diets, and these diets were combined into one group as described in the Cochrane Handbook( Reference Higgins and Green 12 ). In the study carried out by Shai et al. ( Reference Shai, Schwarzfuchs and Henkin 17 ), the LF and LC branches were extracted for meta-analysis, while the MUFA group was discarded and re-analysed in the sensitivity analyses, as duplicate application of the LF dietary intervention data would not be legitimate. Data extraction was conducted independently by both the authors, with disagreements being resolved by consensus. Data processing for this review required the input of the mean and standard deviation of post-intervention values or differences in means. In case of missing data, the authors of the original article were asked for additional information and, if provided, the raw data were used for computations( Reference Wolever, Gibbs and Mehling 16 , Reference Shai, Schwarzfuchs and Henkin 17 ).
Results
Literature search
A total of fourteen studies extracted from 16 608 articles met the inclusion criteria and were analysed in the systematic review( Reference Elhayany, Lustman and Abel 15 – Reference Stern, Iqbal and Seshadri 30 ). Detailed steps of the article selection process used for the present meta-analysis are shown as a flow chart in Fig. 1. In accordance with the overall inclusion criteria, five studies were excluded due to inconsistencies in the mean FG levels of the study populations ( ≥ 5·6 mmol/l) and the corresponding standard deviations, thereby increasing the potential for selection bias (Fig. 1)( Reference Klemsdal, Holme and Nerland 31 – Reference Cardillo, Seshadri and Iqbal 35 ).
Fig. 1 Flow chart for the systematic review article selection process. RCT, randomised controlled trials; FG, fasting glucose.
Study and participant characteristics
All studies included in this systematic review were RCT with a duration ranging between 12 months and 6 years, published between 1978 and 2012, and enrolling a total of 2003 participants. All studies compared a HF regimen defined as a LC diet (six studies), a control/HF diet (four studies) or a MUFA-rich protocol (four studies) with a LF regimen. The reported BMI was >25 kg/m2 in all the studies and the mean age of the participants varied between 52 and 62 years. The reported dropout rates were 25 % for the HF group and 23 % for the LF group. The general study characteristics are given in Table 1. Results obtained using the risk of bias assessment tool are summarised in online supplementary Fig. S1.
Table 1 Characteristics of the randomised controlled intervention trials included in the present meta-analysis
T2D, type 2 diabetes; IFG, impaired fasting glucose; IR, insulin resistance; CH, carbohydrates; HbA1c, glycosylated Hb; HF, high fat; LF, low fat; LC, low carbohydrate; ND, not determined; IGT, impaired glucose tolerance; ADA, American Diabetes Association; PPAR agonist, PPAR-γ agonist; OGTT, oral glucose tolerance test.
* To convert HbA1c to mmol/mol: 10·93 × HbA1c unit (%) − 23·50.
† Two kinds of LF diets (high-glycaemic index and low-glycaemic index; 10 and 30 % of total energy consumption).
‡ In case subjects wished to lose weight.
Outcomes
The pooled estimates of weighted MD for the effects of HF diet consumption compared with those of LF diet consumption on body weight, blood lipids and parameters of glycaemic control are summarised in Table 2. Changes in body weight (see online supplementary Fig. S2), TC levels (see online supplementary Fig. S3), LDL-cholesterol levels (see online supplementary Fig. S4), SBP (see online supplementary Fig. S8), FI levels (see online supplementary Fig. S9), FG levels (see online supplementary Fig. S10), glycosylated Hb levels (see online supplementary Fig. S11), TC:HDL-C ratio and C-reactive protein levels in subjects following a HF diet were not significantly different from those in subjects following a LF diet.
Table 2 Pooled estimates of effect size (95 % CI) expressed as weighted mean difference (MD) for the effects of high-fat v. low-fat diet consumption on cardiovascular and metabolic risk factors
GRADE, Grading of Recommendations Assessment, Development and Evaluation; BW, body weight; TC, total cholesterol; LDL-C, LDL-cholesterol; HDL-C, HDL-cholesterol; CRP, C-reactive protein; DBP, diastolic blood pressure; SBP, systolic blood pressure; FG, fasting glucose; FI, fasting insulin; HbA1c, glycosylated Hb.
* Risk of bias in many studies.
† Sensitivity analysis: including studies with only subjects with type 2 diabetes.
‡ Heterogeneity was observed, but could not be explained.
§ Large study effects were observed; if studies excluded from the analysis, the MD became non-significant.
The HF dietary protocols were found to lead to a significantly more pronounced decrease in TAG levels when compared with the LF dietary protocols (MD − 0·19 mmol/l, 95 % CI − 0·23, − 0·14, P< 0·001; I 2= 0 %, P= 0·58). The subgroup analyses revealed that the decrease in TAG levels was significant in subjects following a MUFA-rich diet than in those following a LF diet (MD − 0·20 mmol/l, 95 % CI − 0·25, − 0·15, P< 0·001; I 2= 13 %, P= 0·33; see online supplementary Fig. S6).
HDL-C (see online supplementary Fig. S5; MD 0·05 mmol/l, 95 % CI 0·01, 0·08, P= 0·01; I 2= 57 %, P= 0·01) and adiponectin (MD 1·10 μg/ml, 95 % CI 0·87, 1·33, P< 0·001) levels were significantly more increased by the HF dietary interventions than by the LF dietary interventions. Subgroup analyses yielded non-significant results.
With respect to blood pressure values, reductions in DBP were significantly more explicit in subjects adhering to a HF diet than in those adhering to a LF diet (MD − 1·30 mmHg, 95 % CI − 1·73, − 0·87, P< 0·001; I 2= 0 %, P= 0·60; see online supplementary Fig. S7). Comparable results for both DBP (MD − 1·38 mmHg, 95 % CI − 1·82, − 0·95, P< 0·001; I 2= 0 %, P= 0·79) and SBP (MD − 1·48 mmHg, 95 % CI − 2·01, − 0·96, P< 0·001; I 2= 0 %, P= 0·45) were obtained when comparing high-MUFA groups with LF groups in subgroup analyses.
With regard to biomarkers of glycaemic control, changes in FI levels did not differ between the HF and LF groups. However, post hoc analysis of subgroups revealed that FI values were significantly more increased in subjects adhering to a usual/HF diet than in those adhering to a LF diet (MD 7·79 pmol/l, 95 % CI 3·24, 12·33, P< 0·001; I 2= 0 %, P= 0·95).
Sensitivity/subgroup analyses
To investigate the effects of HF v. LF diet consumption in patients with manifested type 2 diabetes, a sensitivity analysis was carried out by excluding all RCT enrolling only subjects at a risk of type 2 diabetes (i.e. individuals with impaired FG levels or insulin resistance). A total of eleven studies remained for secondary analyses( Reference Elhayany, Lustman and Abel 15 , Reference Shai, Schwarzfuchs and Henkin 17 , Reference Brehm, Lattin and Summer 19 – Reference Hockaday, Hockaday and Mann 23 , Reference Iqbal, Vetter and Moore 26 , Reference Milne, Mann and Chisholm 29 , Reference Stern, Iqbal and Seshadri 30 , Reference Wolever, Mehling and Chiasson 36 ). The results turned out to be not significantly different from those of the comprehensive meta-analyses except for one parameter: detrimental effects on FI values were no longer valid when comparing usual v. LF protocols in the respective subgroup analysis.
Moreover, additional parameters were specifically altered by HF diet consumption in patients with type 2 diabetes. Both SBP (MD − 1·35 mmHg, 95 % CI − 2·35, − 0·35, P= 0·008; I 2= 3 %, P= 0·39) and FG (MD − 0·41 mmol/l, 95 % CI − 0·74, − 0·08, P= 0·01; I 2= 56 %, P= 0·02) levels were significantly more decreased in subjects following a HF diet than in those following a LF diet. In the Women's Health Initiative as well as the Workforce Diabetes Survey trials( Reference Howard, Curb and Eaton 24 , Reference Shikany, Margolis and Pettinger 25 , Reference Ley, Metcalf and Scragg 27 , Reference Swinburn, Metcalf and Ley 28 ), the HF dietary intervention group was indicated as a control group receiving only ‘usual care’. By implication, one might expect a bias due to the fact that the intervention (i.e. the LF) group received more attention from the investigators, e.g. in form of education, personal goal settings and regular meeting. However, the results of the primary analysis were not affected by the exclusion of these trials.
Subgroup analyses for change scores, hypoenergetic diets, ad libitum diets, study duration (long term and short term), high adherence and low risk of bias were carried out (see online supplementary Tables S1–S7). The majority of the results of the main analysis could be confirmed. Inclusion of only ad libitum diets significantly reduced heterogeneity. Information on the prescription of hypoglycaemic agents was extracted, and it is reported in Table 1. However, due to inconsistent reporting, the data did not allow carrying out further analyses.
Publication bias
The funnel plots (with respect to effect size changes for biomarkers of cardiovascular risk and glycaemic control in response to LF diet consumption) revealed very little asymmetry, suggesting low evidence of publication bias (see online supplementary Figs. S12–S21).
Heterogeneity
Substantial heterogeneity was found with respect to TC (I 2= 67; 73 %), HDL-C (I 2= 57; 65 %), FG (I 2= 82; 56 %) and FI (I 2= 71; 58 %) levels in both primary and secondary analyses (Table 2). It was assumed that substantial heterogeneity might be explained by non-uniform study characteristics of the HF groups such as variations in post-intervention macronutrient intake. To gain insight into these potential correlations, a random-effects meta-regression was performed to examine the associations between LF and HF group parameters and changes in TC, HDL-C, FG, and FI levels. Studies with a higher percentage of energy from carbohydrates were associated with slightly lower differences in TC levels between the two dietary intervention groups (0·026 mmol/l lower TC for every 1 % increase in energy from carbohydrates; 95 % CI 0·050, 0·002; P= 0·036; Fig. 2(a)), those with a higher percentage of energy from fat were associated with slightly higher differences in FI levels between the two dietary intervention groups (2·67 pmol/l higher FI for every 1 % increase in energy from fat; 95 % CI 0·959, 4·38; P= 0·006; Fig. 2(b)), and those with a higher percentage of energy from MUFA and total fat were associated with slightly higher differences in HDL-C levels between the two dietary intervention groups (0·014, 0·012 mmol/l higher HDL-C for every 1 % increase in energy from MUFA and total fat; 95 % CI 0·003, 0·024; P= 0·015; 95 % CI 0·004, 0·020, P= 0·005; Fig. 2(c) and (d)). No such correlations could be detected for the other parameters under investigation.
Fig. 2 Bubble plots showing the dose–response relationship between (a) carbohydrate intake and changes in total cholesterol (TC) levels (P= 0·036), (b) fat intake (P= 0·006) and changes in fasting insulin (FI) levels, (c) MUFA intake and changes in HDL-cholesterol (HDL-C) levels (P= 0·015), and (d) total fat intake and changes in HDL-C levels (P= 0·005). Macronutrient intake is given as the percentage of total energy consumption. The fitted regression lines together with the bubbles represent the estimates from each study, sized according to the precision of each estimate as calculated in the fitted random-effects meta-regression. MD, mean difference. (A colour version of this figure can be found online at http://www.journals.cambridge.org/bjn).
Overall quality of evidence
The overall quality of evidence rated according to the GRADE guidelines ranged from very low to moderate (Table 2). Moderate-quality evidence was found concerning the significant reduction in TAG levels as well as the significant increase in HDL-C levels in subjects following a HF diet. Similarly, moderate-quality evidence could be observed with respect to the non-significant effects of HF regimens on TC and LDL-cholesterol levels. Moderate-quality evidence was observed for the effects of HF diet consumption on the parameters of glycaemic control as outcomes as well. For the remaining outcome parameters (blood pressure, C-reactive protein and adiponectin), low-to-very-low-quality evidence was observed.
Discussion
Type 2 diabetes mellitus is one of the most pressing non-communicable chronic diseases with an estimated prevalence of approximately 235 million affected cases by the year 2030. In the pathogenesis of its manifestation, type 2 diabetes is preceded by a most often undetected phase of pre-diabetes with impaired glucose metabolism, which, however, is susceptible to the benefits of lifestyle changes( 2 ). Various biomarkers have been shown to be predictors of the detrimental consequences of diabetes-associated disorders resulting from micro- and macroangiopathies. In the present systematic review and meta-analysis, long-term intervention studies enrolling either patients with type 2 diabetes or subjects with pre-diabetes and comparing a HF dietary regimen with a LF dietary regimen were analysed. The primary analysis revealed a favourable effect of HF diet consumption with respect to TAG levels (decrease), DBP (decrease), HDL-C levels (increase) as well as adiponectin levels (increase), although the last-mentioned biomarker was measured in only one study including 215 diabetic volunteers( Reference Esposito, Maiorino and Ciotola 21 ). By contrast, an advantage of a LF protocol could be observed when compared with its usual/HF counterpart with respect to FI values in the subgroup analyses. However, this benefit was no longer present following sensitivity analysis including studies with only patients with type 2 diabetes. Thus, a major finding of this meta-analysis is that HF diets exert beneficial effects – when compared with LF diets – on a number of biomarkers considered to be predictors of diabetes-associated complications.
CHD is a highly prevalent manifestation of microangiopathies associated with diabetes and has been reported to have caused 12 % of all premature deaths worldwide in 2004( 37 ). With respect to HDL-C levels, a previous epidemiological study has shown that an increase of approximately 0·025 mmol/l is associated with a decrease in CVD risk of 2 % in men and 3 % in women( Reference Maron 38 ). In the present meta-analysis, HF protocols were found to result in an average higher increase of 0·05 mmol/l in plasma HDL-C levels when compared with LF protocols, indicating a greater risk reduction in CHD by 3·75–5·5 %. Augmented plasma levels of TAG are considered to be univariate predictors of CVD as well( Reference Miller, Stone and Ballantyne 39 ). An increase of 1 mmol/l in plasma levels is associated with a 2-fold increase in the relative risk of CHD( Reference He, Lam and Li 40 ). On combining these data with the results of the present meta-analysis, it was found that the decline in TAG levels following HF diet consumption would be associated with a reduced CHD risk of approximately 10 %.
Increased blood pressure intensifies the risk of stroke and CHD in patients with type 2 diabetes and it could be shown that even minor reductions in blood pressure will reduce the incidence of CVD. Improvements in mean arterial pressure ( − 3 mmHg) were found to be correlated with a reduction in the risk of CVD (5–10 %), stroke (8–15 %) and all-cause mortality (5 %)( Reference Whelton, He and Appel 41 ). Moreover, a single reduction of SBP ( − 3 %) was found to be associated with a decreased risk of fatal myocardial infarctions of approximately 8 %( Reference Stamler, Rose and Elliott 42 ). The present data suggest a beneficial effect of HF diet consumption on DBP in patients with type 2 diabetes and subjects with pre-diabetes, while the sensitivity analysis including only patients with manifested diabetes revealed a beneficial effect on SBP as well (when compared with LF diet consumption).
A significant decrease in FG values ( − 0·41 mmol/l) was found in patients with type 2 diabetes subjected to a HF diet challenge. According to the Asian Pacific Study, attenuations in FG levels of 180 mg/l are correlated with a 23 % diminished risk of CVD( Reference Lawes, Parag and Bennett 43 ). Moreover, the authors of the United Kingdom Prevention Study considered hyperglycaemia to be a more relevant predictor of coronary events in the course of type 2 diabetes when compared with increased insulin levels( Reference Turner, Millns and Neil 44 ). This hypothesis is supported by the findings of the Emerging Risk Factors Collaboration Study reporting a significant correlation between FG levels >5·6 mmol/l and increased mortality risk( Reference Seshasai, Kaptoge and Thompson 45 ).
Among the studies included in the present meta-analysis, one study reported that HF diets increased plasma adiponectin levels. Epidemiological data as well as experimental studies have indicated that decreased plasma adiponectin values are associated with a pronounced risk of insulin resistance and manifestation of type 2 diabetes( Reference Yamamoto, Hirose and Saito 46 ).
When compared with the findings of the present meta-analysis, similar data have been reported in meta-analyses investigating short-term RCT. Thus, Kodama et al. ( Reference Kodama, Saito and Tanaka 6 ) observed significant increases in TAG levels as well as decreases in HDL-C levels in patients with type 2 diabetes subjected to a LF diet challenge than in those of patients subjected to a HF diet challenge. Weighing moderate-fat diets against a regimen with LF intake, Cao et al. ( Reference Cao, Mauger and Pelkman 47 ) found significant changes in TAG levels, HDL-C levels and TC:HDL-C ratio in favour of the moderate-fat protocol. Following a meta-regression analysis, the authors concluded that at least the changes observed in HDL-C levels were correlated with total fat as well as unsaturated fat intake, suggesting that a reduction in carbohydrate intake with simultaneous increase in unsaturated fat (MUFA and PUFA) intake would exert a beneficial effect on plasma lipoproteins. The adherence of individuals assigned to a LC dietary intervention might change over time. Usually, there is good adherence in the short term, but it gets poorer in the long term( Reference Bueno, de Melo and de Oliveira 48 ), which might explain the benefits of LC diet consumption observed by Kodama et al. ( Reference Kodama, Saito and Tanaka 6 ). In the present meta-analysis, macronutrient intakes in the included trials were found to be altered sometimes at the end of the follow-up period. However, in most studies, the dietary protocol at the end of the follow-up period was still distinguishable with respect to HF or LF intakes (Table 1). However, it should be noted that the HF and LF diets had roughly the same dietary macronutrient composition at the end of the follow-up period in the studies carried out by Iqbal et al. ( Reference Iqbal, Vetter and Moore 26 ) and Milne et al. ( Reference Milne, Mann and Chisholm 29 ).
In this meta-analysis, a substantial heterogeneity (I 2>50 %) was found for TC, HDL-C, FG and FI levels (Table 2). Following meta-regressions to examine the associations between HF and LF diet consumption and changes in the outcome parameters, a statistically significant relationship was detected between carbohydrate intake and decreases in TC levels (probably caused by a reduction in SFA intake)( Reference Schwingshackl and Hoffmann 7 ), fat intake and increases in FI levels, and total fat and MUFA intake and increases in HDL-C levels. This is in accordance with the findings from other meta-analyses( Reference Schwingshackl and Hoffmann 49 – Reference Garg 54 ). By comparing the results of four recent meta-analyses, Pagoto & Appelhans( Reference Pagoto and Appelhans 55 ) suggested that investigations dealing with different dietary macronutrient approaches show only small differences between the diets. This statement could be confirmed by the results of the present meta-analysis.
Limitations of the present systematic review
The data of the present meta-analysis refer only to values obtained after an overnight fast and, as such, only represent part of the glycaemic control data; for example, there are no postprandial or post-glucose challenge data. Moreover, the present systematic review did not consider unpublished results, and it cannot be excluded that these results may have had at least a moderate impact on the effect size estimates. Examination of funnel plots revealed very little asymmetry, suggesting that the evidence for publication bias is of low quality. A major limitation of nutritional intervention trials is the heterogeneity of various aspects and characteristics of the study protocols. Therefore, it is not surprising that the RCT included in the present meta-analysis varied regarding the type(s) of diets used (energy restriction, isoenergetic), definitions of LF and HF diets, study population (i.e. age, BMI, type 2 diabetics, abnormal glucose metabolism), intervention time and nutritional assessment as well as long-term follow-up periods (1–6 years.). Following sensitivity analysis including only studies enrolling patients with type 2 diabetes, the beneficial effects of HF diet consumption on TAG levels, HDL-C levels and DBP were found to remain the same as those observed in the conclusive analyses. Moreover, HF protocols were found to exert a more favourable effect than their LF counterparts with respect to SBP and FG levels in patients with manifested diabetes.
With respect to other potential modulating variables, sensitivity analyses and meta-regressions failed to reveal any correlations between the findings of the meta-analysis and age, sex, BMI and study duration (data not shown). These findings must be interpreted in a very conservative manner due to the low number of studies available for the meta-regressions. Not all the studies provided information on the quality of their respective set-up (e.g. method of randomisation and follow-up protocol with reasons for withdrawal; see online supplementary Fig. S1 for the risk of bias assessment according to the Cochrane Collaboration), demanding a conservative interpretation of results. In this context, it should be noted that the RCT varied with respect to dietary assessment methods to validate participant individual intakes. In addition, the drug regimen was not identical for all the participants in the included studies, and the diagnosis and classification of type 2 diabetes differed between the intervention trials. The pooled analysis of adiponectin values of only one study was disputable, but no more data were available. Part of the present meta-analysis was carried out using both post-intervention values and changes in MD; however, this was considered to be an acceptable procedure as described by the Cochrane Collaboration( Reference Higgins and Green 12 ).
This systematic review has some strengths as well. The meta-analysis was conducted following a stringent protocol; for example, the participants were randomly assigned to the intervention groups in all trials. RCT are considered to be the gold standard for evaluating the effects of an intervention and are subject to fewer biases when compared with observational studies.
In conclusion, HF diet consumption was found to exert beneficial effects on TAG levels, DBP and SBP, and HDL-C levels as well as FG levels in subjects who either were pre-diabetic or had manifested type 2 diabetes when compared with LF diet consumption. Therefore, HF and LF diets might not be of equal value in the management of either pre-diabetes or type 2 diabetes, leading to emphasis being placed on the recommendations of HF diets. In this regard, one major issue is the qualitative composition of fat (i.e. higher amounts of MUFA and PUFA and lower amounts of SFA in the percentage of TEC). As a large number of individuals with pre-diabetes or type 2 diabetes are either overweight or obese, nutritional recommendations often include hypoenergetic diets for weight management. Thus, a successful HF strategy has to implement limitations on other nutrients with energetic value, most probably carbohydrates. However, with respect to the high heterogeneity of the RCT included in this systematic review, further long-term intervention trials with a standardised approach are necessary to elucidate the benefits and disadvantages of both dietary regimens.
Supplementary material
To view supplementary material for this article, please visit http://dx.doi.org/10.1017/S0007114514000464
Acknowledgements
The authors are grateful to Iris Shai, PhD, and Thomas Wolever, PhD, for providing raw data for this systematic review and meta-analysis.
The authors received no financial support for their work.
The authors' contributions are as follows: L. S. and G. H. conducted the data analysis, interpreted the results, drafted the manuscript and approved the final version of the manuscript. Both the authors read and approved the final manuscript. The authors have no conflicts of interest to declare.
