Cholesterol, LDL and Longevity: What the Evidence Does and Doesn’t Show
Executive summary
In the single largest cohort study on this question, 12.8 million Korean adults, total cholesterol had a U-shaped relationship with all-cause mortality. The lowest death rates sat at a total cholesterol of roughly 210 to 249 mg/dL (about 5.4 to 6.4 mmol/L), which is higher than the level many patients are told to aim below, and the excess risk at the low end was larger than the excess risk at the high end (Ref 1). Association is not causation, but a simple "lower is always better" message does not survive this data.
In adults over 60, most cohort data show that LDL, the lipoprotein that carries cholesterol, is not positively associated, and is frequently inversely associated, with all-cause mortality (Ref 2). This is association data, it cannot prove direction, and reverse causation is a genuine confounder that I take seriously.
Among 136,905 US patients hospitalised with coronary artery disease, mean admission LDL-C was about 105 mg/dL (about 2.7 mmol/L) and nearly half were below 100 mg/dL (Ref 3). That complicates a simple dose-response reading of LDL, with the important caveat that admission lipids in acutely unwell patients have limitations.
The strongest case for the lipid hypothesis is real and deserves engagement, not dismissal: lifelong low LDL from PCSK9 gene variants tracks with markedly lower coronary disease (Ref 5), and lowering LDL-C with ezetimibe or a PCSK9 inhibitor does reduce cardiovascular events (Refs 6, 7).
The absolute benefit, however, is modest. Adding ezetimibe to a statin cut the seven-year composite event rate by about 2 percentage points (Ref 6). Across statin trials, the median gain in survival within the trial period was about 3 to 4 days (Ref 8), and these particular trials did not demonstrate a clear reduction in all-cause mortality.
Where plaque forms is governed substantially by blood-flow mechanics. Atherosclerosis is focal, clustering at arterial branch points where shear stress is low and disturbed, which places endothelial biology at the centre of the process (Ref 4).
What appears to track better with risk in day-to-day practice is the metabolic environment: the triglyceride-to-HDL-C ratio as a proxy for insulin resistance (Ref 14), fasting insulin, HOMA-IR, and HbA1c read alongside red-cell turnover (Ref 15). A coronary artery calcium score of zero is associated with very low mortality (Ref 16).
None of this is a reason to stop a prescribed statin on your own initiative. It is a reason to have a more detailed, better-informed conversation with your prescriber, and there are many excellent clinicians who will have that conversation with you.
A note on my bias, before anything else
I run a private clinic and I am paid for one-to-one consultations, so I have a commercial interest and you should weigh what follows accordingly. I also read human physiology through a metabolic, low-carbohydrate lens, and I hold postgraduate qualifications in metabolic medicine and anti-ageing medicine, which are separate disciplines rather than one. That lens shapes how I interpret the evidence below. I have tried to state the mainstream case as strongly as its proponents would, then set out where I think it is well supported and where I think it is weak. I may be wrong on parts of this, and I would genuinely welcome evidence that points the other way.
Two points of precision run through the whole piece. First, LDL is a lipoprotein, a particle that ferries cholesterol and fats through the blood. What a standard blood test reports is usually LDL-C, the mass of cholesterol carried inside those particles, and ApoB, which counts the number of atherogenic particles. These are related but not the same thing, and the distinction matters. Second, several of the studies cited here were funded by manufacturers of the drugs being tested, and I have flagged that where it is relevant, in both directions.
What the population-level data actually show
The claim that low cholesterol is uniformly protective does not hold up against the largest datasets we have.
In a prospective cohort of 12,815,006 Korean adults followed for around a decade, with 694,423 deaths recorded, total cholesterol showed a U-shaped association with all-cause mortality in every age and sex band (Ref 1). The total cholesterol range associated with the lowest mortality was 210 to 249 mg/dL for most groups, and the inverse association below 200 mg/dL was actually stronger than the positive association above it. In other words, across this enormous population, moderately high cholesterol was not where mortality was highest, and low cholesterol carried its own excess risk.
Focusing specifically on LDL rather than total cholesterol, a 2016 systematic review of 19 studies covering 68,094 people aged 60 and over found that LDL-C was inversely associated with all-cause mortality in 16 of the cohorts where this was assessed, representing 92 per cent of the participants (Ref 2). Higher LDL in older people was associated with living at least as long, not shorter.
Now the honest critique, because this evidence has real limitations. All of it is observational, so it establishes association, not causation. Reverse causation is the central problem: serious illness, cancer, chronic infection and frailty can lower cholesterol on the way to death, which can manufacture an apparent "low cholesterol, high mortality" signal without cholesterol causing anything. The Korean data are general-population rather than hospitalised patients, which weakens but does not eliminate that concern. The elderly LDL review (Ref 2) is also authored by a group of well-known lipid-hypothesis sceptics and has been criticised for its inclusion criteria and for not fully accounting for statin use and confounding. I think those criticisms deserve to be stated plainly. What I do not think survives is the simple message that lower is always better. The size, consistency and direction of these datasets are not compatible with that claim, even after the caveats.
The heart-attack admission data
A frequently cited observation is that people having cardiac events often do not have high LDL. In an analysis of 136,905 hospitalisations for coronary artery disease across 541 US hospitals, the mean admission LDL-C was 104.9 mg/dL, nearly half of patients had an LDL-C below 100 mg/dL, and only 17.6 per cent were below 70 mg/dL (Ref 3). Notably, more than half had an HDL-C below 40 mg/dL, a metabolic marker rather than an LDL one.
The fair reading, and the fair critique, matter here. This was a cross-sectional snapshot of admission lipids in people who already had coronary disease, not a comparison against a matched population without disease, so it cannot by itself prove that LDL is irrelevant. LDL-C also falls during the acute phase of a myocardial infarction, which can bias admission values downward. And these were coronary-artery-disease admissions broadly, not exclusively confirmed heart attacks. What the data do undercut is the intuition that people who have cardiac events are overwhelmingly the people with the highest LDL. They are not. That should at least prompt humility about using LDL-C in isolation as the number that decides who is at risk.
The strongest case for the lipid hypothesis, taken seriously
I am not interested in a straw man, so here is the mainstream case at its strongest.
First, genetics. People who by chance inherit PCSK9 gene variants that lower their LDL for life have substantially less coronary disease. In the Atherosclerosis Risk in Communities study, nonsense mutations in PCSK9 were associated with a 28 per cent lower LDL-C and an 88 per cent lower risk of coronary heart disease over 15 years in one group, and a separate variant with a 15 per cent lower LDL-C and a 47 per cent lower risk in another (Ref 5). This is a genuinely important finding, and it is the backbone of the Mendelian-randomisation argument that lifelong LDL exposure is causal.
Second, drug trials that lower LDL by mechanisms other than statins. In IMPROVE-IT, adding ezetimibe to a statin after acute coronary syndrome lowered LDL-C further, from a median of 69.5 to 53.7 mg/dL, and reduced the seven-year composite event rate from 34.7 to 32.7 per cent (Ref 6). In FOURIER, the PCSK9 inhibitor evolocumab lowered LDL-C by 59 per cent, to a median of 30 mg/dL, and reduced the primary composite endpoint from 11.3 to 9.8 per cent over a median 2.2 years (Ref 7). Because ezetimibe and PCSK9 inhibitors are not statins and act largely through LDL, these trials are strong evidence that lowering LDL-C itself, not merely some side effect of statins, contributes to fewer events. This is the single best rebuttal to anyone, myself included, who wants to argue that LDL is purely a bystander, and intellectual honesty requires me to put it in front of you clearly.
Here is where I land after taking all of that seriously. The genetic argument is about lifelong exposure across populations and does not translate cleanly into what a specific pharmacological intervention will do for the individual in front of me later in life, and PCSK9 has biological roles beyond LDL. The drug-trial benefits are real but small in absolute terms and are built on composite endpoints that include revascularisation, not only death. IMPROVE-IT's absolute risk reduction was about 2 percentage points over seven years (Ref 6), and FOURIER, despite a striking relative risk reduction, was not shown to reduce cardiovascular or all-cause mortality over its short follow-up (Ref 7). Put most soberly, a systematic review of statin trials found the median postponement of death within the trials was 3.2 days for primary prevention and 4.1 days for secondary prevention (Ref 8). That does not mean the drugs do nothing. It means the absolute benefit is a great deal smaller than the way these medicines are usually described at the bedside, and reasonable people can weigh a few days of average postponement against cost and side effects differently.
Where atherosclerosis forms: endothelium and flow
If LDL concentration were the direct cause of plaque, plaque should appear fairly uniformly wherever the blood carries lipoproteins. It does not. Atherosclerosis is strikingly focal, and it clusters at the outer edges of arterial branch points.
The mechanism that best explains this is haemodynamic shear stress, the frictional force of flowing blood on the cells lining the artery. In a landmark review, higher, smooth arterial shear stress promoted a quiescent, protective endothelial phenotype, whereas the low and disturbed shear stress found at branch points promoted an atherogenic phenotype (Ref 4). The disease tracks with the mechanical-stress pattern, which is not uniform, rather than with the lipid concentration, which is. This is why my own emphasis is on the health of the endothelium and the inflammatory environment around it, rather than on the lipid number alone. Much of this endothelial and shear-stress biology comes from experimental and animal models, so it should be read as mechanism rather than as human outcome data, and humans may differ.
I want to be careful not to overclaim. Shear stress explains where plaque forms; it does not by itself prove that LDL and ApoB are irrelevant to how it progresses. The mainstream "response-to-retention" model holds that ApoB-containing particles retained in the artery wall are a necessary part of lesion growth, and that model is well supported. My honest position is that endothelial damage and inflammation set the stage, and that lipoproteins are involved once that stage is set. Where I differ from the strong lipid view is on primacy, on which comes first and which is the more useful lever, not on whether lipoproteins appear in plaque at all.
LDL versus ApoB: a better number for the same question
ApoB is increasingly promoted as the upgrade to LDL-C, and on its own terms that is fair. Each atherogenic particle carries one ApoB molecule, so ApoB counts particle number, whereas LDL-C measures only the cholesterol mass those particles carry. Two people with the same LDL-C can have very different particle counts, and ApoB captures that. Within the lipid framework, it is the more informative measurement.
My reservation is not that ApoB is a bad measurement. It is that it answers the same category of question as LDL-C: it quantifies circulating particles, which are associated with disease, rather than measuring the disease process itself. I do not routinely order it as a target to drive down, though it often appears on standard panels anyway, and it is worth understanding so that you can follow the conversation when a clinician raises it. That is a matter of interpretation and emphasis, and clinicians I respect weigh it differently.
Statins: absolute benefit and the mitochondrial question
Two things can be true at once. Statins produce a real reduction in events in secondary prevention, and that benefit is smaller in absolute terms than most patients believe. The survival-postponement figures above (Ref 8) are the clearest way to see the second point.
On mechanism and harm, statins inhibit HMG-CoA reductase, the same early enzyme in the mevalonate pathway that also produces coenzyme Q10 (ubiquinone), a molecule central to how mitochondria generate ATP. A review of the statin adverse-effect literature has argued that a mitochondrial mechanism, including reduced CoQ10, may underlie the muscle symptoms that are more frequent with statins than with placebo in randomised trials, and possibly some non-muscle effects too (Ref 9). I want to hedge this properly: that is a review-level hypothesis, not a settled account of population harm, and the mainstream position is that statins are generally well tolerated and that serious adverse effects are uncommon. My own clinical preference leans cautious, but a preference is not proof.
What I will not do is tell you to stop a statin your prescriber has recommended. No disease is caused by a deficiency of statins, and I think the framing around them is often too absolute, but the right move if you have doubts is to discuss the specifics with the clinician who knows your history, not to act on an article. If you have already had a heart attack, this is exactly the situation where the evidence for treatment is strongest and where self-directed changes carry the most risk.
Special cases where the argument gets tested
Familial hypercholesterolaemia. This is where the causal-LDL case is strongest, and it deserves respect. Heterozygous FH affects somewhere between roughly 1 in 200 and 1 in 500 people depending on the population, usually through a faulty LDL receptor, and it is associated with up to a 13-fold higher risk of coronary heart disease (Ref 10). One way to think about the underlying biology is that the receptors, the cellular "letterboxes" that take LDL out of the blood and into cells, are missing or faulty, so cholesterol accumulates in the bloodstream while cells are less able to take delivery. That framing is my interpretation, and it is not the mainstream one, which holds straightforwardly that lifelong high LDL and ApoB drive the excess risk and that lowering them reduces events in FH. I flag my disagreement honestly and add the most important sentence in this section: a confirmed FH diagnosis is precisely the case that must be managed with a prescribing clinician, not on the basis of a blog.
Menopause. Many women see LDL-C rise across the menopausal transition. This is not merely ageing. In the SWAN cohort, total cholesterol, LDL-C and ApoB rose specifically within the year around the final menstrual period, consistent with a menopause-induced change rather than gradual chronological drift, and the pattern held across ethnic groups (Ref 11). My reading is that this is a physiological recalibration as oestrogen declines and lipid handling adjusts, and that treating the number reflexively as a disease misreads it. Hormone replacement is a separate and individual conversation, and I do not see it as the default response to a post-menopausal lipid shift, though it is one tool among several and I am qualified to discuss it.
Lean, low-carbohydrate athletes. Lean people who train hard on low-carbohydrate diets often show dramatically elevated LDL-C. The pattern has been described as the "lean mass hyper-responder" phenotype, defined in one analysis as LDL-C at or above 200 mg/dL with HDL-C at or above 80 and triglycerides at or below 70, occurring in lean, metabolically healthy individuals (Ref 12). The proposed mechanism is prosaic: when little fuel comes from carbohydrate, more fat is trafficked around the body by lipoproteins, so particle numbers rise because the system is doing its job. Strikingly, a 2024 imaging study compared 80 such individuals, mean LDL-C 272 mg/dL on a ketogenic diet for a mean of 4.7 years, against matched controls with an average LDL-C 149 mg/dL lower, and found no significant difference in coronary artery calcium or CT-angiography plaque, and no correlation between LDL-C and plaque in either group (Ref 13). That is a genuinely interesting result. It is also small, short-term and cross-sectional, and the companion progression work suggests plaque can still advance in some of these individuals, so it should be read as preliminary rather than reassuring in a settled way. I would not reflexively treat a lean athlete's isolated high LDL-C without looking at the rest of their metabolic picture, but nor would I claim this question is closed.
What appears to predict risk better in practice
If the disease is fundamentally about endothelial health and inflammation, the more useful question is whether the metabolic environment is driving that inflammation. A coherent panel, read together rather than as isolated numbers, is more informative to me than an LDL-C value alone.
The triglyceride-to-HDL-C ratio is calculable from any standard lipid panel and functions as a rough proxy for insulin sensitivity. In the work of McLaughlin and colleagues, a ratio at or above 3.5 in US (mg/dL) units identified insulin resistance and a preponderance of small, dense LDL particles (Ref 14). One important practical trap: this ratio is unit-dependent, because triglycerides and HDL-C convert between mg/dL and mmol/L by different factors, so a ratio expressed in mmol/L (UK units) is numerically much lower than the same physiology in mg/dL. My own preference is to see the ratio comfortably below 1 in mmol/L units, which is stricter than commonly quoted thresholds; I treat it as a directional signal rather than a hard cut-off.
Fasting insulin is rarely ordered but is arguably the earliest number to move, often rising for years to hold glucose steady before fasting glucose drifts. HOMA-IR, derived from fasting glucose and fasting insulin together, quantifies that relationship and is the most practical measure of what is conventionally called insulin resistance. I prefer the phrase physiological insulin resistance, because in many cases the cells are not broken; they are down-regulating glucose entry as a regulated, protective response, and the "resistance" label implies a malfunction that may not be there. That is my framing, offered as a way of thinking, not as established nomenclature.
HbA1c, the roughly three-month glucose marker, is widely available and useful, but it is not reliable in isolation, and this is the methodological point I would hold myself to as firmly as I hold the lipid hypothesis. HbA1c assumes a standard red-blood-cell lifespan of about 120 days. If red-cell turnover is faster, the reading is falsely low; if slower, falsely high. Measured red-cell survival varies enough between healthy people to shift HbA1c meaningfully for the same average glucose (Ref 15). Pairing HbA1c with a reticulocyte count lets you interpret it in context. Most panels do not do this, and I include myself in the criticism when I have not.
Finally, when a direct look at the artery is genuinely warranted, the coronary artery calcium scan measures calcified plaque itself rather than a marker that travels with it. In a cohort of 25,563 asymptomatic people, a calcium score of zero was associated with very low all-cause and cardiovascular mortality across long-term follow-up (Ref 16). It is not a routine test, it involves a radiation dose, it misses non-calcified "soft" plaque, and it should only follow a conversation about whether it is justified for you specifically. The honest filter is whether the result would change the plan. For most of my patients, the dietary and lifestyle advice is the same whether the score is zero or high, and a test that does not change the decision is a test worth questioning.
What this means in practice
I try to apply one filter to every test and every intervention: does the result actually change what I would do? For most people worried about an LDL number, the sober answer is that the LDL-C in isolation changes very little, and the metabolic panel above changes rather more.
If you have been recommended a statin and want to engage rather than simply comply or refuse, these are measured questions to ask your prescriber, in a friendly and professional spirit: what is the absolute risk reduction, not the relative risk reduction, for someone with my profile; what was the baseline event rate in the trials being relied on; and were those trials conducted in people matched on the metabolic markers I have just described. If the answers are confident and grounded in outcome data, listen carefully. If they are not, that is useful information too.
Your general practitioner is not the problem here. They work inside guidelines from bodies such as NICE, the American College of Cardiology and the American Heart Association, under real medicolegal constraint and limited appointment time, and inside an information environment heavily shaped by industry funding. That structure, not the individual clinician, is what I am critiquing. If you want to see what primary care can look like when it works upstream of the lipid framing, the general-practice service evaluation by Unwin and colleagues reported around 51 per cent of a low-carbohydrate cohort achieving drug-free type 2 diabetes remission over eight years, alongside falls in weight, triglycerides and blood pressure (Ref 18), with the honest caveat that this was an uncontrolled, self-selected service evaluation rather than a randomised trial.
The foundations matter more than any of this, and they are free. Food built around real, minimally processed ingredients that does not chronically spike insulin. Sleep you wake rested from. Daylight on your skin and in your eyes. Regular movement and appropriate strength work. Social connection and lower chronic stress. Supplements are situational tools at most, not the centre of the plan, and I use very few myself. None of the foundations are glamorous, and all of them do more for the endothelium than chasing a single lipid number.
If you would like to work through your own numbers in detail, you do not have to do it with me; there are many excellent clinicians who take this kind of approach. I do offer one-to-one consultations if you would specifically like my input.
Disclosures
I run a private clinic and charge for one-to-one consultations, so I benefit financially when people choose to work with me, and you should weigh this article in that light. I hold a metabolic, low-carbohydrate view of human physiology, which colours my reading of the evidence. I have tried to present the mainstream lipid-hypothesis case at its strongest and to cite the studies that cut against my own position as fairly as those that support it. Several trials cited here were funded by pharmaceutical manufacturers, which I have noted, and one supportive reference on elderly LDL is authored by a group with a declared sceptical position, which I have also noted. I may be wrong on parts of this, and I will update publicly if better evidence emerges.
References
Yi SW, Yi JJ, Ohrr H. Total cholesterol and all-cause mortality by sex and age: a prospective cohort study among 12.8 million adults. Scientific Reports, 2019. https://doi.org/10.1038/s41598-018-38461-y
Ravnskov U, Diamond DM, Hama R, et al. Lack of an association or an inverse association between low-density-lipoprotein cholesterol and mortality in the elderly: a systematic review. BMJ Open, 2016. https://doi.org/10.1136/bmjopen-2015-010401
Sachdeva A, Cannon CP, Deedwania PC, et al. Lipid levels in patients hospitalized with coronary artery disease: an analysis of 136,905 hospitalizations in Get With The Guidelines. American Heart Journal, 2009. https://doi.org/10.1016/j.ahj.2008.08.010
Malek AM, Alper SL, Izumo S. Hemodynamic shear stress and its role in atherosclerosis. JAMA, 1999. https://doi.org/10.1001/jama.282.21.2035
Cohen JC, Boerwinkle E, Mosley TH, Hobbs HH. Sequence variations in PCSK9, low LDL, and protection against coronary heart disease. New England Journal of Medicine, 2006. https://doi.org/10.1056/NEJMoa054013
Cannon CP, Blazing MA, Giugliano RP, et al. Ezetimibe added to statin therapy after acute coronary syndromes (IMPROVE-IT). New England Journal of Medicine, 2015. https://doi.org/10.1056/NEJMoa1410489
Sabatine MS, Giugliano RP, Keech AC, et al. Evolocumab and clinical outcomes in patients with cardiovascular disease (FOURIER). New England Journal of Medicine, 2017. https://doi.org/10.1056/NEJMoa1615664
Kristensen ML, Christensen PM, Hallas J. The effect of statins on average survival in randomised trials, an analysis of end point postponement. BMJ Open, 2015. https://doi.org/10.1136/bmjopen-2014-007118
Golomb BA, Evans MA. Statin adverse effects: a review of the literature and evidence for a mitochondrial mechanism. American Journal of Cardiovascular Drugs, 2008. https://doi.org/10.2165/0129784-200808060-00004
Nordestgaard BG, Chapman MJ, Humphries SE, et al. Familial hypercholesterolaemia is underdiagnosed and undertreated in the general population: consensus statement of the European Atherosclerosis Society. European Heart Journal, 2013. https://doi.org/10.1093/eurheartj/eht273
Matthews KA, Crawford SL, Chae CU, et al. Are changes in cardiovascular disease risk factors in midlife women due to chronological aging or to the menopausal transition? Journal of the American College of Cardiology, 2009. https://doi.org/10.1016/j.jacc.2009.10.009
Norwitz NG, Feldman D, Soto-Mota A, Kalayjian T, Ludwig DS. Elevated LDL cholesterol with a carbohydrate-restricted diet: evidence for a "lean mass hyper-responder" phenotype. Current Developments in Nutrition, 2022. https://doi.org/10.1093/cdn/nzab144
Budoff M, Manubolu VS, Kinninger A, et al. Carbohydrate restriction-induced elevations in LDL-cholesterol and atherosclerosis: the KETO trial. JACC: Advances, 2024. https://doi.org/10.1016/j.jacadv.2024.101109
McLaughlin T, Reaven G, Abbasi F, et al. Is there a simple way to identify insulin-resistant individuals at increased risk of cardiovascular disease? American Journal of Cardiology, 2005. https://doi.org/10.1016/j.amjcard.2005.03.085
Cohen RM, Franco RS, Khera PK, et al. Red cell life span heterogeneity in hematologically normal people is sufficient to alter HbA1c. Blood, 2008. https://doi.org/10.1182/blood-2008-04-154112
Shaikh K, Li D, Nakanishi R, et al. Low short-term and long-term cardiovascular and all-cause mortality in absence of coronary artery calcium: a 22-year follow-up observational study from large cohort. Journal of Diabetes and its Complications, 2019. https://doi.org/10.1016/j.jdiacomp.2019.05.015
McLaughlin T, Abbasi F, Cheal K, et al. Use of metabolic markers to identify overweight individuals who are insulin resistant. Annals of Internal Medicine, 2003. https://doi.org/10.7326/0003-4819-139-10-200311180-00007
Unwin D, Delon C, Unwin J, Tobin S, Taylor R. What predicts drug-free type 2 diabetes remission? Insights from an 8-year general practice service evaluation of a lower carbohydrate diet with weight loss. BMJ Nutrition, Prevention & Health, 2023. https://doi.org/10.1136/bmjnph-2022-000544
