The effects of almond consumption on fasting blood lipid levels: a systematic review and meta-analysis of randomised controlled trials

A systematic review and meta-analysis of randomised controlled trials was undertaken to determine the effects of almond consumption on blood lipid levels, namely total cholesterol (TC), LDL-cholesterol (LDL-C), HDL-cholesterol (HDL-C), TAG and the ratios of TC:HDL-C and LDL-C:HDL-C. Following a comprehensive search of the scientific literature, a total of eighteen relevant publications and twenty-seven almond-control datasets were identified. Across the studies, the mean differences in the effect for each blood lipid parameter (i.e. the control-adjusted values) were pooled in a meta-analysis using a random-effects model. It was determined that TC, LDL-C and TAG were significantly reduced by −0·153 mmol/l (P < 0·001), −0·124 mmol/l (P = 0·001) and −0·067 mmol/l (P = 0·042), respectively, and that HDL-C was not affected (−0·017 mmol/l; P = 0·207). These results are aligned with data from prospective observational studies and a recent large-scale intervention study in which it was demonstrated that the consumption of nuts reduces the risk of heart disease. The consumption of nuts as part of a healthy diet should be encouraged to help in the maintenance of healthy blood lipid levels and to reduce the risk of heart disease.

The following inclusion criteria were applied: (1) a human intervention study that was randomised and controlled (such that the control group/phase could have consisted of either no food or any food(s) without tree nuts or fractions of tree nuts); (2) a full-length article that was published in a peerreviewed journal; (3) the objective of the study (either primary or secondary) was to assess the effects of almond consumption on blood lipid levels; (4) the amount of almonds consumed was reported; (5) the subjects were adults (aged ≥18 years of age) without serious disease such as heart disease; (6) the study duration was ≥4 weeks; (7) fasting blood lipids (i.e. TC, LDL-C, HDL-C and/or TAG) were assessed; and (8) fasting blood lipids were measured using validated methods.
The following exclusion criteria were applied: (1) the publication was of a secondary research study (e.g. systematic review or meta-analysis); (2) the objective of the study (either primary or secondary) was not to assess the effects of almond consumption on blood lipid levels; (3) the subjects had a serious disease (e.g. heart disease, cancer) and were not representative of the general population; (4) the subjects were children or pregnant or lactating women; (5) control-adjusted effects on blood lipids could not be calculated from the data provided; (6) the independent effects of almonds on blood lipid levels could not be isolated (e.g. almonds were co-consumed with other nuts or another nutritional or pharmaceutical intervention); and (7) the study results for the same population group were published in another journal (i.e. the study was a kin publication to another study).

Data extraction and study quality
Study data were extracted independently by two reviewers (T. P. and H. Y. L.), and the consistency of the two datasets was verified by a third reviewer (K. M. V.). Where there were inconsistencies or discrepancies between the two datasets, the original publication was consulted, and a consensus was reached via discussions between the three reviewers (T. P., H. Y. L. and K. M. V.). Data extracted from the studies included study design, country of study conduct, sample size, study population (proportion of males, health status, mean age, mean BMI, mean baseline blood lipid levels (i.e. TC, LDL-C, HDL-C, TAG, TC:HDL-C; LDL-C:HDL-C)), dietary interventions (dose, form of almonds, provision of foods or meals, duration, frequency of intake, pattern of intake), background diets and macronutrient intakes, statistical results, and the mean difference in the effect for each blood lipid parameter (see the Statistical analysis section for details on how the mean difference in the effect was calculated for the crossover and parallel studies).
Health Canada's standardised quality appraisal tool (8) was used to determine the quality of the studies. A quantitative score (zero or one) was assigned to each of the fifteen items included in the tool, and studies with scores of ≥8/15 were considered to be 'higher quality' while studies with scores of ≤7/15 were considered to be 'lower quality'. Study quality was appraised by one reviewer (L. P.).

Statistical analysis
Several of the identified studies had multiple comparisons (e.g. one study may have had three arms, including one control and two different almond doses). Each almond-control comparison, hereinafter referred to as a stratum, was considered a separate trial; however, the control sample size was divided evenly amongst the comparisons so as to avoid inflating the weight of each stratum. For parallel studies, the mean difference in the effect for each blood lipid parameter was calculated as the change from baseline in the control group subtracted from the change from baseline in the almond group. For crossover studies, the mean difference in the effect for each blood lipid parameter was calculated as the blood lipid value at the end of the control phase subtracted from the blood lipid value at the end of the almond phase.
In order to determine the effects of almonds on each of the blood lipid parameters, the results of the studies were pooled in a meta-analysis, with the mean difference in the effect and the inverse of the variance used as the weighting factor. In the majority of the studies, variances for the mean differences were not reported; thus, variances were calculated using information provided in the publication (e.g. using CI or individual variances for the almond and control groups). If, in parallel studies, variances for the changes from baseline were reported separately for the almond and control groups, then a pooled variance for the mean difference was calculated. If, for parallel studies, variances only for the baseline and end of treatment values were reported, then these were used to calculate the variance for the change from baseline, using a correlation coefficient of 0·8. Similarly, for crossover studies, if variances only for the end of treatment values were reported, then the variance for the mean difference was calculated using a correlation coefficient of 0·8. A correlation coefficient of 0·8 was used because this value approximated that calculated from the studies in which variances were provided for the baseline, end of treatment, and change from baseline measures (9)(10)(11) . A random-effects model was used, according to the methods described by DerSimonian & Laird (12) , given that randomeffects models take into consideration the variability in response both within and between studies.
The pooled estimates and accompanying 95 % CI were determined using Comprehensive Meta-analysis Software (version 2.2.064). Publication bias was assessed according to the trim-and-fill method developed by Duval & Tweedie (13) . With this method, asymmetry in the funnel plot is searched for. If the asymmetry is determined to be due to the presence of small studies (with large variances) in which large effect sizes were reported, with an unbalanced number of small studies showing a small effect, then those 'missing' studies are imputed, and the pooled effect size is recalculated.
Subgroup analyses were conducted to evaluate the influence of dose (i.e. <45 v. ≥45 g/d), study design (i.e. parallel or crossover), the control food/diet (i.e. whether it was provided or if subjects were simply instructed to avoid nuts), the duration of the study (i.e. ≥12 weeks v. 4 to <12 weeks (hereinafter referred to as <12 weeks)), and of baseline blood lipid level. Baseline blood lipid levels were categorised dichotomously as 'optimal' or 'not optimal', based on the targets established in the National Cholesterol Education Program Adult Treatment Panel III guidelines (i.e. optimal blood lipid levels were defined as: LDL-C < 2·59; TC < 5·17; HDL-C ≥ 1·03; TAG < 1·69 mmol/l). For crossover studies, the categorisation was based on the reported baseline lipid level; for parallel studies, the categorisation was based on the average of the baseline lipid levels that were reported for each group, weighted by the sample size of each group. Subgroup analyses were conducted when there were at least three strata available for pooling.

Literature search results and overview of included studies
The three literature searches resulted in the identification of 1697 titles, of which eighteen publications met all of the inclusion criteria and none of the exclusion criteria (Fig. 1).

Almond interventions
Across all strata, the average daily intake of almonds ranged from 20 to 113 g/d, and the duration of the almond consumption period ranged from 4 weeks to 18 months. Almonds were required to be consumed every day in all studies except two, in which 28 g (1 oz) of almonds were required to be consumed 5 d per week (24) or 43 g (1·5 oz) of almonds were required to be consumed five to seven times weekly (10) .
Whole, raw (unblanched, unsalted) almonds were consumed in nine strata (Abazarfard et al. (9) ; Sweazea et al. (10) ; Damasceno et al. (19) ; Jenkins et al. strata 1 and 2 (20) ; Ruisinger et al. (27) ; Spiller et al. strata 1 and 2 (17) ; Wien et al. (28) ). In five strata (Cohen & Johnston (24) ; Tan & Mattes strata 1 to 4 (22) ), the almonds that were consumed by the subjects were not specifically described by the study authors as raw, unblanched almonds; however, based on the reported energy value of the almonds, it was determined that the almonds were most probably raw, unblanched almonds. A variety of almonds, namely whole (raw), roasted, and flavoured almonds were consumed in one stratum (Foster et al. (16) ); dry, roasted almonds were consumed in one stratum (Wien et al. (23) ); and almond powder was consumed in one stratum (Tamizifar et al. (21) ). The types of almonds used were not specified in two strata (Kurlandsky & Stote strata 1 and 2 (14) ). In eight strata, all meals and snacks were provided and the almonds were said to have been consumed as a snack (Berryman et al. (11 ) or incorporated into the meals and snacks (Jia et al. strata 1 and 2 (15) ; Li et al. (25) ; Lovejoy et al. strata 1 and 2 (26) ; Sabaté et al. strata 1 and 2 (18) ). The form of almonds that was used was described only by Berryman et al. (11) , who reported administering unsalted, whole, natural almonds with skins, and by Jia et al. strata 1 and 2 (15) , who reported using almond powder. In the remaining five strata in which all meals and snacks were provided (Li et al. (25) ; Lovejoy et al. strata 1 and 2 (26) ; Sabaté et al. strata 1 and 2 (18) ), it is assumed that whole almonds, almond pieces and ground almonds were used to prepare the meals.

Study quality
Based on Health Canada's quality appraisal tool, all of the studies were considered to be 'higher quality' (8) . Across all eighteen publications, the most commonly identified limitations included the lack of reporting on allocation concealment (n 16), the lack of reporting on the method of randomisation and thus the 'appropriateness' of the randomisation method, which also is a quality factor, could not be determined (n 13), as well as the lack of reporting of an intent-to-treat analysis (n 13).

Effects of almonds on fasting blood lipids
The fasting blood lipids that were assessed in each of the twenty-seven strata, as well as other information pertinent to the subgroup analyses (i.e. study design, almond intake, whether a control food/diet was provided, study duration, and baseline blood lipid levels), are summarised in Table 2. In the study by Lovejoy et al. strata 1 and 2 (26) , baseline TAG levels were not reported; thus, a determination as to  Wien et al. (28) R, C, P 24 weeks 52 (M and F); with a medical diagnosis that could benefit from weight reduction***** 55 ± 2·0* 38·0 ± 1·0* Low-energy liquid formula + complex CHO (n 28)  whether TAG levels at baseline were or were not optimal could not be made. In the study by Spiller et al. strata 1 and 2 (17) , TC, LDL-C, HDL-C and TAG were assessed at baseline and at the end of treatment; however, only the results for TC and LDL-C at the end of treatment were reported. Thus, the HDL-C and TAG results could not be included in the meta-analyses, and in order to include the results related to TC and LDL-C in the meta-analyses, the mean difference in the effect for TC and LDL-C had to be calculated by subtracting the end-of-treatment value in the control group from the end-of-treatment value in the almond group (as opposed to subtracting the changes from baseline in the control group from the changes from baseline in the almond group, which was done for all other parallel studies). The effects of almonds on fasting TC levels were assessed in all twenty-seven strata. The daily almond intake was 45 g or greater in 63 % of the strata, the design was crossover in 37 % of the strata, the baseline fasting TC level was not optimal in 52 % of the strata, a control food/diet was provided in 52 % of the strata, and the study duration was <12 weeks in 78 % of the strata ( Table 3). As can be seen in Table 3 and Fig. 2, the reduction in TC was statistically significant when data from all twenty-seven strata were pooled (−0·153 mmol/l; 95 % CI −0·235, −0·070 mmol/l; P < 0·001). As there was no publication bias identified, no adjustment to these values was made. In all of the subgroup analyses, the pooled effect sizes for TC were negative. Statistical significance was observed when pooling those strata in which the almond dose was ≥45 g/d, the baseline TC level was not optimal, the study design was either parallel or crossover, the control food/ diet either was or was not provided, and the duration of the almond intervention period was <12 weeks ( Table 3).
The effects of almonds on fasting LDL-C were assessed in twenty-five strata. The daily almond intake was 45 g or greater in 60 % of the strata, the design was crossover in 40 % of the strata, the baseline fasting LDL-C level was not optimal in 80 % of the strata, a control food/diet was provided in 48 % of the strata, and the study duration was <12 weeks in 76 % of the strata ( Table 3). As can be seen in Table 3 and Fig. 3, the reduction in LDL-C was statistically significant when data from all twenty-five strata were pooled (−0·124 mmol/l; 95 % CI −0·196, −0·051 mmol/l; P = 0·001). As there was no publication bias identified, no adjustment to these values was made. In all of the subgroup analyses, the pooled effect sizes for LDL-C were negative, except for when the five strata in which the baseline blood lipid level was optimal were pooled. For the subgroup analyses, statistically significant pooled reductions in LDL-C were observed when pooling those strata in which the almond dose was ≥45 g/d, the baseline LDL-C level was not optimal, the study design was crossover, a control food/diet was provided, and the study duration was <12 weeks (Table 3). It should be noted that when the control-adjusted changes in LDL-C for the thirteen strata in which the control food/diet was not provided were pooled, the reduction in LDL-C approached statistical significance (P = 0·068).
Almond consumption was not associated with any significant effect on fasting HDL-C, either in the overall analysis in which all twenty-two strata were pooled or in any of the subgroup analyses (Table 3 and Fig. 4).
The effects of almonds on fasting TAG were assessed in twenty-five strata. The almond intake was 45 g or greater in 60 % of the strata, the design was crossover in 40 % of the strata, the baseline fasting TAG level was not optimal in 30 % of the strata, a control food/diet was provided in 48 % of the strata, and the study duration was <12 weeks in 76 % of the strata ( Table 3). As can be seen in Table 3 and Fig. 5, the reduction in TAG was statistically significant when data from all twenty-five strata were pooled (−0·067 mmol/l; 95 % CI −0·132, −0·002 mmol/l; P = 0·042). As there was no publication bias identified, no adjustment to these values was made. In all of the subgroup analyses, the pooled effect sizes for TAG were negative but not statistically significant, except for when the fifteen strata in which the study design was parallel were pooled, and the resultant pooled effect was a statistically significant reduction in TAG (−0·111 mmol/l; 95 % CI −0·204, −0·017; P = 0·020). Through additional sensitivity analyses, it was determined that this effect was dependent on the inclusion of the parallel study by Abazarfard et al. (9) , which included 100 females and was a relatively larger study.
With regards to the ratio of TC:HDL-C, when data from all nine strata were pooled, the reduction in the ratio was statistically significant (see Table 3 and Fig. 6). However, publication bias was detected. Using trim and fill, two studies were found to be missing to the right of the pooled effect size, and with these studies imputed, the pooled effect, though negative (i.e. favourable), was smaller and no longer statistically significant (Table 3). Results for the subgroup analyses were in the same direction of effect (i.e. the pooled effect was negative), with variable statistical significance. With regards to the ratio of LDL-C: HDL-C, when data from all ten strata were pooled, the reduction in the ratio was not significant (−0·089; 95 % CI −0·209, 0·031; P = 0·145) (see Table 3 and Fig. 7). As for the ratio of TC: HDL-C, publication bias was detected, and using trim and fill, two studies were found to be missing to the right of the pooled effect size for LDL-C:HDL-C. With these studies imputed, the pooled effect, though negative (i.e. favourable), was smaller and remained non-statistically significant (Table 3). Results for the subgroup analyses were in the same direction of effect (i.e. the pooled effect was negative), with variable statistical significance.

Discussion
In a meta-analysis that included five randomised control trials (and nine strata), Phung et al. (7) reported that almonds significantly reduce TC and have a strong trend towards reducing LDL-C (P = 0·05). Although Phung et al. (7) also reported a near-significant reduction in HDL-C (P = 0·08) and no effect on TAG, in our analyses, which are based on a total of eighteen publications and twenty-seven strata, the intake of almonds was associated with significant reductions in TC, LDL-C and TAG, and no effects on HDL-C.
In a meta-analysis and dose-response of sixty-one controlled intervention trials, which ranged in duration from 3 to 26 weeks, the consumption of nuts was associated with Table 3. Effects of almonds on blood lipid levels: results of meta-analyses of randomised controlled trials* † TC (mmol/l) LDL-C (mmol/l) HDL-C (mmol/l) TAG (mmol/l) TC:HDL-C LDL-C:HDL-C All strata n 27 n 25 n 22 n 25 n 9 n 10 n 10 n 9 n 10 n 3 Not optimal n 14 n 20 Crossover n 10 n 10 n 10 n 10 n 6 P < 0·001 P = 0·189 P = 0·081 P = 0·241 P < 0·001 Not provided n 13 n 13 n 13 n 13 n 3 † An assessment of publication bias was conducted for each lipid parameter, but only for the meta-analysis that included all strata. Publication bias was not identified for TC, LDL-C, HDL-C or TAG. For the ratio of TC:HDL-C, two studies were found to be missing to the right of the pooled effect, and with the missing studies imputed, the pooled effect was −0·120 (95 % CI −0·289, 0·050). For the ratio of LDL-C:HDL-C, two studies were found to be missing to the right of the pooled effect, and with the missing studies imputed, the pooled effect was −0·059 (95 % CI −0·175, 0·056).
significant reductions in TC, LDL-C, apoB and TAG, with greater effects observed with a nut intake of 60 g/d and in individuals with T2DM (29) . In contrast, in a Cochrane review, Martin et al. (30) reported that the intake of nuts had no effects on LDL-C or HDL-C (for TC and TAG, substantial heterogeneity precluded the pooling of results). The Cochrane assessment was based only on three publications (and four strata): Tey et al. (31) (who provided 42 g of hazelnuts to generally healthy male and female adults for 12 weeks); Abazfarad et al. (9) (who provided 50 g of almonds to overweight and obese premenopausal women for 3 months); and Tey et al. (32) (who provided 30 or 60 g of hazelnuts to overweight and obese male and female adults for 12 weeks). The main objective of the Cochrane review was to assess the effects of nut consumption on the primary prevention of CVD. In none of the studies was the incidence of heart disease assessed; thus, the effects of nut consumption on surrogate measures of CVD risk were examined. With such few studies,  11 journals.cambridge.org/jns and with three of the four strata conducted in generally healthy subjects, it is no surprise that effects on blood lipid levels could not be identified. Based on our systematic evidence-based review and meta-analyses, which included a total of eighteen publications and twenty-seven strata, the intake of almonds was associated with significant reductions in TC, LDL-C and TAG, and no effects on HDL-C. In all of the included studies, almonds or diets enriched with almonds were provided to the subjects; however, the control was variable across the studies. In thirteen of the twenty-seven strata, a control food or diet was not administered to the subjects and the subjects were instructed not to consume nuts. In fourteen of the twenty-seven strata, the subjects were provided with a control food or a control diet. It seems that almonds effectively improve TC and LDL-C, whether the comparison is made with the consumption of no almonds or with a control food or diet that, at the very least, was isoenergetic to the almond intervention. Based on the other subgroup analyses, it seems that the efficacy of almonds in improving TC and LDL-C is greatest with a daily almond intake of 45 g or more and in individuals whose TC and LDL-C levels at baseline are elevated (i.e. not optimal). Although pooling the results of the crossover studies (but not the parallel studies) resulted in a significant reduction in LDL-C, the results should be interpreted with caution, given that the crossover strata were comprised  predominantly of subjects whose baseline LDL-C levels were not optimal, while the parallel strata were comprised predominantly of subjects whose baseline LDL-C levels were optimal. Likewise, although pooling the results of the studies with a duration <12 weeks (but not the studies with a duration ≥12 weeks) resulted in significant reductions in both TC and LDL-C, the results should be interpreted with caution, given that there were only six strata with a duration ≥12 weeks, and in three of these six strata, the intake of almonds was <45 g/d and/or the subjects had optimal levels of TC and/or LDL-C at baseline (10,16,24) . Of the three parallel strata that were 12 weeks or longer in duration and in which the almond intake was ≥45 g/d and the baseline lipid levels were not optimal, there were significant or near-significant reductions in both TC and LDL-C in two of the strata (9,23) .
There is preliminary evidence that the consumption of almonds also leads to favourable changes in the ratio of TC: HDL-C; however, this lipid parameter was assessed only in nine strata, and the improvement was no longer statistically significant once an adjustment for publication bias was made. Preliminary evidence that the consumption of almonds leads to favourable changes in the ratio of TC:HDL-C is consistent with our findings of significant reductions in TC, with no effects on HDL-C. LDL-C as well as the ratio of TC: HDL-C are recognised as surrogate measures of CHD risk. Thus, it is plausible that by improving the blood lipid profile, the consumption of almonds would also be associated with significant reductions in the risk of CHD. While an intervention study on the effects of almonds on the risk of CHD has yet to be conducted, there is evidence from both prospective observational studies and a randomised controlled trial that the consumption of nuts, in general, is associated with significant reductions in the incidence of heart disease (discussed in the following paragraph).
In a meta-analysis of thirteen prospective studies (involving a total of 347 477 individuals and 6127 cases of coronary artery disease (CAD)), the relative risk (RR) of CAD was significantly reduced with the highest v. the lowest consumption of nuts (RR 0·660; 95 % CI 0·581, 0·748); moreover, the protective effect of nuts against the development of CAD was found to be dose-dependent, such that risk decreased by 5 % for every additional serving of nuts consumed per week (33) . In the PREDIMED (PREvención con DIeta MEDiterránea) study, which is a large, multi-centre primary prevention trial of the effects of three diets on CVD risk, the consumption of a Mediterranean diet supplemented with either extra-virgin olive oil or nuts resulted in significant reductions in CVD cases (including cases of myocardial infarction, stroke, or CVD death) relative to a control group instructed to consume a diet low in fat (34) . In a recent cross-sectional study involving 3 312 403 Americans undergoing screening for peripheral arterial disease, those who consumed nuts every day were 21 % less likely to have peripheral arterial disease relative to those who consumed nuts less than once per month; this statistically  13 journals.cambridge.org/jns significant finding was evident even after adjusting for several important variables, such as age, sex, smoking status, obesity, family history of CVD, diet and the presence of diet-related diseases such as diabetes (35) .
The mechanism by which the consumption of nuts leads to favourable alternations in blood lipid levels is not fully understood. Nuts are nutrient dense, have a favourable fatty acid profile, and contain other constituents such as sterols and flavonoids that, collectively, may be important in the mechanism of almonds in improving blood lipid levels and CHD risk. In addition, it is possible that the favourable changes in blood lipid levels with the consumption of almonds are related, at least in part, to concomitant improvements in body weight and body composition. In several of the studies that were included in our meta-analysis, there were significant reductions in body weight with the consumption of almonds relative to the control (11,20,28) . The study by Berryman et al. (11) is of particular interest, given that the subjects were provided with all of their foods during both the almond and control intervention periods, and the diets were rigorously controlled. There were statistically significant improvements in body weight, waist circumference and body composition (including abdominal fat mass) with the 6-week consumption of the almond diet relative to the control diet. Recently, it was demonstrated that the energy value of almonds calculated using the Atwater factors is 32 % greater than the actual energy that is metabolisable from almonds (36) . Similar observations have also been made for pistachios and walnuts (37,38) . This could explain the reductions in body weight that have been observed in some of the studies with the consumption of almonds. If not all of the 'calculated' energy in almonds is actually metabolisable, then in highly controlled experimental studies where the diets are prepared and provided to the study participants (such as in the study by Berryman et al. (11) ), the diets may not have been truly isoenergetic.
The consumption of nuts is encouraged in several 'heart-healthy' diets. Nuts are important constituents of the portfolio diet, which also consists of plant sterols, viscous fibres and soya protein (39) . Likewise, nuts are part of the Mediterranean diet, which consists also of fruits and vegetables, legumes, whole-grain cereals, olive oil, fish and seafood, herbs and spices, and moderate amounts of meat, dairy products and wine (40) . Nuts are constituents of the Palaeolithic diet, which also includes lean meat, fish, fruit, leafy and cruciferous vegetables, root vegetables and eggs (41) . The consumption of nuts, such as almonds, as part of a healthy diet should be encouraged in order to help in the maintenance of normal blood lipid levels and to reduce the risk of heart disease.