Monounsaturated Versus Polyunsaturated Dietary Fat and Serum Lipids
Abstract The objective of this study was to examine whether oils high in monounsaturated or polyunsaturated fats have a differential effect on serum lipid levels, using a meta-analytical approach. Fourteen studies (1983 through 1994) were identified that met six inclusion criteria, the primary criterion being that a study have at least two intervention diets that varied in monounsaturated and polyunsaturated fat content but were otherwise similar in total fat, saturated fat, fiber, and dietary cholesterol. Seven studies included a comparable high-saturated fat diet. Standardized effect sizes (observed treatment difference in mean end-point lipid levels, divided by the pooled SD) were calculated for individual studies, then individual effect sizes were pooled. The results indicated no significant differences in total, LDL, or HDL cholesterol levels when oils high in monounsaturated or polyunsaturated fats were compared directly. Triglyceride levels were modestly but consistently lower on the diets high in polyunsaturated fats (P=.05). Replacement of saturated fat with either monounsaturated or polyunsaturated fat led to significant decreases in total and LDL cholesterol (P<.001), and the pooled effect sizes were comparable for either type of unsaturate (effect sizes ranged from −0.64 to −0.68, ie, roughly a decrease of 0.65 mmol/L [25 mg/dL] relative to the high-saturated fat diets). Neither type of unsaturated fat significantly changed HDL cholesterol or triglyceride levels relative to the high-saturated fat diets. In conclusion, the evidence from this meta-analysis strongly indicates there is no significant difference in LDL or HDL cholesterol levels when oils high in either monounsaturated or polyunsaturated fats are exchanged in the diet. Any dietary recommendations for the use of one in preference to the other should be based on outcomes other than serum cholesterol levels.
- dietary fats, unsaturated
- lipoproteins, LDL cholesterol
- lipoproteins, HDL cholesterol
Reprint requests to Christopher D. Gardner, PhD, Stanford Center for Research in Disease Prevention, Stanford University School of Medicine, 1000 Welch Rd, Palo Alto, CA 94304-1825.E-mail firstname.lastname@example.org.
- Received February 5, 1995.
- Accepted September 15, 1995.
Minimizing dietary saturated fat intake has a beneficial impact on serum cholesterol levels. Calories from saturated fat can be replaced with either unsaturated fats or carbohydrates. With regard to unsaturated fats, it is currently unclear whether monounsaturates or polyunsaturates are preferable for optimal serum lipid levels.
Between 1957 and 1993, Keys et al1 2 3 and Hegsted et al4 5 6 independently developed several equations to predict changes in total-C1 2 3 4 5 6 and LDL-C6 that would accompany changes in dietary fat and cholesterol intake. Each of these predictive equations consistently suggested that polyunsaturated fats lowered serum total-C and LDL-C levels, whereas monounsaturated fats had a neutral effect. These findings were challenged in 1985 when Mattson and Grundy7 reported results indicating that both types of unsaturated fat lowered LDL-C levels when they replaced saturated fat. Their 1985 results also indicated that polyunsaturated fat lowered HDL-C, whereas monounsaturates did not. This study is now frequently cited as evidence that monounsaturates are preferred over polyunsaturates.8 9 10
Since the 1985 report by Mattson and Grundy,7 a substantial number of randomized dietary intervention trials have been conducted to test for a differential effect of monounsaturated and polyunsaturated fats on serum lipids.11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 The majority of these studies reported no significant differences in LDL-C, HDL-C, or triglyceride levels,11 12 13 14 15 16 17 19 20 21 22 24 27 whereas a few studies reported a mix of significant lipid differences favoring either monounsaturates18 26 28 or polyunsaturates.18 23 25 26 28 However, it is possible that an overall pattern of small but consistent differences in serum lipids exists but went undetected by many of the individual studies because of their limited statistical power.
The purpose of this investigation was to address the controversy regarding a differential effect of monounsaturates versus polyunsaturates on serum lipids. A meta-analytical approach was used to pool results from randomized dietary intervention trials that tested for a differential effect on serum lipids of a high-mono versus a high-poly fat diet, exchanging primarily oils while keeping total fat, saturated fat, fiber, and dietary cholesterol intake constant between the intervention diets. Relative to the individual studies, the pooled results increase the statistical sensitivity to detect overall patterns of differences in serum lipids.
Study Identification and Inclusion Criteria
Studies to be included in the meta-analysis were identified by a computerized search (medline, 1966 through 1994) for all publications with a combination of the key words “monounsaturate#” and “polyunsaturate#.” Further studies were identified from the bibliographies of review articles and other studies. Only published data from peer-reviewed journals were considered in these analyses.29 Fifteen studies7 12 13 14 16 17 18 19 20 21 22 23 26 28 30 were identified that met all of the following criteria: (1) dietary intervention trials with human subjects randomized to a high-mono and high-poly fat diet; (2) at least two intervention diets, similar in all respects except for levels of monounsaturated and polyunsaturated fat intake; (3) end-point data (mean±SD or SE) from dietary intervention periods for total-C, LDL-C, HDL-C, and triglycerides; (4) a minimum of 10 participants on each of the diets, to obtain a reasonably accurate estimate of within-group variance; (5) all subjects independent (ie, data from same subjects in multiple studies excluded); and (6) written in the English language. To avoid selection bias, all publications that met these originally established criteria were included in the analysis before examination of results.
Ten additional studies were identified that met most but not all of these criteria and that were therefore excluded from the analyses: four for not randomizing the subjects,31 32 33 34 four for small sample sizes,10 11 24 27 one for not providing the necessary variance data,15 and one for using some of the same subjects in two separate clinical trials.25 In the last case, of two studies reported by Wardlaw et al21 25 (1990 and 1991), only the 1991 trial21 was included because it was determined to have greater overall merit.35 36
All data and relevant study characteristics were extracted with authors’ names kept on separate pages. Mean lipid end-point data with standard variance estimates from each of the dietary intervention periods were commonly reported by all studies and were used in the present study.
The use of mean lipid end-point data in the three parallel design studies19 21 26 raises the issue of whether or not to adjust for lipid differences at baseline. Data were analyzed with and without adjustment for baseline differences, and the results and conclusions were the same by use of either approach. The results reported here are for the unadjusted means.
Two of the crossover studies reported significant findings during one period of the dietary intervention that were not replicated in the other period.23 28 In these cases, only data from the first intervention period were used, and the studies were treated as parallel designs. However, in one case,23 complete lipid end-point data were available for only nine participants in each of the randomized dietary groups from the first intervention period, and therefore this study was excluded because of small sample size. This exclusion left a total of 14 studies for the final meta-analysis.
Five of the studies with a crossover design indicated that there were no sequence or order effects of the diets.7 14 18 21 22 For the crossover design studies that did not report testing for carryover effects, we operated on the assumption that there were none.
The study design of Lichtenstein et al13 included two separate high-mono diets, with olive oil and canola oil. Presenting two separate sets of effect sizes for high-mono diets would violate the original criterion of independence of subjects across separate effect sizes. There was no apparent basis for picking one of the high-mono diets and eliminating the other from analyses. Therefore, the end-point lipid levels for the olive and canola oil diets were averaged, and a pooled SD was used.
Half of the studies reported data for serum lipid levels and half for plasma lipids. Because the standardized “effect sizes” used in the analyses here are unitless, it was unnecessary to make any adjustments for the known systematic differences in serum and plasma lipid levels. For simplicity, no distinction between serum and plasma lipids is made throughout the text, and both are referred to as “serum lipids.”
The method of meta-analysis that was used to determine the effect of dietary fat composition on serum lipids is based on the computation of a standardized effect size (d) as defined by Hedges and Olkin.37 Due to the nature of the reported data from the 14 studies, the effect size used here was the observed treatment difference in mean end-point serum lipid levels, divided by the pooled, within-group SDs of the serum lipid levels. Although the use of within-group SDs ignores the difference in statistical power between crossover and parallel design studies, the average within-subject variability that might have been used to address this difference was not available in most studies.
The study diets varied in their content of three different types of dietary fat: saturated, monounsaturated, and polyunsaturated. Effect sizes were calculated for three possible combinations of dietary fat contrasts: high-mono versus high-poly, high-mono versus high-sat, and high-poly versus high-sat. For each dietary fat contrast, a positive effect size indicates that mean end-point lipids were higher on the first-named type of fat and, conversely, a negative effect size indicates higher lipid levels on the second-named type of fat in the contrast. Thus, for example, a positive effect size for high-mono versus high-poly indicates that the mean serum lipid level is higher with the high-mono diet than with the high-poly diet. Note that if a positive effect size was calculated for a high-mono versus high-poly contrast, and the order of the calculation was reversed to be a high-poly versus high-mono contrast, the absolute magnitude of the effect size would be unchanged but the sign would be reversed to a negative effect size. An effect size of zero indicates no difference between groups.
Effect sizes were computed for individual studies and then pooled if appropriate. Effect sizes were first tested for homogeneity to determine if all studies estimated a common effect across studies (ie, no outliers).37 This test assessed whether the distribution of effect sizes was compatible with the assumption that interstudy differences were attributable to random sampling alone. If no statistically significant indication of heterogeneity was found (P>.05), effect sizes were pooled.37 Weighting of the individual study effect sizes was a function of sample size and represented the reciprocal of the variance of individual effect sizes (ie, larger studies are potentially more precise than smaller studies and therefore contribute more to the overall estimate of effect size).37 A sensitivity analysis was performed by removing the studies one by one from the analysis and recalculating the pooled effect sizes. No single study changed the results of the tests for homogeneity, nor did any given study change any of the findings presented. Ninety-five percent confidence intervals (95% CI) were determined for each individual study37 as well as for the pooled effect size.37
The effect size used here was a standardized difference in means and therefore a unitless value. This standardization process is fundamental to the synthesis of data from different clinical trials37 and provides control for some of the possible differences in serum-lipid assessment between individual study laboratories (eg, differences in scaling). Given that the use of effect sizes in analyses of serum-lipid data is less common than the use of serum concentrations in mmol/L or mg/dL, the following reference points are offered to put the magnitude of the effect sizes reported in the present study into perspective. Effect sizes of 0.1, 0.5, and 1.0 can be literally translated into serum lipid differences of one-tenth, one-half, and one full SD between two test diets. For the sake of this example, it is suggested that a typical SD for total-C, LDL-C, HDL-C, and triglyceride from these studies would be 0.91, 0.91, 0.26, and 0.56 mmol/L (35, 35, 10, and 50 mg/dL), respectively. Therefore, contrasting serum lipids between two diets, an effect size of 0.1, 0.5, and 1.0 would indicate a difference of 0.11, 0.55, and 1.1 mmol/L (3.5, 17.5, and 35 mg/dL), respectively, for either total-C or LDL-C, or a difference of 0.03, 0.13, and 0.26 mmol/L (1.0, 5.0, and 10.0 mg/dL), respectively, for HDL-C, or, finally, a difference of 0.06, 0.28, and 0.56 mmol/L (5.0, 25.0, and 50 mg/dL), respectively, for triglyceride.
Tables 1⇓ and 2⇓ provide a summary of selected characteristics of the 14 studies being reviewed. All were published between 1983 and 1994 and 11 since 1990. Altogether, the 14 studies included 273 men and 166 women between the ages of 18 and 78 years. Study participants were recruited from academic institutions,14 18 20 21 26 clinical settings,7 12 17 19 28 religious groups,22 and the military,16 and in two cases the recruitment pool was not identified.13 30 Five of the studies screened and recruited participants on the basis of relatively elevated levels of either total-C, LDL-C, or triglycerides,12 13 17 19 21 and 3 of the studies included only those participants with total-C levels below a particular cut point18 20 22 ; the cut points varied by study. The remaining 6 studies did not specify any particular inclusion or exclusion criteria. Eleven of the dietary studies were crossovers, and 3 were parallel in design. The dietary intervention periods for the high-mono and high-poly diets ranged from 21 to 84 days. Seven of the studies included a high-sat diet in the design at a comparable level of total fat, which was 6% to 16% kcal higher in saturated fat content relative to the unsaturated fat diets. On the high-mono versus high-poly diets, the level of substitution of monounsaturated for polyunsaturated fat ranged from 5% to 25% kcal, with a mean±SD of 9.7±6.4% kcal. The average level of monounsaturated fat content on the high-mono diets was 19.4±6.1% kcal, with a range of 14% to 30% kcal. The average level of polyunsaturated fat content on the high-poly diets was 15.9±7.1% kcal, with a range of 10% to 30% kcal. In 10 of the studies, virtually all foods were prepared and provided for the participants. The 4 exceptions to this included Foley et al,17 who provided meats, fish, and principal fat sources; Sirtori et al,28 who provided vegetable oils; Wood et al,14 who provided butter and margarines; and Dreon et al,22 who provided instructions by registered dietitians. Dietary fat composition was determined by chemical analysis of duplicate portions in 6 of the studies,12 13 18 19 21 26 and 6 studies calculated diet composition from databases.14 16 17 20 22 28 Mattson and Grundy7 and Becker et al30 used formula diets. All others used whole foods.
The manipulation of fat composition in the various study diets was achieved primarily by the substitution of oils (eg, by adding the oils to various sauces, dressings, and baked items). This enabled the investigators to keep all other major components of the intervention diets comparable (ie, saturated fat, carbohydrate, fiber, dietary cholesterol). Most of the high-mono diets incorporated either olive or canola oil into the diets, whereas most of the high-poly diets used either sunflower, corn oil, or safflower oil. The oils used are summarized in Table 2⇑.
In the tables and figures, the results from individual studies are presented in order of decreasing proportion of kilocalories exchanged between monounsaturates and polyunsaturates on the high-mono versus the high-poly diet (eg, the study with the largest exchange, 25% kcal, Bonanome et al,16 is listed first, at the top). Tables listing all of the serum lipid data from individual studies and all of the individual effect sizes and 95% CI generated from the lipid data are available in a technical report.36
High-Mono Versus High-Poly Comparisons
The 14 individual study effect sizes, the pooled effect sizes, and their 95% CI are presented in Fig 1⇓ for the high-mono versus high-poly contrasts for LDL-C, HDL-C, and triglycerides. Statistically significant findings (P<.05) are illustrated graphically by 95% CI that do not include zero. For LDL-C, 11 of the effect sizes were not significantly different from zero. One study shows a significant positive effect size (ie, LDL-C levels were higher on the high-mono diet),14 whereas 2 of the studies show a significant negative effect size (ie, LDL-C levels were higher on the high-poly diet).18 26 The individual study effect sizes for total-C in the high-mono versus high-poly dietary contrast are virtually identical to those for LDL-C (data not presented).
Similar to the findings for LDL-C, 10 of the 14 effect sizes for HDL-C were not statistically different from zero. Two studies show a significant positive effect size,7 21 and 2 of the studies show a significant negative effect size.26 28
For triglycerides, 9 of the 14 individual study effect sizes were not significantly different from zero, whereas the remaining 5 studies all had significant and positive effect sizes (ie, triglyceride levels were lower on the high-poly diets).18 20 22 26 28 Although the majority of the individual study effect sizes were not significantly different from zero, it is of interest to note the consistency of the pattern. Absolute levels of triglycerides were lower on the high-poly diets in 13 of 14 studies and were identical in 1 study.12
Individual study effect sizes were pooled to determine the best estimate of overall effect. There were no significant deviations from homogeneity for any of the four serum lipids individually (P>.05 for all tests for heterogeneity). Therefore, the effect sizes from all 14 studies were pooled. None of the pooled effect sizes were significantly different from zero. The specific pooled effect sizes and their 95% CI for the high-mono versus high-poly contrasts are as follows: total-C, 0.07 (−0.08, 0.21); LDL-C, −0.01 (−0.16, 0.14); HDL-C, 0.03, (−0.12, 0.17); and triglycerides, 0.14 (0.00, 0.29). These results indicate no significant differences in total-C, LDL-C, or HDL-C when monounsaturates and polyunsaturates are exchanged, primarily by using different plant and vegetable oils. Triglyceride differences between diets were of borderline statistical significance and were lower on the high-poly diets.
For purposes of interpretation, the magnitude of these pooled effect sizes can be roughly translated back into serum lipid differences in mmol/L (mg/dL) between diets. This can be achieved by multiplying the pooled effect sizes by “typical” standard deviations of 0.91, 0.91, 0.26, and 0.56 mmol/L (35, 35, 10, and 50 mg/dL) for total-C, LDL-C, HDL-C, and triglycerides, respectively. The pooled effect sizes represent the best estimate of overall effect and suggest a difference of 0.06, −0.01, 0.01, and 0.08 mmol/L (2.4, −0.4, 0.3, and 7.0 mg/dL), respectively. (Negative values indicate lower serum lipid levels on the high-mono diets, and positive values indicate lower levels on the high-poly diets.) If we take the lower end of the 95% CI into consideration, these differences would be −0.07, −0.15, −0.03, and 0.0 mmol/L (−2.8, −5.6, −1.2, and 0.0 mg/dL), respectively, and differences for the upper end of the 95% CI would be 0.19, 0.13, 0.04, and 0.16 mmol/L (7.4, 4.9, 1.7, and 14.5 mg/dL) for total-C, LDL-C, HDL-C, and triglycerides, respectively.
Subset of Studies With Greatest Monounsaturate Versus Polyunsaturate Substitution
The studies included in this meta-analysis used a wide range of monounsaturated versus polyunsaturated fat substitutions, 5% to 25% of kilocalories. It could be argued that if there actually were a small but differential effect of monounsaturated and polyunsaturated fats on serum lipids, there would be greater sensitivity to detect this when the magnitude of the monounsaturated versus polyunsaturated differential in the study diets was greatest. This possibility was addressed in a post hoc analysis of the 5 studies that reported monounsaturate for polyunsaturate substitutions of 10% to 25% kcal on the high-mono versus high-poly study diets (17.5±6.5% kcal).7 16 17 21 30 Note that for these 5 studies, average levels of monounsaturated and polyunsaturated fat, respectively, were 26.7±5.3% kcal and 7.4±3.0% kcal on the high-mono diets and 9.2±4.3% kcal and 24.3±5.0% kcal on the high-poly diets. The pooled effect sizes and 95% CI were 0.24 (−0.07, 0.55), 0.14 (−0.17, 0.45), 0.26 (−0.04, 0.57), and 0.12 (−0.19, 0.43) for total-C, LDL-C, HDL-C, and triglycerides, respectively. Although none of the pooled effect sizes from this subset of studies were statistically significant, the magnitude of the pooled effect sizes for total-C, LDL-C, and HDL-C were all larger for the subset of 5 studies than for the original group of 14 studies. In each case, the trend was for lower serum levels on the high-poly diets. Therefore, although this post hoc analysis did not reach statistical significance, given the range of the CIs, it is possible that at extremely and perhaps unrealistically high levels of unsaturated fat intakes, polyunsaturates may lower total-C, LDL-C, and HDL-C to a greater extent relative to monounsaturates.
Unsaturated Versus Saturated Fat
Additional post hoc analyses were conducted to address two questions related to the replacement of saturated fats with unsaturated fats. First, how sensitive is the meta-analytical method approach used in the present study for detecting differences when clear differences exist? Here, we assumed that the well-established LDL-C–lowering effect of decreasing saturated fat in the diet would be evident. Second, what is the effect on serum lipids of replacing saturates with monounsaturates relative to replacing saturates with polyunsaturates?
Although the original inclusion criteria required that all of the studies have at least two comparable intervention diets that varied only in the proportion of kilocalories from monounsaturates and polyunsaturates, many of the studies also included a high-sat diet in the study design. It therefore was possible to pool the data from a subset of the original 14 studies to investigate serum lipid levels on a high-sat diet versus the two unsaturated fat diets.
Seven of the 14 studies were excluded from this additional analysis. Four of the studies were excluded because either there was no high-sat diet in the design or its composition was not reported.12 14 19 22 Two of the studies were excluded because the high-sat diet was not only higher in saturated fat but also higher in total fat than the high-mono and high-poly diets.13 20 The study of Sirtori et al28 was excluded because the substitution of saturates for unsaturates was negligible (only 2% kcal) on baseline versus the high-mono and high-poly diets.
Of the seven studies included, the order of the high-sat diet was randomized in sequence with the unsaturated diets in three of the studies.7 14 30 In three of the studies,18 21 26 the high-sat diet was the baseline diet that always preceded the unsaturated diets, and in the study of Foley et al,17 there were two high-sat diet periods, one that preceded and another that followed the unsaturated diets, and serum lipid levels were reported only for the second high-sat diet period.
The seven individual study effect sizes, the pooled effect sizes, and their 95% CI are presented in Fig 2⇓ for the high-mono versus high-sat contrasts and the high-poly versus high-sat contrasts for LDL-C, HDL-C, and triglycerides. When a particular study did not yield an effect size, eg, when there was no appropriate high-sat diet in the study design, that study’s position was left blank in Fig 2⇓. For LDL-C, the data from six of the seven studies yielded significant effect sizes for both types of unsaturated fat. The effect sizes of the seventh study, by Foley et al,17 were borderline significant: −0.22 (95% CI −0.44, 0.00) and −0.22 (95% CI −0.45, 0.00) for the high-mono versus high-sat and the high-poly versus high-sat contrasts, respectively. In every case, the effect sizes were negative, indicating lower LDL-C levels on the high–unsaturated fat diets. The individual study effect sizes for total-C are virtually identical to those for LDL-C (data not presented). For HDL-C, the majority of the 95% CI include zero. The two significant, negative, HDL-C effect sizes for the high-mono diets indicated mean differences between dietary periods of less than one-half SD: −0.4317 and −0.31.26 Three of the high-poly diets produced significant effect sizes in the range of one-half SD (−0.59 to 0.55), indicating either higher30 or lower7 17 HDL-C levels on the high-poly diets relative to the high-sat diets. None of the high-mono versus high-sat effect sizes for triglycerides were significantly different from zero. Only one of the seven high-poly versus high-sat effect sizes for triglycerides was significantly different from zero, and the magnitude of the effect size (−0.24)18 indicated a difference in mean triglyceride levels of less than one-fourth SD.
For the high-mono versus high-sat and the high-poly versus high-sat contrasts, there were no significant deviations from homogeneity among the seven individual study effect sizes for any of the four serum lipids individually (P>.05 for all tests for heterogeneity). Therefore, the effect sizes from all seven studies were pooled. The specific pooled effect sizes and their 95% CI for the high-mono versus high-sat contrast are as follows: total-C, −0.64 (−0.84, −0.44); LDL-C, −0.66 (−0.85, −0.46); HDL-C, −0.07 (−0.27, 0.13); and triglycerides, −0.06 (−0.25, 0.14). The corresponding values for the high-poly versus high-sat contrast are as follows: total-C, −0.68 (−0.87, −0.49); LDL-C, −0.66 (−0.85, −0.46); HDL-C, −0.13 (−0.32, 0.07); and triglycerides, −0.13 (−0.33, 0.07).
The pooled effect sizes indicate that when levels of total fat are held constant and saturated fat is replaced by either monounsaturated or polyunsaturated fat, both total-C and LDL-C are significantly lowered (all P<.001). Even though only seven studies were considered, this meta-analytical approach was able to demonstrate a highly significant statistical difference. In addition, the magnitude of the LDL-C–lowering effect of replacing saturates with monounsaturates was virtually identical to that of replacing saturates with polyunsaturates. Neither HDL-C nor triglyceride levels were significantly changed when saturates were replaced with either monounsaturates or polyunsaturates. The considerable overlap of 95% CI for the high-sat versus high-mono and the high-sat versus high-poly contrasts (Fig 3⇓) indicates that the effects on serum lipids of replacing saturated fat with either monounsaturated or polyunsaturated fats (when levels of total fat are held constant) are statistically indistinguishable.
Further Subset Analyses
Although all of the studies included in the meta-analysis shared a number of important common features, there were also notable differences in study population characteristics and study design. These factors involved the inclusion in the various studies of the following: (1) either men only, women only, or both sexes; (2) individuals with either normal or elevated serum lipid levels or a mixture of both; (3) participants of various age ranges; (4) intervention diets comprising either whole foods or formulas; and (5) a range of scientific merit. Given these potential differences and the identification of only 14 studies that met all of our inclusion criteria, it was not possible to definitively address each of these differences. However, to the extent that it was possible, each of these factors was examined separately.36 As an example, if only the 8 studies that used whole food–treatment diets and provided all of these foods to participants12 13 16 18 19 20 21 26 were considered (ie, excluding the 2 studies that used formula diets7 30 and the 4 studies that had limited control over dietary intake of participants14 17 22 28 ), the effect on results was negligible. The data do not suggest any notable subsets of study participants or dietary intervention approaches that are inconsistent with the general findings.
Keys and Hegsted Equations
Based on the equations of Keys et al1 3 and Hegsted et al,4 6 the observed results from these 14 studies appear to be inconsistent with the predicted serum lipid responses to these high-mono versus high-poly exchanges. The predicted outcome would be an overall lowering of total-C and LDL-C with the high-poly diets relative to the high-mono diets. To directly quantify the magnitude of this apparent discrepancy, the Keys and Hegsted equations were used to calculate predicted serum lipid differences as well as predicted effect sizes between the high-mono and high-poly diets from the 14 studies. To calculate predicted effect sizes from the predicted serum lipid differences, the actual SDs from the observed data were used as the denominators in the equations. The predicted total-C and LDL-C levels were consistently lower on all of the high-poly diets relative to the high-mono diets. When the predicted differences were used, the pooled effect sizes were also significant for total-C and LDL-C, indicating lower predicted lipid levels on the high-poly diets overall (Table 3⇓). Although the predicted lipid differences were significant, the observed lipid differences, as reported earlier, were not significant.
The results of this meta-analysis of 14 studies indicate that there are no significant differences in total-C, LDL-C, or HDL-C levels in response to the dietary exchange of oils high in monounsaturated or polyunsaturated fats. This absence of an effect was observed across a wide range (5% to 25% kcal) of monounsaturate versus polyunsaturate exchanges in the various studies and for a diverse set of participants (eg, wide age range, wide range of baseline serum lipid levels). For serum triglycerides, consistently lower levels were observed within this group of studies for the high-poly diets, and the overall pooled effect size was of borderline significance (P=.05).
For purposes of interpretation, the magnitude of the pooled effect sizes can be roughly translated back into serum lipid differences in mmol/L (mg/dL). If we take this approach, the estimated LDL-C response to an average 9.7±6.4% kcal substitution of monounsaturates for polyunsaturates would be a decrease of 0.01 mmol/L (0.4 mg/dL), with the 95% CI ranging from a decrease of 0.15 mmol/L (5.6 mg/dL) to an increase of 0.13 mmol/L (4.9 mg/dL). For HDL-C, this same level of monounsaturated substitution would yield an increase of 0.01 mmol/L (0.4 mg/dL), with a 95% CI ranging from a decrease of 0.03 mmol/L (1.2 mg/dL) to an increase of 0.04 mmol/L (1.7 mg/dL). For triglycerides, there would be an increase of 0.08 mmol/L (7.0 mg/dL) with monounsaturated fat, with a 95% CI ranging from no change (0.0 mmol/L or mg/dL) to an increase of 0.16 mmol/L (14.5 mg/dL). Given the relatively extreme and unrealistic magnitude of the average caloric exchange between monounsaturated and polyunsaturated fats from this set of studies, it could be inferred that quantitatively smaller and more realistic levels of monounsaturated and polyunsaturated fat substitutions would have even smaller effects on serum lipid levels.
We report these conclusions having assumed that the clinical trials included in this meta-analysis are at least representative and at best a complete group of all of the published clinical trials that met the inclusion criteria for the primary analysis. Although the possibility of a publication bias toward positive findings is generally a concern in the interpretation of meta-analytical results, in this case the existence of unpublished data showing no differences in serum lipids on high-mono versus high-poly diets would only strengthen our conclusions.
In addition to the primary contrast of monounsaturated versus polyunsaturated fats in 14 studies, several ancillary post hoc analyses were carried out to explore issues of secondary interest. First, pooled effect sizes were determined for the subset of 5 studies with the most extreme substitutions of dietary monounsaturated for polyunsaturated fats (≥10% kcal). For this subset of studies, there were still no significant differences in serum lipids for high-mono versus high-poly diets, but there was a suggestion that total-C, LDL-C, and HDL-C were all lower on the high-poly diets. The magnitude of the small triglyceride-lowering effect of polyunsaturates was comparable for both the subset of 5 studies and the original 14 studies. The interpretation of these combined trends is problematic, since an optimal lipid profile includes lower LDL-C and triglycerides but higher HDL-C levels. These post hoc findings suggest that even at extreme dietary shifts between types of unsaturated fat, primarily with use of plant and vegetable oils, the overall impact on an optimal serum lipid profile may be negligible.
Second, results from post hoc analyses of seven studies with a high-saturated fat diet included in their design indicate that replacing saturated fat intake with either monounsaturates or polyunsaturates lowers total-C and LDL-C levels significantly. The magnitude of the effect sizes suggests that total-C and LDL-C were lowered by roughly 0.66 mmol/L (25 mg/dL) when saturated fat intake was lowered by a mean of 10.1±4.0% kcal and replaced by unsaturates. It should be noted that although only seven studies were considered, the pooled effect sizes reached the statistically significant probability level of P<.001 for total-C and LDL-C. In contrast, none of the pooled effect sizes were significantly different from zero for HDL-C or triglycerides. These findings suggest that when calories from saturated fat are replaced with calories from monounsaturated or polyunsaturated fats, primarily from oils, the overall serum lipid profile is improved, comparably so, regardless of the type of unsaturated fat. The limitations of these post hoc analyses include the smaller number of studies involved and the fact that the high-sat diet was either a baseline or a follow-up diet and not part of the randomized diet sequence for four of the seven studies included.
There are at least two possible explanations for the discrepancy between our results indicating no difference in HDL-C between high-mono and high-poly diets and the finding reported by Mattson and Grundy7 that polyunsaturates but not monounsaturates lower HDL-C. First, the Mattson and Grundy finding may apply only when extremely high levels of monounsaturated and polyunsaturated fats are consumed. In their design, which used formula diets, monounsaturates contributed 30% of kilocalories in the high-mono diet and polyunsaturates contributed 30% of kilocalories in the high-poly diet. In contrast, for the remaining 13 studies included in the meta-analysis, of which 12 used whole foods and one used formula diets, the average level of intake of monounsaturates in the high-mono diets was 18.7±5.7% kcal, and the average level of intake of polyunsaturates in the high-poly diets was 14.8±6.2% kcal.
A second explanation for the discrepancy with the Mattson and Grundy7 finding could be that their results apply only to individuals with the particular characteristics of their study population. In fact, even Mattson and Grundy reported no significant differences in HDL-C between diets for the entire group of 20 men in their study. The lower HDL-C found in men who ate the high-poly diet in that study was found only in a subset of 12 men who, while on the high-sat diet, had relatively lower triglyceride levels (1.6±0.5 versus 4.9±1.8 mmol/L [143.0±43.8 versus 433.8±163.7 mg/dL]) and relatively higher LDL-C levels (4.3±1.3 versus 2.9±1.0 mmol/L [165.5±49.9 versus 112.4±36.9 mg/dL]) than the remaining 8 men. However, among the other 13 studies in the meta-analysis, 4 studies selected participants with a similar lipid profile—normal triglycerides (mean <1.7 mmol/L [150 mg/dL]) and high LDL-C (mean >3.4 mmol/L [130 mg/dL])—and none of these 4 studies reported any significant HDL-C lowering on a high-poly diet relative to a high-mono diet.13 14 17 21
Our findings of no differences in serum total-C or LDL-C levels between high-mono and high-poly diets are also in disagreement with the work of Keys et al1 2 and Hegsted et al,4 6 whose equations suggest that polyunsaturates actively lower total-C and LDL-C whereas monounsaturates are neutral. In an earlier response to reports that monounsaturates effectively lower LDL-C when replacing saturates, Hegsted et al6 postulated that “. . . the fall in serum cholesterol frequently observed when oils high in monounsaturated fatty acids are incorporated in the diet is due to changes in the dietary content of the saturated and polyunsaturated fatty acids.” We have addressed this possibility by including in the meta-analysis only those dietary intervention studies that kept levels of saturated fat (as well as levels of total fat and dietary cholesterol) constant between the high-mono and high-poly diets. Having applied the Keys and Hegsted equations to the 14 studies included in the current analyses, we found that the predicted total-C–lowering and LDL-C–lowering effect of the high-poly diets overestimated the observed serum lipid responses. It is possible that these predictive equations are most useful when applied to dietary manipulations more similar to those used in the same studies that generated the regression equations—studies that involved substantial shifts in total fat, saturated fat, and dietary cholesterol, and that in many cases included greater quantities of saturated fat and cholesterol. In the specific case where only monounsaturated and polyunsaturated fats are exchanged while total fat, saturated fat, and dietary cholesterol levels are kept constant, and at the levels described in Table 1⇑, the predictive equations appear to exaggerate the observed responses.
Our results are similar to those reported earlier by Mensink and Katan.38 These investigators used data from 27 published studies to estimate the effects of various dietary fat interventions on serum lipid levels. The 27 individual trials included dietary manipulations of total fat, total carbohydrate, saturated fat, monounsaturated fat, polyunsaturated fat, and fiber intake. Multiple regression analyses were used to isolate the independent effects of saturated, monounsaturated, and polyunsaturated fat. The regression coefficients generated by this analysis38 suggested that both LDL-C and HDL-C levels would be slightly lower if there were a substantial substitution of polyunsaturates for monounsaturates, but, as in the present study, the differences were not statistically significant. Using the average substitution of polyunsaturates for monounsaturates in the current meta-analysis of 14 studies (9.7% kcal), we calculated the serum lipid responses predicted by the regression coefficients of Mensink and Katan.38 Unlike our experience with the Keys1 2 3 and Hegsted4 5 6 equations, the serum lipid responses predicted by the Mensink and Katan equations fell well within the 95% CI of the observed effect sizes in the current analyses. The better agreement with the Mensink and Katan38 results may be partially due to the similarity between that study and the present one in selecting only intervention trials that kept dietary cholesterol constant.
Although the Mensink and Katan38 conclusions are basically in agreement with our own, there are several important differences between their analyses and ours. First, their inclusion criteria were less stringent than our own because their focus was broader: they included studies with dietary manipulations of total fat, carbohydrate, and fiber as well as of unsaturates. Second, because their data sources involved the variety of dietary changes noted above, they used multiple regression analyses to statistically isolate the effects of monounsaturates and polyunsaturates. In the present meta-analysis, the effect of these unsaturates was directly isolated by selecting only those studies that held the other factors constant. Finally, because the present study is more recent, it has the advantage of being able to include seven recent investigations of monounsaturated versus polyunsaturated fat that were published since 199212 13 14 16 17 19 20 and were not available for the Mensink and Katan report.38
A limitation of the current meta-analysis may have involved pooling the results from a set of studies that differed in certain study population characteristics (eg, normolipidemic and hyperlipidemic) and various aspects of study design (eg, whole-food and formula diets). The available published data are insufficient to definitively state whether the general findings of the present investigation apply equally to all possible subgroups (eg, there were no data for hypertriglyceridemic women). To the extent that it was possible, we grouped various sets of studies by age, sex, baseline serum lipid status, food source, and scientific merit. These data do not suggest any notable subsets of study participants or dietary intervention approaches that are inconsistent with the general findings.36
The fact that the studies included in this meta-analysis used primarily plant and vegetable oils to isolate and manipulate the dietary content of monounsaturated and polyunsaturated fat leads us to acknowledge other limitations in the interpretation of these results. First, it is known that some oils differ in their content of factors other than monounsaturates and polyunsaturates that modify cholesterol levels. For example, olive oils can have substantially more squalene than other oils. Micronutrient contents of the oils were not provided in these studies and therefore could not be taken into account. In addition, it is possible that foods, as opposed to oils, that are good sources of monounsaturated and polyunsaturated fats may be associated with other micronutrients or dietary factors that have a differential effect on serum lipids.
There may be important physiological consequences of altering dietary unsaturated fat composition that have not been addressed in this meta-analysis. There is evidence to suggest that the intake of monounsaturated fatty acids relative to diets rich in polyunsaturates increases the resistance of serum low-density lipoproteins to in vitro peroxidation.10 16 There are also reports, with mixed findings, that the two unsaturates may or may not have a differential influence on apolipoproteins,11 13 14 18 20 26 lipoprotein subclasses,7 11 18 22 blood glucose,12 28 indexes of thrombosis,27 28 39 40 and cancer.41 42 43 Another issue not addressed in this meta-analysis is the distinction between n-6 polyunsaturates (plant sources) and n-3 polyunsaturates (marine sources). Although n-3 content was not reported in all studies, it is possible that the modest triglyceride-lowering effect observed with the high-poly diets may have been attributable at least partially to a higher n-3 polyunsaturate content of those diets.44
In conclusion, the evidence from this meta-analysis of 14 clinical trials indicates that there is no significant difference in serum total-C, LDL-C, or HDL-C levels between diets relatively high in monounsaturated versus polyunsaturated fats when fat intake is derived primarily from common plant and vegetable oils. Triglyceride levels may be slightly lowered by polyunsaturates. Future research in this area should take into consideration the now substantial number of trials that have replicated this finding. Further work may be needed to address the possibility that previously unidentified subgroups of the population are exceptions to this general pattern, and in that case it will be important to clearly define participant selection and inclusion criteria and to avoid the statistical power limitations of small sample sizes in study designs. However, it now may be more appropriate to shift the research emphasis in this area toward outcomes other than serum cholesterol levels on which to base dietary recommendations for the use of oils high in one unsaturated fat in preference to another.
Selected Abbreviations and Acronyms
Dr Gardner was supported by an institutional national research service award (no. 5 T32 HL07034) from the National Heart, Lung, and Blood Institute. We would like to acknowledge the contributions of Ingram Olkin, PhD, Steve Fortmann, MD, Sally Mackey, MS, RD, and Marilyn Winkleby, PhD, for their critical review of the text, and Michaela Kiernan, PhD for her assistance with editorial review and the figures.
Keys A. Serum-cholesterol response to dietary cholesterol. Am J Clin Nutr. 1984;40:351-359.
Hegsted DM. Serum-cholesterol response to dietary cholesterol: a re-evaluation. Am J Clin Nutr. 1986;44:299-305.
Hegsted DM, Ausman LM, Johnson JA, Dallal GE. Dietary fat and serum lipids: an evaluation of the experimental data. Am J Clin Nutr. 1993;57:875-883.
Mattson FH, Grundy SM. Comparison of effects of dietary saturated, monounsaturated, and polyunsaturated fatty acids on plasma lipids and lipoproteins in man. J Lipid Res. 1985;26:194-202.
Artaud-Wild SM, Connor SL, Sexton G, Connor WE. Differences in coronary mortality can be explained by differences in cholesterol and saturated fat intakes in 40 countries but not in France and Finland: a paradox. Circulation. 1993;88:2771-2779.
Reaven P, Parthasarathy S, Grasse BJ, Miller E, Steinberg D, Witztum JL. Effects of oleate-rich and linoleate-rich diets on the susceptibility of low density lipoprotein to oxidative modification in mildly hypercholesterolemic subjects. J Clin Invest. 1993;91:668-676.
Ginsberg HN, Karmally W, Barr SL, Johnson C, Holleran S, Ramakrishnan R. Effects of increasing dietary polyunsaturated fatty acids within the guidelines of the AHA Step 1 Diet on plasma lipid and lipoprotein levels in normal males. Arterioscler Thromb. 1994;14:892-901.
Nydahl MC, Gustafsson I-B, Vessby B. Lipid-lowering diets enriched with monounsaturated or polyunsaturated fatty acids but low in saturated fatty acids have similar effects on serum lipid concentrations in hyperlipidemic patients. Am J Clin Nutr. 1994;59:115-122.
Lichtenstein AH, Ausman LM, Carrasco W, Jenner JL, Gualtieri LJ, Goldin BR, Ordovas JM, Schaefer EJ. Effects of canola, corn and olive oils on fasting and postprandial plasma lipoproteins in humans as part of a National Cholesterol Education Program Step 2 diet. Arterioscler Thromb. 1993;13:1533-1542.
Wood R, Kubena K, O’Brien B, Tseng S, Martin G. Effect of butter, mono- and polyunsaturated fatty acid-enriched butter, trans fatty acid margarine, and zero trans fatty acid margarine on serum lipids and lipoproteins in healthy men. J Lipid Res. 1993;34:1-11.
Kris-Etherton PM, Derr J, Mitchell DC, Mustad VA, Russell ME, McDonnell ET, Salabsky D, Pearson TA. The role of fatty acid saturation on plasma lipids, lipoproteins and apolipoproteins, I: effects of whole food diets high in cocoa butter, olive oil, soybean oil, dairy butter and milk chocolate on the plasma lipids of young men. Metabolism. 1993;42:121-129.
Bonanome A, Pagnan A, Biffanti S, Opportuno A, Sorgato F, Dorella M, Maiorino M, Ursini F. Effect of dietary monounsaturated and polyunsaturated fatty acids on the susceptibility of plasma low density lipoproteins to oxidative modification. Arterioscler Thromb. 1992;12:529-533.
Valsta LM, Jauhiainen M, Aro A, Katan MB, Mutanen M. Effects of a monounsaturated rapeseed oil and a polyunsaturated sunflower oil diet on lipoprotein levels in humans. Arterioscler Thromb. 1992;12:50-57.
Wahrburg U, Martin H, Sandkamp M, Schulte H, Assmann G. Comparative effects of a recommended lipid-lowering diet vs a diet rich in monounsaturated fatty acids on serum lipid profiles in healthy young adults. Am J Clin Nutr. 1992;56:678-683.
Wardlaw GM, Snook JT, Lin M-C, Puangco MA, Kwon JS. Serum lipid and apolipoprotein concentrations in healthy men on diets enriched in either canola oil or safflower oil. Am J Clin Nutr. 1991;54:104-110.
Berry EM, Eisenberg S, Haratz D, Friedlander Y, Norman Y, Kaufmann NA, Stein Y. Effects of diets high in monounsaturated fatty acids on plasma lipoproteins: The Jerusalem Nutrition Study: high MUFAs vs high PUFAs. Am J Clin Nutr. 1991;53:899-907.
Chan JK, Bruce VM, McDonald BE. Dietary α-linolenic acid is as effective as oleic acid and linoleic acid in lowering blood cholesterol in normolipidemic men. Am J Clin Nutr. 1991;53:1230-1234.
Wardlaw GM, Snook JT. Effect of diets high in butter, corn oil or high-oleic acid sunflower oil on serum lipids and apolipoproteins in men. Am J Clin Nutr. 1990;51:815-821.
McDonald BE, Gerrard JM, Bruce VM, Corner EJ. Comparison of the effect of canola oil and sunflower oil on plasma lipids and lipoproteins and on in vivo thromboxane A2 and prostacyclin production in healthy young men. Am J Clin Nutr. 1989;50:1382-1388.
Sirtori CR, Tremoli E, Gatti E, Montanari G, Sirtori M, Colli S, Gianfranceschi G, Maderna P, Dentone CZ, Testolin G, Galli C. Controlled evaluation of fat intake in the Mediterranean diet: comparative activities of olive oil and corn oil on plasma lipids and platelets in high-risk patients. Am J Clin Nutr. 1986;44:635-642.
Becker N, Illingworth DR, Alaupovic P, Connor WE, Sundberg EE. Effects of saturated, monounsaturated, and 118 -6 polyunsaturated fatty acids on plasma lipids, lipoproteins and apoproteins in humans. Am J Clin Nutr. 1983;37:355-360.
Masana L, Camprubi M, Sarda P, Sola R, Joven J, Turner PR. The Mediterranean-type diet: is there a need for further modification? Am J Clin Nutr. 1991;53:886-889.
Mata P, Garrido JA, Ordovas JM, Blazquez E, Alvarez-Sala LA, Rubio MJ, Alonso R, de Oya M. Effect of dietary monounsaturated fatty acids on plasma lipoproteins and apolipoproteins in women. Am J Clin Nutr. 1992;56:77-83.
Mata P, Alvarez-Sala LA, Rubio MJ, Nuño J, de Oya M. Effects of long-term monounsaturated- vs polyunsaturated-enriched diets on lipoproteins in healthy men and women. Am J Clin Nutr. 1992;55:846-850.
Sirtori CR, Gatti E, Tremoli E, Galli C, Gianfranceschi G, Franceschini G, Colli S, Maderna P, Marangoni F, Perego P, Stragliotto E. Olive oil, corn oil, and n-3 fatty acids differently affect lipid, lipoproteins, platelets and superoxide formation in type II hypercholesterolemia. Am J Clin Nutr. 1992;56:113-122.
Gardner CD, Kraemer HC. Monounsaturated versus polyunsaturated dietary fat and serum lipids: a meta-analysis. Palo Alto, Calif: Department of Statistics, Stanford University; July 1994. Technical report No. 297.
Hedges LV, Olkin I. Statistical Methods for Meta-Analysis. Orlando, Fla: Academic Press; 1985:78,81,86,110,113,123.
Mensink RP, Katan MB. Effect of dietary fatty acids on serum lipids and lipoproteins: a meta-analysis of 27 trials. Arterioscler Thromb. 1992;12:911-919.
Kwon J-S, Snook JT, Wardlaw GM, Hwang DH. Effects of diets high in saturated fatty acids, canola oil, or safflower oil on platelet function, thromboxane B2 formation, and fatty acid composition of platelet phospholipids. Am J Clin Nutr. 1991;54:351-358.
Vatten LJ, Bjerve KS, Andersen A, Jellum E. Polyunsaturated fatty acids in serum phospholipids and risk of breast cancer: a case-control study from the Janus serum bank in Norway. Eur J Cancer. 1993;29A:532-538.