In the final installment of this review, we will look at clinical events.
O'Keefe and colleagues state (1):
Trials from both the setting of primary prevention and secondary prevention show that the risk of suffering a CHD event during the course of the study was closely correlated with on-treatment LDL.
To support this, they provided the following two graphs of statin trials:
These analyses are often called "meta-regressions," and they appear to support a strong relationship between LDL cholesterol and coronary heart disease (CHD) events.
But there are many issues here.
Problem #1: The Endpoint
Combining studies with different endpoint definitions might "compromise the interpretation of results across trials" (2).
Thus, if we want to make a comparison across trials — as O'Keefe et al. did — then the definition of CHD events should be similar in each study.
As DuBroff and de Lorgeril noted (3):
Some have argued that there is a linear relation between low-density lipoprotein (LDL) levels and CHD events [referring to O'Keefe et al.]. This analysis may be inaccurate because it combines different types of CHD events from diverse studies into one endpoint even though each study defines CHD events differently.
Worse still, it is difficult to tell what O'Keefe and colleagues mean by "CHD events" when their estimates do not completely match those of the original trials.
The CARE trial, for example, reported a CHD event rate of 13.2% in the placebo group and 10.2% in the statin group (3). But O'Keefe and colleagues give an unexplained event rate of over 15% for the placebo group and around 10% for the statin group.
Problem #2: Flawed Models
According to O'Keefe's graphs, there are linear relationships between LDL and CHD events. Apparently, CHD risk decreases as LDL goes lower, without any diminishing benefits.
But this has no basis in reality.
A linear relationship is implausible because it allows for CHD event rates of 0% (or lower) and 100% (or greater). Furthermore, biology is "neither linear nor simple" (4).
Compressing the complexity of biological processes into the simplicity of a statistical model is often unrealistic, and we should not seek to infer detailed understanding of biology from statistical models (5).
In fact, a linear relationship is contradicted by the very same statin trials that O'Keefe and colleagues cite.
For instance, the researchers of WOSCOPS explicitly stated (6):
The relationship between reduction of LDL and the reduction of risk was not linear.
And, in another paper, they found no relationship between LDL and CHD events.
To quote (7):
Although LDL cholesterol level is currently used to select patients for statin therapy and to monitor treatment response, it was notable that neither baseline nor change in LDL cholesterol predicted future coronary events.
For other trials, the AFCAPS investigators noted "little relation" between on-treatment LDL and CHD risk, and that LDL "is not predictive unless considered in conjunction with HDL-C" (8).
As stated:
Interestingly, LDL-C and TC did not achieve significance as predictors of risk either at baseline or at year 1.
The CARE trial also reported a non-linear relationship, with no association at LDL levels below approximately 125 mg/dL (9):
A central finding of this study is that the relationship between LDL during treatment and coronary events is not a linear one but rather appears not to decline further below a concentration of ≈125 mg/dL.
Likewise, researchers of LIPID did not find a robust association between on-study LDL and CHD (10). A later report also suggested little to no association between LDL and cardiovascular mortality (11):
(Image: Based on data from LIPID. Note that LDL is represented by the red curve, not the blue)
And many other trial analyses are consistent with these. PROSPER, for example, showed that baseline, on-treatment, and change in LDL were unrelated to cardiovascular risk (12).
Problem #3: Ecological Bias
Similar to what we saw in Part 2, O'Keefe's graphs are study-level correlations and thus ecological in nature. As such, aggregation bias and confounding bias are major issues.
Note well, although the graphs contain data from "randomized" trials, this is irrelevant. When the data are analyzed across trials — as O'Keefe et al. did — the benefits of randomization are lost.
In the words of Hayward, Hofer, and Vijan (13):
An ecological study involves comparing differences between groups of individuals (aggregate data), thus ignoring differences between individuals within those groups (for example, noting that clinical trials that achieved greater average change in LDL cholesterol level have also tended to achieve greater relative cardiovascular benefits). The simplicity of presenting aggregate data can have great intuitive appeal, but it is actually an extremely weak source of epidemiologic evidence, having been characterized as “dangerous at best and disastrous at worst.”
A recent paper (2022) from Japanese researchers also acknowledged issues with ecological comparisons, even citing O'Keefe et al. in the process.
To quote (14):
“The lower the better” hypothesis is basically derived from the meta-regression analyses [ecological comparisons] of randomized controlled trials of “statins versus placebo”, “more versus less statins”, and “non-statin lipid-lowering therapy versus placebo”, suggesting that lower on-treatment LDL-C was associated with lower cardiovascular event rates [O'Keefe et al. is cited here]. However, the results from the meta-regression analyses might not be robust, because there are big differences in the risk profiles of the patients enrolled in the individual trials, and in the intensity of lipid-lowering therapy between the trial arms and across the trials.
As Hayward and coworkers pointed out, when each trial population is compared to its own control — thus avoiding across-trial biases — the relationship between LDL and cardiovascular events "changes substantially" and becomes "less strong and less uniform."
We can see this in the following graph:
(Image: Each trial compared to its own control with correct estimates for LIPID, CARE, and HPS. Note how shallow the lines are compared to O'Keefe's graphs)
But even this has limitations because it does not consider potential non-lipid and non-LDL lipid mechanisms of statin therapy and other in-trial biases.
Problem #4: Meta-Regressions Are Highly Subjective
Like a meta-analysis, meta-regressions attempting to correlate variables depend on many decisions, including study selection, choice of endpoint, and how the data are analyzed.
Population-level statistical evidence comes from randomised trials and meta-analyses that are shot through with biases, rendering such evidence very often unreliable (15).
O'Keefe and colleagues, however, do not attempt to justify their decisions or discuss the methodological issues, sampling biases, and major conflicts of interest involved in statin trials.
With different trial selection and decisions, we could equally conclude that there are no consistent study-level associations between LDL and clinical events.
In fact, two recent meta-regressions, published in 2022 by Byrne et al. and Ennezat et al. — both independent groups — could not find a robust association between LDL and clinical events.
Here is the conclusion from Byrne et al. (16):
Because the meta-regression analysis yielded inconsistent results, we concluded that our meta-regression was inconclusive in proving or disproving an association between the magnitude of LDL-C reduction and the size of treatment effect.
And here are conclusions from Ennezat et al. (17):
Our findings indicate that all-cause and CV mortality reduction was not related to LDL-C relative or absolute reduction (RR, ARD) or to baseline and achieved LDL-C levels in meta regression models. . . . clinical benefits did not consistently differ between trials that achieved LDL-C levels below 70 and even 55 mg/dL and those with achieved LDL-C greater than 70 or 116 mg/dL.
I will also note that O'Keefe and colleagues contradicted themselves when talking about the alleged benefits of statin therapy.
In one sentence, they claimed that trials using statin therapy have demonstrated "remarkable reductions" in CHD events. Yet, in the very next sentence, they stated:
Approximately three out of four CHD events occurred despite the statin therapy.
I think most readers would agree that if most events are still occurring despite statin therapy, then it is not accurate to describe the alleged risk reductions as "remarkable."
Problem #5: "Observational" Studies
According to O'Keefe et al.:
Observational studies show a continuous positive relationship between CHD risk and LDL levels that extends well below the average range seen in modern populations without any definite threshold where lower LDL concentrations are not associated with lower risk.
But the paper cited in support of this statement did not evaluate LDL cholesterol and only included male subjects up to 39 years of age (i.e., a non-representative group consisting of many with genetic disorders) (18).
Furthermore, O'Keefe et al. failed to realize that data from the largest cohort included in the paper (MRFIT screenees) contradicted their graphs. We can see this in the following graph from MRFIT 1989:
(Image: Taken from reference 19. Note the U-shaped relation for overall mortality and the flat lower portions of the CHD curve. The absolute difference in CHD risk comparing cholesterol levels <160 mg/dL to the 240-259 mg/dL group is a mere 0.53%)
Nevertheless, since O'Keefe et al. accept epidemiological cohorts, they should have no problem with a study conducted by Lee et al. in 2020, which contained over 5.6 million subjects aged 20–39 (20).
Lee et al. found that the 84 to 101 mg/dL LDL range was associated with the lowest risk — a finding at odds with the alleged 50 to 70 "optimal" range. Moreover, after adjustments for other risk factors, there was no evidence of an association between LDL and clinical events:
After adjusting for age, sex, smoking, heavy drinking, obesity, physical inactivity, family history of premature CVD, and history of hypertension and diabetes mellitus . . . triglycerides [not LDL] remained the sole independent determinant of clinical events.
But there's more. . .
A cohort of almost 15 million subjects found that LDL levels of 55–69 mg/dL were not associated with less CHD compared to LDL levels of 70–114 mg/dL (21). Instead, a low LDL level was associated with increased cardiovascular mortality, stroke, heart failure, sudden cardiac death, and possibly aneurysm rupture risk.
(Image: Data for all-cause mortality taken from reference 21. Based on 536,975 deaths. Note the higher death rate at lower LDL levels and the lower death rate over a wide range of LDL levels above 109 mg/dL)
Similarly, Rong et al. showed that LDL levels less than 70 mg/dL were associated with a higher risk of all-cause, cardiovascular, and stroke mortality compared to LDL levels of 100–129 mg/dL (22). Also, those with LDL levels less than 70 mg/dL did not have a lower risk of dying from CHD.
Yuan et al. also found non-linear relationships between LDL and clinical events (23): A J-shaped association for heart attacks, no association for ischaemic stroke, an inverse association for haemorrhagic stroke (lower risk with higher LDL), and a reversed J-shaped association for all-cause mortality (higher risk with lower LDL).
Therefore, many epidemiological cohorts do not suggest that LDL levels of 50-70 mg/dL might be better for CHD. If anything, when looking at overall health (all diseases), the data raise the hypothesis that "optimal" LDL levels are higher than 70 mg/dL.
Problem #6: PROVE-IT Did Not Verify Anything
PROVE-IT was a trial conducted in patients with acute coronary syndrome comparing the effects of standard statin therapy (40 mg of pravastatin daily) to intensive statin therapy (80 mg of atorvastatin daily).
Based on this trial, O'Keefe et al. stated:
The recently published PRavastatin Or atorVastatin Evaluation and Infection Therapy (PROVE-IT) trial is the strongest verification of the lower is better hypothesis.
They further provided the following graph from PROVE-IT showing an apparent 16% reduction in cardiovascular events for lowering LDL to 62 mg/dL with intensive statin therapy (atorvastatin):
But we have some issues here.
First, the alleged benefit seen in the graph was driven by soft endpoints susceptible to bias (mainly revascularization). The number of CHD deaths was small and other endpoints such as heart attacks and strokes appeared minimally affected.
For example, the rate of heart attacks was 6.6% in the high-dose atorvastatin group and 7.4% in the standard-dose pravastatin group. This translated into a minuscule 0.8% absolute risk reduction that was not statistically significant.
Second, PROVE-IT did not and could not test the "lower is better" hypothesis, because the researchers employed two different statins at different intensities. Thus, we could equally conclude that statin intensity matters, not LDL lowering (24):
The guidelines also refer to the PROVE-IT, TNT, A to Z, IDEAL and SEARCH studies as evidence that lowering LDL to 2 mmol/L [77 mg/dL] or less results in the lowest risk for secondary prevention patients. Unfortunately, none of these studies prospectively tested different LDL targets. Rather, they tested different doses and different statins, demonstrating that the intensity of the statin matters, not the degree of LDL lowering.
Indeed, the hypothesis that statin intensity matters is consistent with a recent analysis by Toyota et al. (14).
In short, Toyota et al. found that, at the same statin dose adjusting for a multitude of variables, on-treatment LDL levels less than 70 mg/dL were not associated with lower cardiovascular risk compared to higher on-treatment LDL levels.
To quote:
The major findings of the present study were as follows: (1) very low on-treatment LDL-C level (<70 mg/dL) was not associated with lower cardiovascular event risk compared with moderately low on-treatment LDL-C level (70–100 mg/dL) in patients receiving the same doses of statins; (2) high LDL-C (≥100 mg/dL) was not associated with higher cardiovascular event risk compared with a moderately low LDL-C level (70–100 mg/dL) in the pitavastatin 1 mg/day stratum . . .
These results are illustrated in the following graph:
(Image: Taken from reference 14. Note the lack of association in the 1 mg/day stratum. The higher risk for the grey bar in the 4 mg/day stratum was attributed to unreported non-adherence rather than LDL)
Thus, the PROVE-IT trial, cited by O'Keefe and colleagues as the "strongest verification of the lower is better hypothesis," actually verified nothing about LDL at all.
Conclusion
In the end, O'Keefe et al. could not provide any credible mechanistic, pathological, epidemiological, or clinical trial evidence to support their claims.
Contrary to the title of their paper, the optimal LDL level is not 50 to 70 mg/dL and lower LDL levels are neither better nor physiologically normal.
References
1) O'Keefe, J. H., Cordain, L., Harris, W. H., Moe, R. M., & Vogel, R. (2004). Optimal low-density lipoprotein is 50 to 70 mg/dl: lower is better and physiologically normal. Journal of the American College of Cardiology, 43(11), 2142-2146.
2) Jakobsen, J. C., Wetterslev, J., & Gluud, C. (2021). Considerations on the strengths and limitations of using disease-related mortality as an outcome in clinical research. BMJ Evidence-Based Medicine, 26(3), 127-130.
3) DuBroff, R., & de Lorgeril, M. (2015). Cholesterol confusion and statin controversy. World journal of cardiology, 7(7), 404.
3) Sacks, F. M., Pfeffer, M. A., Moye, L. A., Rouleau, J. L., Rutherford, J. D., Cole, T. G., ... & Braunwald, E. (1996). The effect of pravastatin on coronary events after myocardial infarction in patients with average cholesterol levels. New England Journal of Medicine, 335(14), 1001-1009.
4) Sturmberg, J. P. (2020). From probability to believability. Journal of Evaluation in Clinical Practice, 26(4), 1081-1086.
5) Tennant, P. W. G., Arnold, K. F., Berrie, L., Ellison, G. T. H., & Gilthorpe, M. S. (2017, July). Advanced modelling strategies: challenges and pitfalls in robust causal inference with observational data. In Advanced Modelling Strategies: Challenges and pitfalls in robust causal inference with observational data. Leeds Institute for Data Analytics.
6) Shepherd, J., & Park, J. S. (1998). Prevention of heart disease: is LDL reduction the outcome of choice? No, there is more. Value in Health, 1(2), 120-124.
7) Ford, I., Shah, A. S., Zhang, R., McAllister, D. A., Strachan, F. E., Caslake, M., ... & Mills, N. L. (2016). High-sensitivity cardiac troponin, statin therapy, and risk of coronary heart disease. Journal of the American College of Cardiology, 68(25), 2719-2728.
8) Gotto Jr, A. M., Whitney, E., Stein, E. A., Shapiro, D. R., Clearfield, M., Weis, S., ... & De Cani, J. S. (2000). Relation between baseline and on-treatment lipid parameters and first acute major coronary events in the Air Force/Texas Coronary Atherosclerosis Prevention Study (AFCAPS/TexCAPS). Circulation, 101(5),
9) Sacks, F. M., Moye, L. A., Davis, B. R., Cole, T. G., Rouleau, J. L., Nash, D. T., ... & Braunwald, E. (1998). Relationship between plasma LDL concentrations during treatment with pravastatin and recurrent coronary events in the Cholesterol and Recurrent Events trial. Circulation, 97(15), 1446-1452.477-484.
10) Simes, R. J., Marschner, I. C., Hunt, D., Colquhoun, D., Sullivan, D., Stewart, R. A., ... & Tonkin, A. (2002). Relationship between lipid levels and clinical outcomes in the Long-term Intervention with Pravastatin in Ischemic Disease (LIPID) Trial: to what extent is the reduction in coronary events with pravastatin explained by on-study lipid levels?. Circulation, 105(10), 1162-1169.
11) Hilvo, M., Meikle, P. J., Pedersen, E. R., Tell, G. S., Dhar, I., Brenner, H., ... & Laaksonen, R. (2020). Development and validation of a ceramide-and phospholipid-based cardiovascular risk estimation score for coronary artery disease patients. European heart journal, 41(3), 371-380.
12) Packard, C. J., Ford, I., Robertson, M., Shepherd, J., Blauw, G. J., Murphy, M. B., ... & Westendorp, R. G. (2005). Plasma lipoproteins and apolipoproteins as predictors of cardiovascular risk and treatment benefit in the PROspective Study of Pravastatin in the Elderly at Risk (PROSPER). Circulation, 112(20), 3058-3065.
13) Hayward, R. A., Hofer, T. P., & Vijan, S. (2006). Narrative review: lack of evidence for recommended low-density lipoprotein treatment targets: a solvable problem. Annals of internal medicine, 145(7), 520-530.
14) Toyota, T., Morimoto, T., Iimuro, S., Fujita, R., Iwata, H., Miyauchi, K., ... & Kimura, T. (2022). Low-Density Lipoprotein Cholesterol Levels on Statins and Cardiovascular Event Risk in Stable Coronary Artery Disease―An Observation From the REAL-CAD Study―. Circulation Journal, CJ-22.
15) Stegenga, J. (2022). Evidence of effectiveness. Studies in History and Philosophy of Science, 91, 288-295.
16) Byrne, P., Demasi, M., Jones, M., Smith, S. M., O’Brien, K. K., & DuBroff, R. (2022). Evaluating the association between low-density lipoprotein cholesterol reduction and relative and absolute effects of statin treatment: a systematic review and meta-analysis. JAMA Internal Medicine.
17) Ennezat, P. V., Guerbaai, R. A., Maréchaux, S., Le Jemtel, T. H., & François, P. (2022). Extent of LDL-cholesterol Reduction and All-cause and Cardiovascular Mortality Benefit: a Systematic Review and Meta-analysis. Journal of Cardiovascular Pharmacology.
18) Stamler, J., Daviglus, M. L., Garside, D. B., Dyer, A. R., Greenland, P., & Neaton, J. D. (2000). Relationship of baseline serum cholesterol levels in 3 large cohorts of younger men to long-term coronary, cardiovascular, and all-cause mortality and to longevity. Jama, 284(3), 311-318.
19) Colpo, A. (2006). The Great Cholesterol Con: Why Everything You've Been Told about Cholesterol, Diet and Heart Disease is Wrong!. Lulu. com.
20) Lee, H., Park, J. B., Hwang, I. C., Yoon, Y. E., Park, H. E., Choi, S. Y., ... & Kim, H. K. (2020). Association of four lipid components with mortality, myocardial infarction, and stroke in statin-naïve young adults: a nationwide cohort study. European Journal of Preventive Cardiology, 27(8), 870-881.
21) Yi, S. W., An, S. J., Park, H. B., Yi, J. J., & Ohrr, H. (2022). Association between low-density lipoprotein cholesterol and cardiovascular mortality in statin non-users: a prospective cohort study in 14.9 million Korean adults. International Journal of Epidemiology.
22) Rong, S., Li, B., Chen, L., Sun, Y., Du, Y., Liu, B., ... & Bao, W. (2022). Association of Low‐Density Lipoprotein Cholesterol Levels with More than 20‐Year Risk of Cardiovascular and All‐Cause Mortality in the General Population. Journal of the American Heart Association, 11(15), e023690.
23) Yuan, S., Huang, X., Ma, W., Yang, R., Xu, F., Han, D., ... & Lyu, J. (2022). Associations of HDL-C/LDL-C with myocardial infarction, all-cause mortality, haemorrhagic stroke and ischaemic stroke: a longitudinal study based on 384 093 participants from the UK Biobank. Stroke and Vascular Neurology, svn-2022.
24) Lindblad, A. J., Kolber, M. R., Garrison, S., Cotton, C., & Allan, G. M. (2015). Prevention and Management of Cardiovascular Disease Risk in Primary Care: Evidence Review of 12 Key Clinical Questions. Alberta College of Family Physicians.