"A meta-analysis of prospective epidemiologic studies showed that there is no significant evidence for concluding that dietary saturated fat is associated with an increased risk of CHD or CVD."
The above quote was the conclusion from Siri-Tarino et al. (2010) — a meta-analysis that evaluated the relation between saturated fat and cardiovascular events (1).
As you might expect, some researchers were not exactly happy with this conclusion, to say the least (2-4).
One common objection was that some analyses adjusted for blood cholesterol levels. Allegedly, adjusting for serum cholesterol is an "over-adjustment," which would attenuate the relation between saturated fat and cardiovascular disease.
In essence, critics claim that if you statistically adjust for serum cholesterol, you remove the mechanism by which saturated fat causes cardiovascular disease: an increase in serum cholesterol levels.
For example, here is a quote from Scarborough et al. criticizing the Siri-Tarino analysis (3):
It is well established that saturated fat intake is associated with increased concentration of serum cholesterol, and that serum cholesterol concentrations are associated with CHD and CVD. Therefore, serum cholesterol concentrations lie on the causal chain between saturated fat intake and CHD and CVD and to adjust for serum cholesterol concentrations in a meta-analysis will obscure the effect of saturated fat intake on these health outcomes.
This argument is hardly convincing.
First, serum cholesterol and LDL levels are very poor predictors of cardiovascular disease (see here). The main premise, in fact, is likely false (i.e., the premise that cholesterol/LDL causes cardiovascular disease).
Second, the argument is logically invalid. We cannot assume transitivity or a causal chain. As one epidemiologist notes, foods containing saturated fat could be neutral or beneficial "even if LDL IS causal." Be careful with reductionist thinking! (see here).
Third, statistical adjustments cannot adequately compensate for the severe biases in nutritional epidemiology and involve "far more subjective judgment than many users realize" (5). Thus, establishing a causal chain from epidemiological data alone is not possible.
Observational studies suffer from a wide variety of biases that may not be adequately addressed even using state-of-the-art statistical modeling techniques [6].
Actually, causal chains are also difficult to establish in well-controlled randomized trials (due to confounding between the proposed mechanism and outcome). It often happens that a treatment appears beneficial or harmful, yet the mechanisms remain unclear.
But for the sake of argument, let us continue...
"Addressing the Controversy"
Although rarely cited, Siri-Tarino and colleagues did respond to the cholesterol over-adjustment criticism. When they excluded studies that adjusted for cholesterol, the results did not differ significantly from the summary result for all studies.
As stated (7):
We agree with Scarborough et al that it is appropriate to consider the possibility that inclusion of serum cholesterol concentrations in multiple regression models may attenuate the relation of saturated fat to cardiovascular disease (CVD) in observational cohort studies. However, using data from the subset of studies in our meta-analysis in which the models did not include blood cholesterol concentration [9 coronary heart disease (CHD) studies and 6 stroke studies; n = 291,126], the results did not differ significantly from those that we reported for all 21 studies (n = 347,747).
[Image: Taken from Siri-Tarino, reference 8. Note that the original effect for CVD is 1.00 (0.89, 1.11), which is almost identical to the subgroup estimate of 1.02 (0.86, 1.19)]
Other researchers also noted the lack of meaningful differences. Here is Sanders (2013) commenting on the issue (9):
This review [Siri-Tarino 2010], although meticulously conducted, has been controversial and criticised for adjusting for serum cholesterol in the analysis. However, even without adjustment for serum cholesterol there are no significant trends to indicate any increase in risk and the size effects are small.
Of course, the Siri-Tarino meta-analysis is not new, being published in 2010. But subsequent meta-analyses suggested the same.
Siri-Tarino Confirmed
In 2014, Chowdhury et al. reported no significant differences between studies adjusting for conventional risk factors (including lipids) and those adjusting for conventional non-lipid risk factors (P = 0.47) (10):
There was no material difference in the combined RRs according to sex, year of baseline survey, dietary assessment tool, duration of follow-up, outcome definition, or degrees of statistical adjustment.
De Souza et al. in 2015 also found no significant differences between studies adjusting for "TC or LDL-C" and those that did not. These lack of differences held for the endpoints of total CHD (P = 0.361), CHD mortality (P = 0.972), and ischemic stroke (P = 0.860).
As they stated (11,12):
This demonstrates that the overall results of our synthesis are robust and are not substantially affected by different approaches to covariate adjustment.
For total mortality, Kim et al. (2021) reported no significant differences based on lipid adjustments (13):
We performed a subgroup analysis by adjustment for serum lipids (e.g., hypercholesterolemia), but significant differences were not observed.
The largest published analysis, however, was done by Zhu, Bo, and Liu in 2019 (14). Using their data, I found no significant differences between studies adjusting for serum cholesterol (or baseline hypercholesterolemia) and those that did not.
(Image: Subgroup analysis based on data from Zhu, Bo, and Liu)
Notably, the point estimates in all these subgroup analyses are between 0.80 and 1.20. Even by NutriGrade's low standards, these estimates would be regarded as non-meaningful (see here).
Why not unadjusted models?
To evaluate the possible effect of lipid over-adjustment, we cannot use unadjusted or minimally adjusted models (e.g., age adjusted only). Why? Because this would lead to the opposite problem: under-adjustment.
As De Souza points out, saturated fat consumption generally associates with unhealthy habits. Therefore, unadjusted models could lead to overestimated associations and a biased result against saturated fat (12):
. . . unadjusted models overestimate associations, because estimates reflect other determinants of the health outcomes. Notably, in our study those in the highest categories of saturated fat intakes tended to exercise less, smoke more, and eat less fibre.
Besides properly adjusted models, the over-adjustment issue should be addressed within individual cohorts, not between cohorts with subgroupings or meta-regressions.
Individual Cohorts
Esrey and colleagues stated decades ago that the diet-lipid-heart hypothesis may be “overly simplistic," because diet-CHD relations did not change when they adjusted for serum lipids (15). And numerous similar findings have been published.
Gillman et al. found an inverse relation between saturated fat and stroke with "comparable results" after further adjustments for total cholesterol and HDL cholesterol (16).
Pietinen et al. reported a reduced risk of coronary death associated with a higher intake of saturated fat (in one model), but they also stated that adjustments for lipids made no difference (17):
In the multivariate models, there were significant inverse associations between coronary death and the intake of saturated fatty acids and a positive association for polyunsaturated fatty acids and linoleic acid (p for trend both < 0.05). Further adjustment for serum total and high density lipoprotein cholesterol did not change the associations.
Xu et al. found a significant association between saturated fat and CHD mortality in the younger but not older age group. No difference was seen when they excluded lipids from the model (18):
Omitting HDL cholesterol and LDL cholesterol from the model did not change the results.
Without adjustments for LDL cholesterol, Virtanen et al. observed that saturated fat associates with increased LDL cholesterol but not increased CHD events/mortality (19).
To quote:
Although SFA intake was associated with higher baseline LDL cholesterol concentrations . . . there was no independent association with SFA intake and risk of CHD. This is in line with the results from the reviews of prospective cohort studies.
Similarly, in four cohorts, saturated fat (chocolate in one) was associated with increased LDL. In all four, however, there were no harmful associations between saturated fat and heart events (20-23).
One of these four cohorts also found that adjustment for the total cholesterol to HDL ratio had virtually no effect on the point estimate [0.83 (0.74, 0.93) without adjustment vs. 0.84 (0.75, 0.95) with adjustment]:
Our results did not materially change after including the baseline total cholesterol:HDL cholesterol ratio or systolic blood pressure in the models.
I also previously reported on two studies where no harmful association was found between milk intake and cardiovascular events/disease despite increased LDL levels (see here).
More recently, Wang and colleagues did not find a relationship between saturated fat and CHD mortality (24). The point estimate in a model adjusting for intermediate factors (including total cholesterol) was almost identical to a model not adjusting for these factors [1.02 (0.94–1.10) vs. 1.00 (0.93–1.08)].
On the other hand, Blekkenhorst reported an association between saturated fat and cardiovascular mortality. But even here, "neither total nor LDL cholesterol concentrations were related to ASVD mortality" (25).
Lastly, in a "substitution" analysis, Kvist et al. evaluated dairy products and noted that hypertension, hypercholesterolemia, and diabetes may be a part of the causal link as "intermediate factors." But again, they found that adjustment for potential intermediate factors showed "similar pattern of associations" (26).
Other Nutrients
The above findings also apply to other nutrients.
To give some examples: Maki reported that adjustment for non-HDL-C had "virtually no impact" on point estimates for egg intake (27), and Gao et al. found "no evidence" to support the idea that risk factors mediated associations for diet (28).
Likewise, Choi et al. stated that adjustment for LDL-C and non-HDL-C "did not substantially alter" their finding (29).
Haugsgjerd et al. also found that adjustments for intermediate factors "did not materially influence the association" for carbohydrates (30), and Erkkilä et al. stated for PUFA intake that "adjustment for LDL and HDL cholesterol and triglyceride concentrations had little effect on the associations" (31).
Conclusions
In terms of associations in nutrition, adjustments for serum cholesterol/LDL generally do not matter. The over-adjustment claim is flawed in theory with no credible empirical evidence to support it.
If anything, the majority of analyses are consistent with the words of Dehghan and colleagues (32):
Predicting the net clinical effect based on considering only the effects of nutrient intake on LDL cholesterol is not reliable in projecting the effects of diet on cardiovascular disease events or on total mortality.
After all, cholesterol and LDL in themselves do not drive cardiovascular disease. According to conventional wisdom, it is really the modified (e.g., oxidized) lipoproteins that are the true culprits (33,34):
LDL itself has no significant atherogenic properties [35].
Thus, while meta-analyses in nutritional epidemiology certainly have major flaws, inclusion of models adjusting for serum cholesterol is not one of them.
References
1) Siri-Tarino, P. W., Sun, Q., Hu, F. B., & Krauss, R. M. (2010). Meta-analysis of prospective cohort studies evaluating the association of saturated fat with cardiovascular disease. The American journal of clinical nutrition, 91(3), 535-546.
2) Katan, M. B., Brouwer, I. A., Clarke, R., Geleijnse, J. M., & Mensink, R. P. (2010). Saturated fat and heart disease. The American journal of clinical nutrition, 92(2), 459-460.
3) Scarborough, P., Rayner, M., van Dis, I., & Norum, K. (2010). Meta-analysis of effect of saturated fat intake on cardiovascular disease: overadjustment obscures true associations. The American journal of clinical nutrition, 92(2), 458-459.
4) Stamler, J. (2010). Diet-heart: a problematic revisit. The American journal of clinical nutrition, 91(3), 497-499.
5) Christenfeld, N. J., Sloan, R. P., Carroll, D., & Greenland, S. (2004). Risk factors, confounding, and the illusion of statistical control. Psychosomatic medicine, 66(6), 868-875.
6) Berger, V. W., Bour, L. J., Carter, K., Chipman, J. J., Everett, C. C., Heussen, N., ... & Uschner, D. (2021). A roadmap to using randomization in clinical trials. BMC medical research methodology, 21(1), 1-24.
7) Siri-Tarino, P. W., Sun, Q., Hu, F. B., & Krauss, R. M. (2010). Reply to P Scarborough et al. The American Journal of Clinical Nutrition, 92(2), 459.
8) https://www.yumpu.com/en/document/read/22217668/astrup-a-et-al-am-j-clin-american-heart-association
9) Sanders, T. A. (2013). Reappraisal of SFA and cardiovascular risk. Proceedings of the Nutrition Society, 72(4), 390-398.
10) Chowdhury, R., Warnakula, S., Kunutsor, S., Crowe, F., Ward, H. A., Johnson, L., ... & Di Angelantonio, E. (2014). Association of dietary, circulating, and supplement fatty acids with coronary risk: a systematic review and meta-analysis. Annals of internal medicine, 160(6), 398-406.
11) De Souza, R. J., Mente, A., Maroleanu, A., Cozma, A. I., Ha, V., Kishibe, T., ... & Anand, S. S. (2015). Intake of saturated and trans unsaturated fatty acids and risk of all cause mortality, cardiovascular disease, and type 2 diabetes: systematic review and meta-analysis of observational studies. Bmj, 351.
12) https://www.bmj.com/content/351/bmj.h3978/rr-18
13) Kim, Y., Je, Y., & Giovannucci, E. L. (2021). Association between dietary fat intake and mortality from all-causes, cardiovascular disease, and cancer: A systematic review and meta-analysis of prospective cohort studies. Clinical Nutrition, 40(3), 1060-1070.
14) Zhu, Y., Bo, Y., & Liu, Y. (2019). Dietary total fat, fatty acids intake, and risk of cardiovascular disease: a dose-response meta-analysis of cohort studies. Lipids in health and disease, 18(1), 1-14.
15) Esrey KL, Joseph L, Grover SA. Relationship between dietary intake and coronary heart disease mortality: lipid research clinics prevalence follow-up study. J Clin Epidemiol 1996;49:211–6. - Notably, these observations are nothing new.
16) Gillman, M. W., Cupples, L. A., Millen, B. E., Ellison, R. C., & Wolf, P. A. (1997). Inverse association of dietary fat with development of ischemic stroke in men. Jama, 278(24), 2145-2150.
17) Pietinen P, Ascherio A, Korhonen P, et al. Intake of fatty acids and risk of coronary heart disease in a cohort of Finnish men. The Alpha-Tocopherol, Beta-Carotene Cancer Prevention Study. Am J Epidemiol 1997;145: 876–87.
18) Xu J, Eilat-Adar S, Loria C, et al. Dietary fat intake and risk of coronary heart disease: the Strong Heart Study. Am J Clin Nutr 2006;84:894–902.
19) Virtanen, J. K., Mursu, J., Tuomainen, T. P., & Voutilainen, S. (2014). Dietary fatty acids and risk of coronary heart disease in men: the Kuopio Ischemic Heart Disease Risk Factor Study. Arteriosclerosis, thrombosis, and vascular biology, 34(12), 2679-2687.
20) Puaschitz, N. G., Strand, E., Norekvål, T. M., Dierkes, J., Dahl, L., Svingen, G. F. T., ... & Nygård, O. (2015). Dietary intake of saturated fat is not associated with risk of coronary events or mortality in patients with established coronary artery disease. The Journal of nutrition, 145(2), 299-305.
21) Praagman, J., Beulens, J. W., Alssema, M., Zock, P. L., Wanders, A. J., Sluijs, I., & Van Der Schouw, Y. T. (2016). The association between dietary saturated fatty acids and ischemic heart disease depends on the type and source of fatty acid in the European Prospective Investigation into Cancer and Nutrition–Netherlands cohort, 2. The American journal of clinical nutrition, 103(2), 356-365.
22) Ho, Y. L., Nguyen, X. M. T., Yan, J. Q., Vassy, J. L., Gagnon, D. R., Gaziano, J. M., ... & VA Million Veteran Program LD YL H XM TNDRGJQYJLVPWFWKCJMG. (2021). Chocolate consumption and risk of coronary artery disease: the Million Veteran Program. The American journal of clinical nutrition, 113(5), 1137-1144.
23) Sonestedt, E., Wirfält, E., Wallström, P., Gullberg, B., Orho-Melander, M., & Hedblad, B. (2011). Dairy products and its association with incidence of cardiovascular disease: the Malmö diet and cancer cohort. European journal of epidemiology, 26(8), 609-618.
24) Wang, Y., Fang, Y., Witting, P. K., Charchar, F. J., Sobey, C. G., Drummond, G. R., & Golledge, J. (2023). Dietary fatty acids and mortality risk from heart disease in US adults: an analysis based on NHANES. Scientific Reports, 13(1), 1614.
25) Blekkenhorst, L. C., Prince, R. L., Hodgson, J. M., Lim, W. H., Zhu, K., Devine, A., ... & Lewis, J. R. (2015). Dietary saturated fat intake and atherosclerotic vascular disease mortality in elderly women: a prospective cohort study. The American journal of clinical nutrition, 101(6), 1263-1268.
26) Kvist, K., Laursen, A. S. D., Overvad, K., & Jakobsen, M. U. (2020). Substitution of milk with whole-fat yogurt products or cheese is associated with a lower risk of myocardial infarction: the Danish Diet, Cancer and Health cohort. The Journal of nutrition, 150(5), 1252-1258.
27) https://mbclinicalacademy.com/tag/non-hdl-cholesterol/
28) Gao, M., Jebb, S. A., Aveyard, P., Ambrosini, G. L., Perez-Cornago, A., Carter, J., ... & Piernas, C. (2021). Associations between dietary patterns and the incidence of total and fatal cardiovascular disease and all-cause mortality in 116,806 individuals from the UK Biobank: A prospective cohort study. BMC medicine, 19(1), 1-12.
29) Choi, Y., Gallaher, D. D., Svendsen, K., Meyer, K. A., Steffen, L. M., Schreiner, P. J., ... & Jacobs, D. R. (2022). Simple Nutrient-Based Rules vs. a Nutritionally Rich Plant-Centered Diet in Prediction of Future Coronary Heart Disease and Stroke: Prospective Observational Study in the US. Nutrients, 14(3), 469.
30) Haugsgjerd, T. R., Egeland, G. M., Nygård, O. K., Igland, J., Sulo, G., Lysne, V., ... & Tell, G. S. (2020). Intake of carbohydrates and SFA and risk of CHD in middle-age adults: the Hordaland Health Study (HUSK). Public Health Nutrition, 1-15.
31) Erkkilä, A., de Mello, V. D., Risérus, U., & Laaksonen, D. E. (2008). Dietary fatty acids and cardiovascular disease: an epidemiological approach. Progress in lipid research, 47(3), 172-187.
32) Dehghan, M., Mente, A., Zhang, X., Swaminathan, S., Li, W., Mohan, V., ... & Amma, L. I. (2017). Associations of fats and carbohydrate intake with cardiovascular disease and mortality in 18 countries from five continents (PURE): a prospective cohort study. The Lancet, 390(10107), 2050-2062. - "Predicting the net clinical effect based on considering only the effects of nutrient intake on LDL cholesterol is not reliable in projecting the effects of diet on cardiovascular disease events or on total mortality.”
33) Lands, B. (2022). Lipid nutrition:“In silico” studies and undeveloped experiments. Progress in Lipid Research, 85, 101142.
34) Akhmedov, A., Sawamura, T., Chen, C. H., Kraler, S., Vdovenko, D., & Lüscher, T. F. (2021). Lectin-like oxidized low-density lipoprotein receptor-1 (LOX-1): a crucial driver of atherosclerotic cardiovascular disease. European heart journal, 42(18), 1797-1807.
35) Nakajima, K., & Tanaka, A. (2018). Postprandial remnant lipoproteins as targets for the prevention of atherosclerosis. Current Opinion in Endocrinology & Diabetes and Obesity, 25(2), 108-117.