In previous posts, I have shown that the "lower the better" hypothesis is not credible. Alleged cause-and-effect relationships between blood cholesterol and cardiovascular disease do not hold up to scrutiny (e.g., see here, here, or here).
More importantly, however, the "lower the better" hypothesis also fails when we look at the objective endpoint of all-cause mortality.
In 1992, for example, researchers of the Conference on Low Blood Cholesterol analyzed the relationship between blood cholesterol and all-cause mortality from nineteen cohort studies consisting of subjects aged 35 to 69 with nine to thirty years of follow-up.
After adjustments for age, diastolic blood pressure, cigarette smoking, body mass index, and alcohol intake (where available), the results showed a U-shaped relationship for men and no relationship for women. In stratified analyses (for smoking and alcohol status), the results remained similar.
Thus, for both men and women, there was no indication that low blood cholesterol was better. If anything, low levels (<160 mg/dL) might be worse in some cases.
Reverse Causation
Over the decades, "reverse causation" has been the standard reply to the lack of all-cause mortality benefits at low cholesterol levels.
According to lipid-lowering supporters, it is not low cholesterol levels that cause disease. Instead, it is preclinical disease that causes low cholesterol levels.
To address this potential bias, researchers of the Conference on Low Blood Cholesterol excluded early deaths (i.e., deaths occurring within five years), but the results "did not change."
They also noted that for some studies, the higher mortality at low cholesterol levels persisted even after the first ten or more years of deaths were removed from the analysis.
Thus, reverse-causation bias did not seem likely, and a true cause-and-effect relationship between low blood cholesterol and some diseases remained possible:
Effect-cause [reverse causation] is not likely to explain many of the low-cholesterol associations, because the conference report analyses excluded deaths occurring during the first 5 years after the cholesterol measurement [Hulley, Walsh, and Newman (1992)].
But these cohorts were just the beginning.
In 1994, Krumholz and colleagues published a community-based cohort study of older participants and found that high cholesterol levels were not associated with mortality. For women, those with the highest levels (≥240 mg/dL) had the best survival whereas those with the lowest levels had the worst survival (<200 mg/dL).
When the researchers removed early deaths to assess the possibility of reverse causation they found no substantial difference in the results. Also, there were "no significant differences" between the cholesterol groups at baseline for overall health status, functional status, cognitive function, psychosocial support, or socioeconomic status.
Three years later, Weverling-Rijnsburger and colleagues confirmed these results in another cohort. They found that "each 1 mmol/L increase in total cholesterol corresponded to a 15% decrease in mortality." To address reverse causation, they excluded early deaths and looked at a subgroup with similar cholesterol concentrations over time, but the results did not change.
As they stated:
Taken together, the results probably cannot be explained by disease, known or unknown, that causes both low total cholesterol concentrations and increased all-cause mortality.
In 2001, the Honolulu Heart Program with over 20 years of follow-up found that high cholesterol levels were not associated with higher mortality. In fact, those who maintained a low cholesterol level over the 20-year period had the highest risk of dying.
Again, lipid-lowering supporters were not happy, claiming "reverse causation" bias. But as the authors noted in their reply, reverse causation was not a good explanation:
The central, surprising, finding from our study was that low serum cholesterol, persisting for 20 years, increased subsequent all-cause mortality. We have no logical explanation for this finding, nor, apparently, do any of our correspondents. . . . Nonetheless, to suggest that chronic medical and psychiatric disorders cause poor appetite and low bodyweight, leading to low serum cholesterol for 20 years, followed by increased mortality defies logic.
Similarly, Langsted and colleagues (2011) conducted a long-term study with 31 years of follow‐up. The results showed that elevated triglycerides were associated with increased mortality, but not elevated cholesterol levels.
According to the investigators, reverse causation did not seem likely:
It is unclear why only elevated nonfasting triglycerides and not elevated nonfasting cholesterol are associated with total mortality. The lack of increased mortality with increased cholesterol levels cannot be explained by confounding as we adjusted for the same cardiovascular risk factors in the analyses for both lipids, and elevated triglyceride levels are clearly associated with increased mortality . . . Also, this discrepancy is unlikely to be explained by reverse causation . . . This is because we also tried to exclude individuals who died within 1 or several years after the lipid measurements, leaving the risks of total mortality as a function of increasing cholesterol levels largely unchanged . . . and because we followed the cohort for up to 31 years during which time 58% of all women and 69% of all men died.
As for more recent cohorts, Sung and colleagues (2019) looked at LDL and mortality in a Korean cohort. After exclusion of early deaths and adjustments for confounders, low LDL levels (<70 mg/dL) were associated with a higher risk of death. The researchers also noted that the young age of the cohort "helped decrease the influence of potential reverse causality."
One year later, Johannesen and colleagues (2020) assessed the relationship between LDL levels and all-cause mortality in a Danish cohort. After adjustments for many factors (including diseases such as cancer), they found a U-shaped relationship between LDL and mortality, with the lowest risk at an LDL level of 140 mg/dL. The results also remained similar even though the researchers "tried to account for reverse causation in numerous analyses."
In 2021, Liu and colleagues used data from NHANES to assess the association between LDL and all-cause mortality in a US cohort. Low LDL levels (<70 mg/dL) were associated with increased all-cause mortality, a result that persisted "after adjusting for potential confounders" and the exclusion of those who died within three years "in order to prevent reverse causality."
On stratification, Liu et al. found that low LDL levels were still associated with higher mortality among participants who did not have cancer at baseline — an illness that allegedly contributes to reverse causation. The point estimates for the lowest cholesterol group, in fact, were similar between those with and without known cancer.
More Cohorts
The above is only a sample of the database.
Hamazaki et al. (2015) and Ravnskov et al. (2020) already reviewed a large number of cohorts showing no beneficial associations or even harmful associations between lower LDL levels and mortality.
(Image: Taken from Ravnskov et al.)
And there are many others published since.
In the Shanghai Aging Study (2022), for instance, there was an inverse association between LDL and mortality (i.e., the higher the better). The results remained after the exclusion of deaths within the first two years and the exclusion of functionally impaired participants. The difference in survival between high and low LDL levels also expanded over time, which "strengthens the evidence against reverse causality."
Also in 2022, Yi and colleagues published a large cohort of almost 15 million individuals without known cardiovascular disease, cancer, or lipid-lowering medication use. Based on 536975 deaths, the results showed that lower LDL levels were associated with higher mortality:
(Image: Low LDL levels are associated with increased mortality. Note also that if we use a higher reference level where comorbidities are less common (e.g., an LDL of 160 mg/dL), lower LDL levels still do not associate with benefit. Excluding the first three years of deaths had little effect on the associations)
This finding complemented a cohort of 12.8 million adults by Yi, Yi, and Ohrr (2019). In that cohort, there were U-shaped associations between blood cholesterol and mortality "regardless of sex or age." Even for extremes of the curve, the rate of death was greater for those with low cholesterol levels (<150 mg/dL) than those with very high cholesterol levels (>280 mg/dL).
Incidentally, these results are quite similar to that of the Global Cardiovascular Risk Consortium (2023), where researchers analyzed data from 112 cohort studies conducted in 34 countries and 8 geographic regions:
(Image: Low blood cholesterol is associated with higher mortality. Compared to the reference level of 116 mg/dL, even extremely high cholesterol values do not associate with higher risk)
Also worth mentioning is a cohort by Zeng and colleagues published in 2022. They excluded individuals with cancer, acute myocardial infarction, and heart failure at baseline. Based on the dietary data and body mass index, there was no evidence of malnutrition in the lower non-HDL cholesterol groups, and the baseline risk factor data were generally favorable. Moreover, the average age of the population was only 44.29 years.
Despite all these factors that should minimize "reverse causation" bias, those with lower non-HDL cholesterol levels had a higher risk of death (compared to the reference level of 130–159 mg/dL).
Many of these cohorts also show that a low LDL level is not associated with lower cardiovascular mortality — the very condition a low LDL level is supposed to prevent.
For instance, Rong and colleagues (2022) published a cohort with up to 27 years of follow-up. They found that low LDL levels (<70 mg/dL) were associated with increased risks of all-cause mortality, cardiovascular mortality, and stroke mortality. Excluding participants taking lipid-lowering drugs or those with comorbidities "yielded similar results."
(Image: Taken from Liu and colleagues. The lack of beneficial relationships between all-cause mortality and low cholesterol is not surprising. Many studies have failed to find beneficial associations between low cholesterol and cardiovascular mortality/events or atherosclerosis)
Possible Mechanisms
Researchers have proposed several mechanisms by which a low LDL level may be disadvantageous.
Wu and colleagues (2023) noted that lipoproteins such as LDL "may have protective functions." LDL may protect against the "lethal effects of inflammation," and its reduction is associated with an increased risk of diabetes and hemorrhagic stroke.
Vainshelboim and Myers (2023) observed that cholesterol has "many essential physiological roles both at the cellular and organism levels," and low levels in some cases "may lead to dysregulation of homeostasis and promote illness."
Lv, Yin, and Shi (2022) likewise stated:
Several potential explanations are as follows: (1) Opposite to “bad cholesterol”, LDL-C may play an important role in many physiological and biochemical processes. There are several studies have confirmed that LDL-C can improve the immune response to bacteria and viruses and tissue repair capabilities. Meanwhile, LDL-C is actively involved in the formation and transport of sterol hormones (such as androgens and estrogen) and the transport of fat-soluble vitamins. (2) Low LDL-C may increase the non-cardio-cerebrovascular mortality of the oldest old. For example, low LDL-C is closely related to cancer, injury, suicide, depression, liver disease, and chronic kidney disease [note: we could add many other conditions, including atrial fibrillation and heart failure].
Finally, Hsu and colleagues (2021) stated that a drop in lipid levels may indicate "reduced resistance to oxidative stress," and that decreased lipid levels may interfere with the binding of bacterial toxins, which "increases the risks of infection."
Thus, the conclusions of Wen and colleagues (2022) may be correct. A low cholesterol level "is not a cause for celebration."
Other Thoughts
Although lipid-lowering supporters refer to "reverse causation," the alleged bias is more appropriately called "confounding bias" or "confounding by pre-existing disease," where the disease causes both low cholesterol and mortality.
Second, there is really no good way of handling these biases in observational epidemiology. Strategies such as excluding early deaths or restricting the data to specific groups can introduce biases rather than solve them.
Third, although some researchers have questioned adjusting for albumin, the adjustment does not seem to matter in most cases. The lack of benefit for low cholesterol persisted in many studies that adjusted for albumin, matched for albumin, stratified by albumin, or excluded low albumin values.
Fourth, I suspect that U-shaped relationships would flatten considerably if cohorts exclude those with genetic lipid disorders and properly address physical activity, cardiorespiratory fitness, body composition, and dietary oxidative products (all objectively measured).
Fifth, contrary to what some researchers claim, genetic studies do not provide better evidence (e.g., so-called Mendelian randomization). Such designs rely on very strong assumptions and do not necessarily produce more valid results than traditional epidemiology.
According to VanderWeele (2021):
A view is sometimes expressed that so-called natural experiments, as ostensibly used in mendelian randomization analyses, provide stronger forms of evidence for causal inference than regressions using observational data. This is an oversimplification. Mendelian randomization analyses make numerous assumptions about the absence of direct effects of the variant on the outcome not through the exposure, and the absence of various selection biases. These assumptions are different from those in regressions with observational data but just as susceptible to violations.
In any case, some genetic data show an association between low cholesterol levels and increased liver cancer, pancreatic cancer, endometrial cancer, Parkinson’s disease, depression, chronic kidney disease, hemorrhagic stroke, diabetes, and dementia.
Sixth, the most comprehensive meta-regression on the topic (a statistical tool lipid-lowering supporters love) did not show any relationships between decreases in LDL (absolute or percentage reductions) and all-cause mortality (expressed as relative or absolute risk):
(Image: Taken from Ennezat et al.)
Conclusions
In general, low cholesterol levels are not associated with an all-cause mortality benefit. Instead, low levels are often associated with a greater risk of death.
While these harmful associations at low cholesterol levels do not prove causation, the conclusion is still clear: There is no suggestion whatsoever from these cohorts that low cholesterol levels are beneficial, and the hypothesis that low cholesterol levels are harmful under some circumstances has not been excluded.
Besides all-cause mortality, it would be interesting to explore the evidence for specific diseases in more detail. Some researchers, for example, argue that the relationship between low cholesterol and liver cancer may be causal given the plausible mechanisms and robust relationships (i.e., strong and consistent relationships with a dose response).
Thus, claims that low blood cholesterol will save us from diseases and prolong survival generally do not hold up in observational epidemiology. Whether this is true in randomized trials will be the topic of the next post.