Thursday, 31 January 2013

Diet-Heart: Saturated Fat and Blood Cholesterol

Several hundred metabolic ward experiments have established beyond plausible doubt that isocaloric substitution of saturated fat for carbohydrate, polyunsaturated fat and monounsaturated fat raises serum (blood) LDL and total cholesterol (Fig. 1).1 Many of these experiments were carried out in locked facilities where the participants mode of life, energy balance and body weight were maintained, with the focus purely on isocaloric exchange of fat for starch or one kind of fat for another.2 This is one of the main pillars that the diet-heart hypothesis was built on.

Figure 1. Serum total cholesterol and dietary saturated fat in 395 solid food dietary and 32 liquid formula dietary experiments, by different experimental design

In Part I of this review, I addressed a misleading meta-analysis of prospective cohort studies examining the association between saturated fat intake and the risk of cardiovascular disease authored by Siri-Tarino et al. and Stephan Guyenet’s interpretation of this meta-analysis. In this part of the review I will address the evidence regarding the influence that saturated fat has on serum LDL and total cholesterol in humans and some of the primary arguments against the evidence that saturated fat raises serum cholesterol that Guyenet presented in a review of the literature that I previously commented on.3

One of the primary arguments against the evidence that saturated fat raises serum cholesterol that Guyenet presented is in regards to the duration of most of the metabolic ward experiments that are cited in the literature. For example, Guyenet stated: 
When references are provided, they nearly always point to the same type of study: short-term controlled diet trials, in which volunteers are fed different fats for 2-13 weeks and their blood cholesterol measured 
Guyenet’s statement is misleading as he appears to be emphasizing that there is a lack of long term controlled trials with findings that conform to the expectations from shorter term metabolic ward experiments. There have actually been a number of randomized controlled dietary trials with a duration of between one year and up to and beyond eight years that have demonstrated a significant decrease in serum LDL and total cholesterol, independent of weight loss when saturated fat was replaced with either carbohydrates or unsaturated fats. Examples include the National Diet-Heart Study and the Veterans Admin study, among many other trials included in a recent Cochrane review.4 It would be incorrect to suggest that detailed reviews neglect these long-term trials, although researchers may be more inclined to cite shorter-term metabolic ward experiments because of their more rigorous design. 

In the great majority of the metabolic ward experiments practically all of the serum cholesterol response to a change in dietary fat was completed within two to three weeks, and any further fluctuations in serum cholesterol can mainly be explained by intra- and interindividual variation, seasonal trends or laboratory error.2 Furthermore, in long term controlled trials the observed changes in serum LDL and total cholesterol and saturated fat intake actually conform to the expectations from the metabolic wards experiments that Guyenet attempts to downplay.5 6 This explains the universal agreement that intake of saturated fat (lauric, myristic and palmitic acids in specific) raises serum LDL and total cholesterol. 

Another one of the primary arguments against the evidence that saturated fat raises serum cholesterol that Guyenet presented is in regards to the null findings from certain observational studies. For example, Guyenet stated: 
…why do researchers almost never cite observational studies to support the idea that dietary saturated fat increases blood cholesterol? … One reason may be that in most instances, when researchers have looked for a relationship between habitual saturated fat intake and blood cholesterol, it has been very small or nonexistent.
There are actually several different types of observational studies that have studied the influence that saturated fat has on serum cholesterol. For example, secular trends have demonstrated a clear association between changes in saturated fat intake and changes in serum cholesterol within nations all-throughout the world, independent of other factors [reviewed previously]

The observational studies that Guyenet described in his review of the literature used cross-sectional methodology in which saturated fat intake and serum cholesterol are typically only measured at one specific point in time, and not repeatedly over a prolonged period over time. Guyenet’s suggestion that these studies that he cited examined ‘habitual saturated fat intake’ is unfounded and provides very scant evidence to suggest as he did that the findings from short term metabolic ward studies cannot be extrapolated to the long term. As Guyenet should already be aware, the shortcomings of the methodology used in the cross-sectional studies he cites was thoroughly addressed several decades ago and should no longer be considered controversial.7 8 9 10 11 A number of these shortcomings will be addressed here as Guyenet has conveniently failed to address them.


The Problem of Interindividual Variability


One of the greatest shortcoming of cross-sectional studies which has contributed towards unnecessary confusion has been caused by the lack of consideration  of the interindividual (between-individual) variability of serum lipids in response to diet. Steinburg has described how this shortcoming can obscure the findings in cross-sectional studies:12 
One explanation is that the variability from person to person in terms of response to dietary factors is so great that it dilutes out the correlation, as in one of Ahrens’ metabolic ward protocols, almost all would show the same qualitative responses but to different degrees... [Jacobs et al.] pointed out the large variability encountered from individual to individual in responses to changes in dietary fat. Jacobs et al. used a mathematical model to show that a zero correlation is what you would actually expect. They also applied their model to a set of experimental data and showed that, again, a zero correlation might be expected. In short, a zero correlation in a population study does not necessarily negate the possible role of dietary fat in helping to determine plasma cholesterol levels in individuals.
Table 1 provides a good example of the significant between-individual variability of serum lipids in response to diet in reference to the National Diet-Heart Study.12 

Table 1. Percentage of subjects with indicated percentage drop in serum cholesterol in the National Diet-Heart Study

This shows that even if the investigators know the exact composition of a person’s diet they will not be able to predict their level of serum cholesterol, but will be able to predict whether that person’s serum cholesterol will go up or down if that person modifies their dietary intake. As the association between diet and coronary heart disease in part relies upon serum lipids, this suggests that the between-individual variability of serum lipids in response to diet would likely have obscured the findings from the prospective cohort studies addressed in Part I of this review.8 


The Problem of Dietary Modification


Another important shortcoming of cross-sectional studies that can obscure the association between diet and serum cholesterol in observational studies can be caused by the participants decreasing their intake of saturated fat in response to unfavorable serum lipid concentrations. This has actually been shown to bias the cross-sectional association between saturated fat and serum cholesterol towards a negative association. This problem was addressed by Shekelle et al. in a paper from the Chicago Western Electric Company study:9
In the Chicago Western Electric Company study, diet was assessed at the initial examination, in 1957-1958, of 1900 middle-aged men and again at their second examination about one year later. At the first examination, lipid composition of the diet, as summarized by a score based on the formula of Keys, Anderson and Grande, was positively associated with level of serum cholesterol. Between the first and second examinations, however, hypercholesterolemic men were more likely than others to have reduced intake of dietary saturated fatty acids and cholesterol. As a result, at the second examination the cross-sectional linear association between the diet score and serum cholesterol concentration was significantly positive for men with initial levels of serum cholesterol less than 250 mg/dl, significantly negative for men with initial levels of 250 mg/dl or higher and not significantly different from zero for all men together.
Guyenet failed to mention this limitation despite referring to the Chicago Western Electric Company study in his review. Guyenet claimed that although this study found a positive association between saturated fat and serum cholesterol, there was no association with heart attack deaths. Guyenet failed to mention that the lack of a significant association with heart attack deaths may be explained at least in part by the fact that the researchers actually adjusted for serum cholesterol, and because the participants with elevated serum cholesterol were likely to have reduced their saturated fat intake. This study nevertheless found that saturated fat was associated with an 11% non-significant increased risk, and dietary cholesterol and the Keys score with a statistically significant increased risk of  fatal coronary heart disease.13 Furthermore, this study also found that dietary cholesterol was associated with a statistically significant increased risk of all-cause mortality, a finding consistent with several other studies including the Nurses’ Health Study.14 15 16

The purpose of the paper from the Health Professionals Follow-up Study that Guyenet cited was to address the influence of saturated fat intake on the risk of coronary heart disease, not on serum cholesterol. The researchers were aware that the health professionals, of which most had previously measured their serum cholesterol may have made dietary modifications if they were previously aware of having elevated cholesterol, and therefore excluded any participants with a baseline history of elevated cholesterol.17 The fact that electric company employees modified their diet in response to unfavorable serum lipids in the 1950’s strongly suggests that health professionals in the 1980’s would also have made dietary modifications, thus explaining the unexpected distribution of saturated fat intake of participants with baseline history of elevated serum cholesterol in this study.


The Problem of Intraindividual Variability


Another important shortcoming of cross-sectional studies that can obscure the association between diet and serum cholesterol is caused by intraindividual (within-individual) variability of serum lipids and dietary intake. Hegsted and Nicolosi showed that the spontaneous within-individual variability of serum cholesterol even when consuming a constant diet is relatively large, typically varying by 5 to 10%.11 It has been estimated that three samples of blood are required in order estimate the true serum cholesterol of an individual, assuming that standardized methods are being used. Even just sitting from a standing position has been shown to decrease serum cholesterol concentrations by 6%.18

Balogh et al. concluded that in order to estimate within 20% of the actual dietary intake of 90% of a studied population, there is a requirement of at least 22 days of 24-hour dietary recalls for saturated fat.19 Inaccurately measuring within-individual variability in dietary intake has been shown to lead to a miss-classification of subjects into ranges of usual dietary intakes, biasing correlation coefficients towards null.20 Liu et al. showed that just a single measure of a dietary variable and a single measure of serum cholesterol as was done in the Tecumseh Study reduces the correlation coefficient by 50%, and is therefore almost meaningless.7 It should be noted that almost half of the cross-sectional studies that Guyenet cited used only a single 24-hour recall to assess dietary intake, including the Tecumseh Study as well as the Evans Country study, the Bogalusa Heart Study, and the Japan-Honolulu-San Francisco study.20 21

Jacobs et al. demonstrated the problem of intraindividual variability on a graph (Fig. 2), showing how the predicted association between diet and serum cholesterol becomes obscured by the introduction of typical variations in serum cholesterol and diet.8

Figure 2. A: The open circles represent the relationship that exists in five typical individuals between diet score and predicted serum cholesterol. Vertical arrows represent random variations in serum cholesterol and horizontal arrows random variations in diet scores. B: The open circles demonstrate how the perfect correlation in A is completely obscured by the introduction of typical sources of intraindividual variability.

Excluding the Health Professional Follow-up Study, in the two cross-sectional studies that Guyenet cited with the largest number of participants, the Israel Ischemic Heart Disease Project (n=8829) and the Japan-Honolulu-San Francisco study (n=9844), when the researchers considered the participants as groups rather than as individuals in order to minimize intraindividual variability from obscuring the results, saturated fat was highly significantly and positively associated with serum cholesterol.22 23 Similar results were also found in the Seven Countries Study (n=12,700).18 The findings from these studies actually conform closely to the expectations from the metabolic ward experiments that Guyenet attempts to downplay.

It should be noted that each of these studies alone studied a significantly larger number of participants than all of the remaining cross-sectional studies cited by Guyenet combined. Of these remaining studies the Chicago Western Electric study which made up approximately half of the sample size (n=1900) also found a statistically significant association. Furthermore, Guyenet failed to cite at least half a dozen other cross-sectional studies that studied between 650 and 23,000 participants that found a significant association between saturated fat and serum cholesterol.18 24 25 Guyenet has since updated his post to note that he was made aware of a number of studies that he missed but nevertheless claimed that they are largely consistent with his conclusions.

In the review, Guyenet made several misleading statements in regards to the data from the Framingham study: 
One of the longest-running, most comprehensive and most highly cited observational studies, the Framingham study was organized by Harvard investigators and continues to this day. When investigators analyzed the relationship between saturated fat intake, serum cholesterol and heart attack risk, they were so disappointed that they never formally published the results. We know from multiple sources that they found no significant relationship between saturated fat intake and blood cholesterol or heart attack risk
Guyenet cited a 1976 paper from the Tecumseh study as a reference for his findings from the Framingham study, but failed to mention that the Framingham study has produced many formal publications since this paper was published. In 1992 Posner et al. published a paper from the Framingham study addressing the association between saturated fat and heart attack risk, however this study used a single 24-hour food recall and therefore is of limited value. In this study the younger but not older cohort saturated fat was associated with a marginally significant increased risk of coronary heart disease.26

The findings described in the paper from the Tecumseh study of the failure to find a significant association between diet and serum cholesterol in the Framingham study may have addressed an early publication of Framingham which excluded participants with very high or low serum cholesterol concentrations which could have been influenced by diet. A reanalysis of all of the men’s data in 1992 resulted in a larger sample size and range of serum cholesterol concentrations found a significant association between dietary cholesterol and total fat intake and serum cholesterol.18 However, the more recent analysis of women from the Framingham study produced contrasting findings for dietary cholesterol, which resulted in much confusion that has been provoked by the likes of Atkins and Sugar Busters. Dr. William Castelli, a former director of the Framingham Heart Study responded to this misuse of the Framingham study data:27
The data are diet history data. Very weak science!!!... Better science, where I lock you up in a metabolic ward has taught us that lowering the saturated fat, the cholesterol in the diet lowers cholesterol.
The Framingham study is well respected in part for enhancing the medical communities knowledge of risk factors for coronary heart disease including the Framingham Risk Score, not for the use of high quality dietary assessment methods. As suggested by Castelli, the dietary data from the Framingham study should not be misused as Guyenet has done in order to attempt to downplay the findings from metabolic ward experiments.

In response to my comments regarding the concerns about the quality of dietary assessment methods used in the cross-sectional studies that Guyenet cited in his review, he replied stating: 
You like good measurement methods, so let's focus on the two studies that did 7-day weighed food records, the Bankers study and the Caerphilly study. The first found no association between animal fat intake and serum cholesterol across a several-fold variation in intake. The second found a significant but extremely weak association between SFA intake and serum cholesterol. "The percentage of variance in the plasma lipid concentrations which could be explained solely by the dietary variables was very small, ranging from 1 to 7 per cent". That sounds quite consistent with the conclusions of my post.
As already explained, a near null finding is what the expected result would be in these cross-sectional studies regardless of the quality of the dietary and lipid assessment methods used due to the problem of between-individual variability. However, a significant limitation of the British bank men study (n=99), the smaller of the two studies Guyenet refers to was that there was a prolonged time period between the dietary survey and the blood sample in a portion of the studied men. In Table 2 the first column shows the data on all the men in the survey, and column two shows the data only on those men whose blood samples were taken either during or within one day of completing the dietary survey.7 

Table 2. Serum cholesterol and dietary factors in all men compared to men whose cholesterol was measured close to the dietary survey in the British bank men study


It is clear that the correlation between animal fat and serum cholesterol was a lot stronger in those men whose blood samples were taken during or within one day of completing the dietary survey, suggesting that the results from the ‘All Men’ group was obscured by dietary changes made in the prolonged time period between the dietary survey and the blood sample in some of the men. Furthermore, it was estimated that the combined within-person variability for dietary factors and serum cholesterol reduced the correlation by 35-40%.7 

In regards to the Caerphilly study (n=653), the larger of these two studies, Guyenet appears to be extrapolating the conclusions of the authors that the between-individual variability in diet explained only a small portion of the between-individual variability in the serum cholesterol levels of the studied population to claim that the association between saturated fat and serum cholesterol although statistically significant was ‘extremely weak’. The authors of this study argued in a later paper that studies on homogeneous populations consuming a diet uniformly high in saturated animal fat such as their own, the lack of a contribution of between-individual variability in saturated animal fat intake to coronary heart disease may merely reflect 'the absence of clear differences in mean intakes of saturated fatty acids'.28 As the between-individual variability in animal fat intake was small it is to be expected that animal fat would only explain a small portion of the between-individual variability in serum lipids, and that other factors including genetics may explain a larger portion of the variance. In the Seven Countries Study where saturated fat ranged from 3% to about 22% of calories, saturated fat intake explained approximately 89% of the variance in serum cholesterol between the 16 cohorts.18 Although animal fat intake may not explain much of the variation in serum cholesterol between individuals in the Caerphilly study, it still may have largely explained the uniformly high serum cholesterol of the studied population as a whole.

When critically examined, the findings from these studies are 'quite consistent' with the diet-heart hypothesis, not with Guyenet’s unfounded conclusions. Furthermore, these were not the only studies that I informed Guyenet of that used a 7-day food record. The EPIC-Norfolk study included a 7-day food record from 6,416 participants which according to the researchers found a similar association between diet and serum lipids as the food frequency questionnaire from 22,914 participants which was used to gain sufficient power for the analysis by separate genotypes. This study found that saturated fat was associated with a significantly higher serum LDL and total cholesterol for all four individual genotypes studied.24 

If Guyenet is to suggest that observational studies that use multiday food records are the highest quality form of observational evidence to test the diet-heart hypothesis, then high quality observational studies support both the hypothesis that saturated fat raises serum cholesterol and increases the risk of coronary heart disease. As Stamler described in regards to the Siri-Tarino et al. meta-analysis of prospective cohort studies and the risk of saturated addressed in Part I of this review:29
Five used dietary history or multiday food record; the RR was >1.00 in all 5 studies, even though 3 were adjusted for serum or dietary lipids.
Plant Positive has produced an informative video addressing measurement error in the newly released YouTube series Nutrition Past and Future, which provides further details addressing how intra- and interindividual variability can obscure the findings between diet and risk factors and heart disease in reference to the hand-picked observational studies that Gary Taubes included in his book, Good Calories Bad Calories. Coincidentally the observational studies addressing saturated fat and serum cholesterol included in Good Calories Bad Calories are mostly the same studies that Guyenet included in his review.

Cholesterol Confusion 7 The Measurement Problem


Longitudinal Observational Studies


Several prospective cohort studies that measured diet and serum cholesterol multiple times over several years have also found that changes in saturated fat intake was associated with changes in serum cholesterol.30 31 These studies are more informative than cross-sectional studies which typically only measure diet and serum cholesterol at a single point of time making them more prone to being obscured by intra- and interindividual variability.

The Blue Mountain Eyes Study was a prospective cohort study that measured diet and serum cholesterol three times throughout the 10 year follow-up, and found a statistically significant association between changes in saturated fat intake and serum cholesterol. As expected due to the problem of intra- and interindividual variability, the cross-sectional analysis of only the baseline data showed a weak non-significant association between saturated fat intake and serum cholesterol.30 In response to my comments regarding the positive association between saturated fat and serum cholesterol in this study, Guyenet responded stating:
Contrary to what you said, the Blue Mountain Eye study did not find any association whatsoever between SFA intake and TC/LDL. What they found was an association between TC/LDL and a CHANGE in SFA intake over time. That result is difficult to interpret due to the fact that in the same study, people who consistently ate more SFA didn't have higher TC/LDL than those who consistently ate less. You seem to have a tendency to misrepresent findings to support your position. I want to be clear with you, I will not tolerate that on this blog.
It is clear that Guyenet has again confused the cross-sectional analyses that examined only baseline saturated fat intake with habitual intake. The majority of the participants actually modified their intake of saturated fat to a substantial degree during the two subsequent assessments of dietary intake, demonstrating that the baseline measurement of diet as being a very poor marker of ‘people who consistently ate more SFA’.30 It is unfortunate that Guyenet is unwelcome to genuine criticisms of his arguments against the diet-heart hypothesis and his stance on this matter suggests that any further debate on this topic would likely have proved to be meaningless.


Isocaloric Substitution of Macronutrients


In reference to the meta-analysis of feeding experiments by Katan et al. Guyenet also attempted to downplay the findings by stating: 
As a side note, many of these studies were of poor quality, and were designed in ways that artificially inflated the effects of saturated fat on blood lipids. For example, using a run-in period high in linoleic acid, or comparing a saturated fat-rich diet to a linoleic acid-rich diet, and attributing the differences in blood cholesterol to the saturated fat. Some of them used hydrogenated seed oils as the saturated fat.
The claim that the meta-analysis authored by Katan et al. used ‘hydrogenated seed oils as the saturated fat’ is unfounded. Katan et al. clearly stated that one of the inclusion criteria for the experiments was that 'Food intake had to be thoroughly controlled and described’ and were also clear that diets rich in hydrogenated fat were analyzed separately.32 In a later meta-analysis of controlled experiments, Katan et al. found that trans fats from industrial hydrogenated fats and from ruminant sources found in meat and dairy both raise the ratio of serum LDL to HDL cholesterol.33 Such findings however has not had any  appreciable effect on the dietary recommendations of the cholesterol skeptics.

As emphasized in Part I on this review, a decrease of saturated fat intake suggests an increase in intake of other sources of energy in order to maintain caloric balance, and therefore examining saturated fat intake is meaningless without considering what sources of energy it is replacing. This is clearly why metabolic ward experiments compare saturated fat with other macronutrients. The purpose of the experiments that Guyenet refers to as being of ‘poor quality’ was to examine the effect of the isocaloric substitution of saturated fat for linoleic acid, not to ‘artificially inflate the effects of saturated fat on blood lipids’. Metabolic ward experiments have also demonstrated that isocaloric substitution of saturated fat for monounsaturated fat and carbohydrates also raises serum cholesterol but less appreciatively than when replaced for linoleic acid.1 32

As shown in Figure 1, saturated fat raised serum cholesterol in each type of solid diet experimental design to a similar degree.1 It is not these tightly controlled metabolic ward experiments that were of poor quality,  but it is Guyenet’s interpretation of these experiments that is of poor quality and it is Guyenet who has selected studies that artificially deflate the effects of saturated fat on blood lipids.


Criticisms of the Diet-Heart Hypothesis


Stephan Guyenet specifically chose to limit the focus of his review of the literature to a study design that he should have been well aware has near zero statistical power to detect a relationship between diet and serum cholesterol even if one existed. Guyenet should also have been aware that the majority of his audience are unaware of these design flaws and are therefore are likely to regard his conclusions as fact. These criticisms presented here also apply to other cholesterol skeptics, including Michael Eades and Chris Kresser who cited Guyenet's review as evidence to downplay the diet-heart hypothesis.34 35

The criticisms of the diet-heart hypothesis largely rely on the misinterpretation of studies that have near zero statistical power to detect a statistically significant relationship or suffer from significant methodological issues.


Diet-Heart Posts


Part I - Diet-Heart: A Problematic Revisit
Part III - Diet-Heart: The Role of Vegetarian Diets in the Hypothesis
Part IV - Cracking Down on Eggs and Cholesterol
Part V - Cracking Down on Eggs and Cholesterol: Part II


Please post any comments in the Discussion Thread