by Rob Palmer
Welcome to the third article in this series assessing claims made in “The Carnivore Code,” by Dr. Paul Saladino. If you missed parts one and two, then you can find them here and here, respectively. Of note, the first article contains an explanation of the purpose and method of this series, all of which is longer than I wish to repeat here. The main takeaways from those posts were: 1) Dr. Saladino cited evidence that supported his rebuttal of the claims made by Dan Buettner of The Blue Zones about the diets of the Nicoyans being “blue zone”-esque, though this evidence didn’t support Dr. Saladino’s characterization of the Nicoyans as being renowned for eating meat. 2) Dr. Saladino claimed that insulin resistance is associated with low levels of HDL and elevated levels of triglycerides, though the study he cited showed that severe idiopathic insulin resistance is associated with normal levels of HDL and elevated triglycerides.
In the spirit of honest communication, the discrepancies between the claims made in the book and the evidence cited concern me. That said, it’s only been two articles, so I have more work to do before I can feel confident about the relative veracity of the book.
As I mentioned in the second article, whether or not HDL and triglycerides are indicators of cardiovascular health and/or insulin resistance seems to be a claim that warrants rigorous investigation for a variety of compelling reasons. It seems there are two critical steps to take: 1) determine if there is an association, and 2) determine how strong the association is. (If anyone knows of papers that try to answer those questions, then please share them!) After reflecting upon the last article, I’m still amply motivated to devote at least one more post to this subject, and that’s what I’ll avail to do in this article.
The claim and citation for this article come from the same paragraph of the book as the claim from the last post. There’s a sentence which claims that HDL levels correlate with insulin sensitivity, and this sentence has three citations. I’ve decided to pick the third citation for the ostensible and arbitrary reason that it’s the third article in the series, though I may review one or both of the other studies in future posts. (I’ll leave debates about post hoc confabulation and the existence or nonexistence of free will for another article.)
The referenced study, “Insulin Resistance and Cardiovascular Events With Low HDL Cholesterol,” (2003) is an analysis of data from the Veterans Affairs High-Density Lipoprotein Intervention Trial (VA-HIT). For background, the VA-HIT gave either gemfibrozil or a placebo to men with known coronary heart disease, low HDL cholesterol, low LDL cholesterol, and variable triglycerides levels. (Noe that this study seems poorly-suited for making claims about HDL because people with normal or high levels of HDL were excluded. This shortcoming will become more apparent upon further analysis, though I want to flag it early.) In addition to assessing for diabetes, the researchers assessed insulin resistance by calculating HOMA-IR values (from fasting insulin and fasting glucose levels). Participants were considered insulin resistant if their HOMA-IR value was in the highest tertile, which seems to have been 10.20 and above. (Whether or not this definition of insulin resistance seems valid to you is a separate, though interesting, matter.)
Below is the table that’s most relevant to the claim Dr. Saladino made. It seems important to note that this study had 2,531 participants, which gives it a high statistical power. Consequently, seemingly minor and clinically meaningless differences can be statistically significant. That qualification aside, the group that had neither diabetes nor insulin resistance had a statistically higher mean HDL level than did any of the other groups. The mean difference in all cases, however, was approximately one mg/dl—32 versus 31, give or take one- or two-tenths of a mg/dl. This difference may be statistically significant in this massive sample, though this difference seems so minute as to be clinically insignificant. Additionally, there were no statistically significant differences between the mean HDL of participants who weren’t insulin resistant and had diabetes (column 2) and the mean HDL of participants who either had diabetes and insulin resistance (column 1) or who had insulin resistance without diabetes (column 3). In short, the HDL level seems to correlate weakly with insulin resistance in this sample. To reiterate an earlier point, however, the weakness of this correlation may be in part due to the fact that the study excluded participants with normal levels of HDL.
As for triglycerides, there seems to be a stronger correlation with insulin resistance, though this correlation isn’t quantified either. Participants with diabetes and insulin resistance (column 1) and participants with insulin resistance without diabetes (column 3) had higher levels of triglycerides than did participants without insulin resistance with or without diabetes (column 2 and 4, respectively). The authors summarize the findings, “Both triglycerides and HDL cholesterol were also linearly related to HOMA-IR levels. However, the range of mean values for both of these lipid measurements from the lowest to highest decile level of HOMA values was relatively small: for HDL cholesterol, values ranged from a mean of 32.4 to 30.5 mg/dl; for triglycerides, from a mean of 133 to 182 mg/dl.”
Thus, this study doesn’t show me enough to help me make a confident assessment of this issue, though it doesn’t seem to support the claim that there’s a strong or clinically meaningful correlation between HDL, triglycerides, and insulin resistance. Though this finding technically supports the claim Dr. Saladino made (i.e., that there is a correlation), the book led me to believe that the correlation is strong and meaningful—rather than a single mg/dl of HDL. Thus, the book seems to misrepresent the study. As mentioned earlier, there were two other citations for sentence from which I found this study, so perhaps the other studies are more compelling. (I hope they are, since I expect it would make writing these articles easier.)
Rather than end the article here, I want to discuss other parts of this paper that seemed more interesting and relevant to previously discussed topics. For example, the authors found a strong correlation between waist circumference and fasting insulin levels in patients who weren’t taking insulin (r = 0.968, see the figure below). Additionally, there was a strong correlation between waist circumference and BMI (r = 0.996) as well as a strong correlation between fasting insulin and HOMA-IR (r = 0.943 for all subjects, and r = 0.969 for subjects without a history of diabetes). At the risk of making an inaccurate inference, it seems that a person’s BMI and/or waist circumference is a simple and free way to estimate a person’s fasting insulin and HOMA-IR. (If you’re more familiar with Pearson correlation coefficients, then please let me know if this claim seems valid.) I don’t know how much stronger the correlation is between BMI, waist circumference, fasting insulin, and HOMA-IR than is the correlation between HDL, triglycerides, fasting insulin, and HOMA-IR, though my tentative guess based on this data is that there’s a clinically meaningful difference. Though BMI can be a poor proxy for body composition in people with high levels of lean body mass as well as in “skinny fat” people, it’s likely to be valid in the majority of the population and can be omitted in cases of exception. Consequently, maybe we ought to give more attention to BMI and waist circumference and possibly give less attention to HDL and triglycerides.
One of the other parts of this study I find interesting is its analysis of the five-year cardiovascular event rate stratified by levels of insulin resistance, HDL, and triglycerides. The mean cardiovascular event rate was statistically significantly higher in participants with insulin resistance (31.7%) compared to participants without insulin resistance (20.7%, RR 1.62), and this trend remained significant when excluding participants with diabetes (27.7% vs 19.8%). Thus, insulin resistance seems to have a strong predictive value of cardiovascular events. And what about HDL and triglycerides? Whether or not participants had high or low HDL values had no correlation with cardiovascular events when controlling for insulin resistance. For example, participants with insulin resistance had approximately the same rate of cardiovascular events regardless of their HDL and triglycerides, and the same was true of participants without insulin resistance.
These relationships also seem to suggest that there’s at best a weak correlation between HDL, triglycerides, and insulin resistance (as measured by HOMA-IR). If there were a strong correlation, then we would expect that low and HDL and high triglycerides would be associated with worse outcomes (given the strong association of insulin resistance). To quote the authors, “The occurrence of new CHD events was much less dependent on the levels of either HDL cholesterol or triglycerides at baseline than on the presence or absence of insulin resistance as measured by the calculation of HOMA-IR.” That said, the authors note that some data suggest that the combination of insulin resistance and elevated triglycerides is worse than either insulin resistance or elevated triglycerides on their own.
Lastly, I enjoyed the discussion of the various effects of gemfibrozil when stratifying for insulin resistance—precision medicine, some might say. In short, gemfibrozil improved outcomes more in participants who had insulin resistance compared to participants who didn’t have insulin resistance (28% vs 20% reductions in five-year cardiovascular events, respectively). Additionally, in the absence of insulin resistance, cardiovascular events were more reduced in participants with low levels of HDL than at high levels as well as more reduced at high levels of triglycerides as compared to low triglycerides. As an aside, I wish the authors included the number needed to treat, though don’t we all. Regardless, it’s nice to have this sort of data in order to determine which patients are most likely to benefit from a particular medication.
To sum up the pertinent parts of this piece, this study shows that there may be a weak and clinically insignificant correlation between HDL, triglycerides, and insulin resistance, with the caveat that these participants had low HDL. Thus, it seems to be a misrepresentation to cite this study when asserting that HDL and triglycerides are a proxy for insulin resistance, as Dr. Saladino did. A close reading of the paper suggests that waist circumference may be a much better predictor of fasting insulin and insulin resistance, the latter of which was a strong predictor of cardiovascular events in the study’s population.
Post-script: I looked at the other two studies cited with the study I discussed in this article, and their titles suggest that they may be more relevant to the claim Dr. Saladino made about HDL, triglycerides, and insulin resistance than was this study. Consequently, I may review one or both of them in future articles.
Robins, S. J., Rubins, H. B., Faas, F. H., Schaefer, E. J., Elam, M. B., Anderson, J. W., & Collins, D. (2003). Insulin Resistance and Cardiovascular Events With Low HDL Cholesterol: The Veterans Affairs HDL Intervention Trial (VA-HIT). Diabetes Care, 26(5), 1513–1517. https://doi.org/10.2337/diacare.26.5.1513
Saladino, P. (2020). The Carnivore Code: Unlocking the Secrets to Optimal Health by Returning to Our Ancestral Diet. New York: Houghton Mifflin Harcourt Publishing Company.