The meta-analysis by Siri-Tarino et al (2010) initiated the birth of neo-diet-heart denialism online. The new gospel is that saturated fats do not contribute to heart disease. Inspired by HealthyLongevity (well, I copied his material), I decided to wrap a forum post pointing out some of the key concepts of diet-heart research. Having rudimental knowledge over these themes helps the inquiring reader to avoid being confused by the professional denialists working hard to confuse their global audience (Gary Taubes, Denise Minger, Stephan Guyenet, etc).
1) Interindividual varition; individual differences in the response to the doseThe differences in fat intake are usually minor in homogeneous cultures. However, the cholesterol levels in every population follows a normal distribution: a same diet will lead to wide variety of cholesterol levels between individuals, even though everything else is kept constant. If the differences in the fat intake are small in a given population, it cannot explain the large differences in serum cholesterol levels in a model. Jacobs et al (1) showed mathematically that null association is expected between diet and serum cholesterol levels in cross-sectional population studies (even when there exist cause and effect): despite the null associations in such studies, the intake of saturated fats is one of primary determinants of serum cholesterol levels of an individual (Pedersen et al 2011).
2) Intraindividual variation (regression dilution bias); spontaneous variation within individual”Similar lack of attention to intraindividual variability is frequent in surveys and epidemiological studies and leads to error, controversy, and waste of time and money.” (Keys, 1988)
For a long time, Keys held an intellectual monopoly over this theme. In the past, the researchers didn't properly understood that there's a substantial, spontaneous day-to-day (5 to 10%) variation of cholesterol levels within individual, even though the diet is held constant. Hegstedt & Nicolosi (1987) showed that it takes at least 3 blood cholesterol measurement at the baseline of the survey in order to reliable assess the serum cholesterol level of an individual. In addition, twenty-two randomly collected 24-h dietary recalls are required to estimate the true individual mean intake of saturated fat within ±20 %, while most studies have only one recall or a food frequency measure of saturated fat intake available (2). In older studies, participants were often misclassified in wrong categories in regards to both SFA intake and serum cholesterol levels because of inadequate emphasis on intraindividual variation. In other words, an inadequate number of cholesterol measurements and dietary recalls were performed for the participants at the baseline of the survey. This resulted often in null associations of both diet and serum cholesterol levels and eventually between diet and the risk of CHD.
The Seven Countries StudyWhen Keys (1988) decided to analyze the participants of the large Israel Ischemic Heart Disease study in the form of small groups based on place of birth, he found a strong correlation between the intake of saturated fat and serum cholesterol levels (3). This association did not exist when the participants were analyzed as an individual basis. Some information is always lost with the use of medians, however, by using and comparing medians, the researchers can effectively eliminate the effect of spontaneous variation within individuals (part 2) and the large individual differences in dose-response and the statistical fallacy brought by such large differences between individuals (part 1). Keys understood these two key concepts of diet-heart research already in the late 1950's. He was several decades ahead the curve. The differences in the intake of saturated fat explained 89% of the variation in serum cholesterol levels between the 16 different cohorts used in the 7CS. The differences in the intake of SFAs were large, from 3 percent (of calories) to 22 percent (of calories). The lowest mean serum cholesterol levels at the baseline was measured in one of the Japanese cohort (143mg/dl) and the highest in one of the Finnish cohort (262mg/dl). These ecologic correlations described the real world in much greater accuracy than did the poorly conducted population studies from homogeneous cultures. The ecologic correlations between diet and serum cholesterol levels were similar to the results obtained from tightly controlled metabolic-ward feeding trials. Neo-diet-heart denialists such as Guyenet and Taubes attempt to confuse the reader by showing a cascade of old cross-sectional and prospective cohorts in order to demonstrate that the intake of SFAs do not affect the risk of CHD: The poor audience is left out with an impression that the science is a democracy where every study bears an equal value*
*BonusIt has been pointed out (4) that regression dilution bias dilutes the real association of serum cholesterol concentration and the risk of CHD in several large-scale prospective population studies such as the Framingham and Whitehall. Moreover, the infamous meta-analysis by Siri-Tarino et al used poorly conducted sub-studies that did not pay attention to regression dilution bias and/or were adjusted for serum cholesterol levels (see Stamler 2010; Katan et al 2011; Scarborough et al 2010; Kromhout et al 2011 and Pedersen et al 2011). Modern forward-looking studies, with sophisticated dietary recall techniques, paying careful attention to intraindividual variation, such as the Health Professional’s Follow-up (5) found that saturated fat intake was associated with a statistically significant 72% increased risk of fatal coronary heart disease for high compared to low intake (after maximum adjustment).
Links to some of the references used:
1)
http://aje.oxfordjournals.org/content/110/1/77.short2)
http://arc.crsociety.org/read.php?3,2065283)
http://ajcn.nutrition.org/content/48/5/1161.short4)
http://aje.oxfordjournals.org/content/1 ... 1.full.pdf5)
http://www.ncbi.nlm.nih.gov/pubmed/8688759