402 ‒ NMR blood analysis: how mortality risk and more can be assessed from a single blood sample
In a Nutshell
NMR spectroscopy of a single blood sample measures lipoprotein particle concentrations, sizes, and inflammatory markers like GlycA, plus branched-chain amino acids and citrate. These data generate the LPIR insulin-resistance score, the MVX metabolic-vulnerability score, and accurate LDL-P, ApoB, and lipid values. Mortality, cardiovascular events, and diabetes risk track far more closely with particle number and these composite scores than with standard cholesterol numbers, so therapy should be guided by particle counts and MVX, not LDL-C.
These notes were generated by AI and may contain inaccuracies.
Peter Attia welcomes Jim Otvos, creator of the NMR LipoProfile test (LDL-P and HDL-P), noting that he has followed Otvos's work since Tom Daypring introduced it to him in May 2011. Attia explains that many listeners have likely had an LDL-P or HDL-P test without realizing it originated with Otvos.
Otvos earned a PhD in chemistry and biochemistry. He spent 20 years in academia using NMR spectroscopy as a structural tool, first at the University of Wisconsin-Milwaukee and then at North Carolina State University starting in 1990. NMR machines exist in every chemistry department in the country and are used by organic chemists to determine the structure of synthesized molecules.
In 1986, a paper published in the New England Journal of Medicine claimed that a simple NMR test could diagnose cancer (positive or negative) regardless of cancer type. The test measured the width of prominent signals in the NMR spectrum of blood plasma at half-height: narrow signals indicated cancer, broad signals indicated no cancer. The paper lacked mechanistic explanation for why this relationship should exist.
Otvos obtained six leftover plasma samples from healthy individuals at a local hospital. Half showed narrow signals and half showed broad signals. The three individuals with narrow signals were women who had recently given birth, demonstrating that pregnancy was a false positive not mentioned in the original paper.
Out of scientific curiosity, Otvos began measuring plasma samples and noticed the supposed cancer diagnostic signal was not symmetrical but had lumps, bumps, and shoulders. The signals were traced to lipids and lipoprotein particles. The narrow signal in cancer patients was actually due to higher triglycerides and lower HDL cholesterol—both known characteristics of cancer patients from literature published 20 years earlier.
Otvos received $100,000 in funding from Siemens Medical Systems after a one-hour presentation. With this funding, samples were obtained from people with and without cancer. Major lipoproteins (VLDL, LDL, and HDL) were separated and analyzed. VLDL signals consistently appeared to the left of LDL signals, which appeared to the left of HDL signals. The superposition of these signals with varying relative concentrations created the different composite signal shapes observed in plasma samples.
Otvos published a paper in Clinical Chemistry in 1990 or 1991 demonstrating that NMR signals from isolated VLDL, LDL, and HDL were identical whether from cancer patients or healthy individuals. The signals were reproducible but appeared at slightly different positions for different lipoprotein classes.
Otvos filed a patent on using the NMR signal to deduce lipoprotein concentrations, encouraged by Siemens advisors despite his lack of interest in commercialization. The patent was issued. Clinical translation would require a commercial entity since no NMR machines existed in clinical laboratories.
From the early 1990s to 1995-1996, Otvos received NIH grants supporting analytic development. Beyond differentiating VLDL, LDL, and HDL, the methodology could distinguish size subspecies within each class (smaller, medium-sized, larger particles). This was initially considered infeasible due to close signal positions.
Otvos connected with Ron Krauss at UC Berkeley's Donner Laboratory, who used gradient gel electrophoresis to differentiate LDL by size and showed that small dense LDL was associated with greater cardiovascular risk at any given LDL cholesterol level. Krauss sent 45 samples with electrophoresis tracings showing pattern A (large LDL) and pattern B (small LDL). NMR analysis successfully distinguished large from small LDL.
The ability to measure small dense LDL (pattern B) with three-fold greater cardiovascular risk at any LDL cholesterol level drove commercialization. Otvos felt unqualified for commercialization but recognized NMR's potential to provide information beyond standard lipid panels.
Standard lipid panels use chemistry-based assays. For triglycerides, lipase hydrolyzes fatty acids from glycerol, and the remaining glycerol triggers a color change proportional to glycerol amount. Since triglycerides contain three fatty acids per glycerol molecule, glycerol measurement allows imputation of triglyceride concentration. However, individuals with high free (non-esterified) glycerol will show falsely elevated triglycerides.
Cholesterol assays add reagents causing color change or UV spectrum shift, quantified against a standard curve. Total cholesterol measures all cholesterol in VLDL, LDL, and HDL without differentiation. To measure LDL cholesterol or HDL cholesterol specifically, lipoproteins must be separated first, then cholesterol assays performed on separated fractions.
For many years, HDL cholesterol was measured by precipitating VLDL and LDL, then assaying cholesterol in the remaining HDL. LDL cholesterol measurement required ultracentrifugation to separate VLDL from LDL—a step most clinical laboratories cannot perform. The Friedewald equation estimated LDL cholesterol as: total cholesterol minus HDL cholesterol minus (triglycerides divided by 5).
Most laboratories still use the Friedewald equation despite recognition that dividing triglycerides by 5 provides inaccurate VLDL cholesterol estimates. An NIH equation developed with Otvos's input, used by LabCorp and other major laboratories, provides improved accuracy by incorporating non-HDL cholesterol and triglycerides.
NMR spectra display frequency on the x-axis and signal amplitude on the y-axis. Different molecular groups show signals at different frequencies depending on chemical environment. Signal amplitude is proportional to the number of contributing nuclei. Organic chemists use NMR to deduce molecular structure from known or predicted spectral patterns.
Otvos focused on a single signal from terminal methyl groups of fatty acid chains carried in lipoprotein particles. The original New England Journal authors suggested cancer caused structural alterations in lipoproteins affecting the signal, but Otvos disproved this by showing signals from isolated lipoproteins were identical regardless of cancer status.
Despite identical lipid composition, signals from the same methyl groups appear at slightly different frequencies depending on particle size. Larger particles produce signals with slightly lower frequency; smaller particles produce signals with slightly higher frequency. This relationship is highly reproducible due to physical-chemical properties.
A simple NMR spectrum acquired in 30 seconds captures the composite signal. A computer model containing reference signals for all VLDL, LDL, and HDL size subspecies decomposes the composite signal into constituent parts. The sum of deduced signal intensities from all subspecies recreates the measured composite signal shape.
NMR measures lipoprotein particle concentration (reported as nanomoles per liter) rather than lipid mass concentration (milligrams per deciliter). The signal reflects particle number regardless of variable cholesterol and triglyceride content within particles. A mole contains 6.02 × 10²³ particles; typical LDL particle concentrations are on the order of 10¹⁶ particles per liter.
Otvos's studies in large populations, including Framingham, replicated Ron Krauss's findings: at any given LDL cholesterol level, individuals with predominantly small dense LDL had significantly higher cardiovascular risk than those with large LDL. Multi-linear regression analyses confirmed that LDL particle size adds predictive value beyond LDL cholesterol alone.
LDL particles are always fully saturated with lipid regardless of size. A smaller LDL particle therefore carries less cholesterol per particle than a larger LDL particle. At any given level of LDL cholesterol, people with small dense LDL have higher cardiovascular risk. However, these individuals also have more LDL particles overall, meaning their LDL particle number (LDLP) is higher than would be predicted from the LDL cholesterol measurement alone. An alternative explanation for the apparent extra risk associated with small dense LDL is therefore that there are simply more particles rather than the size itself being the determining characteristic of atherogenicity.
Imagine four lipoproteins each containing three units of cholesterol, totaling 12 units of cholesterol. Alternatively, imagine three larger lipoproteins each containing four units of cholesterol, also totaling 12 units. Both scenarios deliver the same total cholesterol, but one person has four particles while the other has three. The question becomes whether the higher-risk person faces elevated risk because their particles are smaller or because they have more particles that happen to be smaller.
To test whether particle size or particle number drives risk, researchers stratify people according to LDL particle number and then compare those with small LDL versus large LDL at the same particle concentration. When LDL particle number is held constant, there is no difference in cardiovascular risk between individuals with small LDL versus large LDL. The risk is identical whether someone has ten big and ten small particles or fifteen big and five small particles at the same total particle count.
If only assessing cardiovascular risk, cholesterol information combined with size information is equivalent to particle information combined with size information. Size does not add predictive value beyond particle number. However, if clinicians believe small LDL is particularly harmful and attempt to enlarge particles therapeutically while LDL particle concentration remains at an acceptable level, they may pursue unnecessary interventions expecting clinical benefit.
Several drug classes have effects that enlarge LDL particles, including CETP inhibitors, niacin, and various HDL-targeted drugs. Triglyceride-lowering drugs also increase LDL particle size. Clinical trials of niacin were expected to show efficacy because they modestly raised HDL cholesterol, lowered triglycerides, and made both LDL and HDL particles larger. However, outcome studies demonstrated no efficacy. This lack of benefit has been reproduced with fibrates and other drugs targeting particle size.
Patients sometimes report high LDL cholesterol and high LDL particle number on NMR testing but claim their particles are all large, using terms like "fluffy" or "buoyant." They argue they have pattern A rather than pattern B and therefore need not worry despite LDL particle numbers exceeding 2,000 nanomoles per liter. This is a fallacious idea. Data demonstrate that LDL particle number, not size, determines risk. Familial hypercholesterolemia provides genetic evidence: affected individuals have very high LDL cholesterol and LDL particle levels with large particles, yet they experience early cardiovascular mortality, with homozygotes dying in their thirties.
A 2007 Framingham study and a 2011 MESA study examined discordance between LDL cholesterol and LDL particle number. Discordance occurs when LDLP and LDLC disagree in either direction. To define discordance, LDL cholesterol values in milligrams per deciliter are converted to percentiles, allowing direct comparison with LDL particle number percentiles. Individuals within plus or minus 12 percentile units are considered concordant. This threshold was selected to create approximately equal-sized groups: roughly 30 percent concordant, 30 percent discordant with LDLP higher than LDLC, and 30 percent discordant with LDLC higher than LDLP.
Kaplan-Meier curves display cumulative incidence of cardiovascular events over time. In the MESA analysis, the concordance group showed intermediate risk. When LDL particle number exceeded LDL cholesterol, cardiovascular risk was higher than predicted by LDL cholesterol alone. When LDL cholesterol exceeded LDL particle number, cardiovascular risk was lower than predicted. Risk tracked with particle number rather than cholesterol content. Large LDL particles also enter the arterial wall and, when oxidized, deposit their cholesterol content. The discordance analyses demonstrated that cardiovascular risk follows particle number, not cholesterol concentration.
Current cardiovascular risk equations incorporate total cholesterol, HDL cholesterol, diabetes status, smoking status, and hypertension. Adding LDL particle number to these models does not improve risk assessment beyond what the standard equations already provide. HDL cholesterol in the model captures some information about LDL particle number because low HDL cholesterol correlates with higher LDL particle concentrations relative to LDL cholesterol. The HDL cholesterol variable therefore indirectly reflects LDL particle burden through metabolic associations rather than through a causal pathway.
A histogram analysis examined the relationship between the number of metabolic syndrome criteria met and the probability of discordance between LDL cholesterol and LDL particle number. With zero metabolic syndrome criteria, discordance probability is low. As the number of criteria increases from one to five, the probability of discordance rises monotonically. Individuals meeting all five metabolic syndrome criteria are virtually guaranteed to be discordant, with higher LDL particle number than LDL cholesterol would suggest. This indicates that greater discordance signals underlying metabolic dysfunction and elevated insulin resistance.
Although LDL particle number does not improve risk assessment beyond standard equations, it provides superior information for risk management. When managing high-risk patients, the goal is determining how much LDL lowering is required. LDL particle number or ApoB serves as a better biomarker for whether LDL-associated risk has been adequately controlled. Many patients achieving very low LDL cholesterol have not achieved equivalently low LDL particle numbers or ApoB levels and may therefore benefit from more aggressive LDL-lowering therapy such as adding a PCSK9 inhibitor.
As LDL levels are lowered with statins or other therapies, cholesterol within remaining LDL particles is increasingly replaced by triglyceride through lipid exchange. Triglyceride-rich particles donate triglyceride to LDL particles in exchange for cholesterol ester. Consequently, LDL particle number decreases, but LDL cholesterol decreases even more because both particle count and cholesterol content per particle decline. This explains why discordance becomes more pronounced at lower LDL concentrations.
The LPIR score was developed as a composite measure incorporating six parameters: VLDL, LDL, and HDL size and subclass concentrations. The score ranges from 0 to 100, with higher scores indicating greater insulin resistance. The score was designed to assess insulin resistance and diabetes risk more effectively than the simple triglyceride-to-HDL cholesterol ratio. The rationale parallels cardiovascular disease prevention: insulin resistance over time leads to beta cell dysfunction and eventual hyperglycemia. Primary prevention of diabetes requires identifying insulin resistance before glucose elevation occurs, as conversion to diabetes at a glucose threshold of 126 milligrams per deciliter indicates approximately 50 percent loss of beta cell function.
The prevention approach involves identifying causal markers that indicate disease risk before the disease itself becomes measurable. The smoking and lung cancer example illustrates this principle: smoking represents a causal marker for lung cancer, and treatment should begin immediately upon identifying the causal driver rather than waiting for measurable disease. This principle applies equally to elevated LDL cholesterol, APOB, LDL particles, and hypertension in cardiovascular disease.
Insulin resistance functions as the equivalent causal marker for diabetes. Waiting until glucose levels rise represents waiting for the biomarker that defines diabetes, at which point cellular damage at the beta cell level in the pancreas may already be occurring. An oral glucose tolerance test can identify insulin resistance years before glucose elevation occurs, but this test has fallen out of clinical favor due to requiring two hours and being too cumbersome for routine use.
The LDL particle versus LDL cholesterol argument faced tremendous resistance from the medical establishment despite its compelling logic. Medical school training on cholesterol creates barriers to adopting new biomarkers. The same resistance pattern applies to diabetes, where prediabetes is viewed as a risk factor rather than the disease in a less manifest form.
NMR analysis has demonstrated the relationship between the insulin resistance score and likelihood of future diabetes. The insulin resistance score can predict whether someone who is prediabetic is more or less likely to transition to diabetes. The transition to prediabetes does not guarantee progression to diabetes, as demonstrated by cases of stable prediabetes persisting for 25 years without progression.
The LPIR (Lipoprotein Insulin Resistance) score has been validated on longitudinal data to predict type 2 diabetes development. The LPIR score performs better than fasting insulin for assessing insulin resistance. A study in the Diabetes Prevention Program, though not yet published, demonstrated that LPIR and related NMR measures are independently predictive and superior to insulin measurements.
The Diabetes Prevention Program was a 20-year-old study that examined lifestyle intervention and metformin effects on diabetes transition. All participants had NMR analysis performed. Lifestyle change and weight loss significantly reduced diabetes incidence, while metformin was less effective but still significant compared to placebo. The LPIR score was significantly reduced by lifestyle intervention and less significantly by metformin.
Branched-chain amino acids are also related to insulin resistance and can improve the LPIR score. A diabetes risk index (DRI) has been developed that integrates branched-chain amino acids with the LPIR score.
The diabetes risk index is commercially available through LabCorp. Liposcience, founded in the mid-1990s, was acquired by LabCorp approximately 10 years ago. The transition from ordering Liposcience tests directly to receiving results through LabCorp has occurred, though broad clinical translation has not been achieved.
Liposcience began as a university spin-off with the goal of making NMR testing available to any laboratory worldwide rather than becoming a large testing laboratory. Initial operations used research NMR spectrometers, but the vision required developing an analyzer that could perform NMR testing efficiently and rapidly, ideally in 30 seconds or less.
The transition from a laboratory testing company to an in vitro diagnostics (IVD) company required developing the Vanta NMR analyzer. This analyzer was designed to operate like a regular chemistry analyzer, requiring no NMR expertise from operators. FDA clearance for LDL particle measurement and the Vanta analyzer was obtained in 2011.
Vanta analyzers were distributed to major laboratories including LabCorp, Mayo Clinic, Cleveland Clinic, Scripps Clinic, and ARUP. The business model involved laboratories paying Liposcience for each analysis performed on the analyzers.
Liposcience went public, leading to LabCorp's acquisition primarily for financial reasons to eliminate per-test payments. LabCorp, being a laboratory testing company rather than an IVD company, did not share the vision of widespread NMR analyzer distribution. The IVD business model ended in 2014 when LabCorp took back analyzers from other reference laboratories.
LabCorp does not engage in the marketing and awareness creation activities that Liposcience performed to prove clinical efficacy of new tests. This has resulted in Liposcience technology becoming largely invisible despite continued use by those already familiar with the testing.
The Vanta analyzers, manufactured approximately 15 years ago, represent the only generation of NMR analyzers. Superconducting magnets last a long time without degradation, but the computers and moving parts create concerns about longevity. No second-generation analyzers have been developed.
Each Vanta analyzer costs approximately $400,000 to $500,000. The machines have no consumables associated with assays, making them cost-effective for high-volume settings. A single NMR scan can provide information on cardiovascular risk, inflammation level, diabetes risk, and overall mortality risk.
NMR spectrum data can be processed through machine learning algorithms to produce accurate lipid panel and APOB information at no incremental cost beyond the basic NMR scan. This represents significant cost savings compared to performing APOB testing separately through chemistry methods.
The MVX score measures six components: small HDL particles, GlycA, citrate, and the three branched-chain amino acids (leucine, isoleucine, and valine). When normalized to a 0-100 scale with higher scores indicating worse outcomes, the MVX score demonstrates remarkable predictive value for mortality, cardiovascular death, and liver disease.
The score has been validated in both high-risk catheterized patient populations and healthy populations. The development of MVX was not based on mechanistic understanding of the individual components but rather on the efficiency of NMR analysis for measuring baseline samples from large longitudinal studies.
Baseline samples from major studies have been analyzed using NMR, including MESA (Multi-Ethnic Study of Atherosclerosis) with over 20 years of follow-up, the Women's Health Study with 26,000 women, Birmingham clinical trials, Framingham Offspring Study, and intervention trials including JUPITER and the Diabetes Prevention Program.
NMR spectra are stored and can be reanalyzed when new extraction capabilities are developed, allowing retrospective analysis of large clinical trials for new biomarker validation. The Diabetes Prevention Program data demonstrated that reductions in LPIR score and branched-chain amino acids powerfully predict reduced diabetes risk, providing evidence for causality.
A Duke University cardiac catheterization study recruited 7,000 people over several years into a biorepository when they presented to the catheterization laboratory with chest pain or other qualifying cardiac issues.
Stored blood samples from patients who underwent coronary angiography at Duke were analyzed using NMR. These samples were collected at baseline and stored at minus 80 degrees, which maintains perfect stability for NMR analysis. Research revealed that small HDL particles were particularly powerful in relating to the likelihood that someone would die during the roughly five or ten-year follow-up period of the study.
The gly signal is another signal in the NMR spectrum that is not located where the signal is interrogated for LDL particles. For ten to fifteen years, this signal was not considered relevant. The gly signal is superimposed on a broad signal being used to interrogate the fatty acid composition of plasma, specifically how much monounsaturated fats were in the blood. A method was developed to quantify how big the gly signal was so it could be subtracted from the other signal, but this quantification provided perfectly good information about the gly signal itself.
Using MESA data and software developed to measure this sharp signal, analysis revealed that the gly signal relates to all sorts of things including mortality very strongly. Literature review showed that this signal was identified twenty years ago as an inflammatory marker. The signal comes from the carbohydrate, the glycan decoration on many proteins in the blood, most of which are acute phase proteins that increase in concentration during inflammatory conditions. The signal is a composite that quantifies the most abundant four or five acute phase proteins contributing to it, and the glycan decoration serves various signaling purposes.
The gly signal reflects not only the levels of acute phase proteins but also how much glycan is on them, which is also connected to inflammation. This signal provides a very useful measure of steady state systemic inflammation level. Unlike CRP, which is very volatile and goes up and down day to day, the gly signal is a very stable parameter because it integrates lots of different things. Clinical recommendations for CRP use suggest taking the average of two or three different measurements, but nobody does this, whereas the gly signal does not suffer from this biological variability problem. When added to CRP in prediction models, the gly signal typically assesses outcomes more strongly than CRP does, though CRP tends to add independently.
The gly signal correlates strongly with IL-6 and other interleukins. However, this marker is very unspecific and will not tell anything about local inflammation as opposed to systemic inflammation. The gly signal offers a simple and very cheap marker of the things that systemic inflammation contributes to, as it comes along for the ride with other NMR information.
When seeing a brand new patient with hs-CRP of two or two and a half, this is more disconcerting than if it's 40, because a level of 40 is so high that it's clearly in response to something acute like getting over a cold or receiving a vaccine. Very high CRP levels speak to acute inflammation which is less pathologically concerning, while low simmering levels are more concerning. The gly signal is very different from this pattern. CRP levels could go up a thousandfold during an infection, but gly levels show a much more muted response because the signal contains so much information in it.
If someone has an active infection, the gly level will be higher by twofold, not a thousandfold, so it is not immune to those changes. In people with inflammatory diseases such as rheumatoid arthritis and psoriasis, gly levels are significantly elevated and are responsive to anti-inflammatory treatment. The gly signal has all the characteristics of a useful biomarker to assess systemic inflammation, which is increasingly understood as an important contributor to mortality risk even though the mechanistic fine points are not fully understood.
The discovery approach for these biomarkers is different from typical biomarker discovery. Rather than starting from bottom up with a mechanistic idea and a particular enzyme to target as therapy, this approach uses large databases of population studies to discover things that were not being looked for in the first place. This top-down approach is more efficient for doing epidemiology, and measuring these things by NMR is not as expensive or difficult as traditional methods.
The MVX biomarker has three components. The lipoprotein component is small HDL particles. The inflammatory component is the gly signal. The third component is the metabolic component consisting of citrate and the three branched-chain amino acids. In the CathGen population, which had a 17% five-year mortality rate in people with a mean age of 60, less than half died from cardiac issues, with 60% dying from non-cardiac causes. Gly and small HDL particles predicted mortality in this population.
Looking at all other measurable parameters, branched-chain amino acids (isoleucine, leucine, and valine) and ketone bodies and plasma protein individually had significant associations with mortality. However, these were not used as part of the MVX composite because the goal was to find things that contributed independently and additively to the other markers. The inflammatory part (gly and small HDL particles) was combined into a subscore called the inflammatory vulnerability index (IVX). The branched-chain amino acids and citrate added independently to the prediction of mortality, and together with the inflammatory markers formed the MVX composite.
Literature review revealed powerful evidence about mortality risk in people with several acute diseases, especially kidney disease and dialysis patients, heart failure patients, and anything with cachexia. The literature describes this vulnerability as coming from a so-called malnutrition-inflammation syndrome, also called protein-energy wasting syndrome. Inflammation is probably the context that allows these dysregulated metabolic situations to exist. Branched-chain amino acids are related to mortality in the opposite direction from their relationship to diabetes risk, where high levels speak to insulin resistance, obesity, and diabetes risk, but low levels speak to mortality risk.
Subgroup analysis within the 7,000 people in the CathGen study showed that the MVX association with mortality exists equally strongly in men and women, in people with and without obesity, diabetes, heart failure, previous myocardial infarction, coronary catheterization occlusion, and kidney disease. The relationship is not affected by any of these conditions and is equally strong, if not stronger, in people without chronic disease versus those who are disease-free. The hazard ratios were bigger in the CathGen study than in the MESA study of healthy people.
The two subparts of MVX are the IVX inflammation part and the other four parameters combined to form MMX, the metabolic malnutrition index. While there is synergy between these components as part of a syndrome, it might be clinically useful to distinguish whether a high MVX score is due more to inflammatory reasons or metabolic malnutrition wasting reasons, as different therapies might be better suited for each case. Both subscores are reported on the same 0-100 scale as the aggregate score.
The 2023 first publication of MVX in the cardiac catheterization cohort included replication in a completely independent cardiac catheterization population in Utah, Salt Lake City, which replicated very well. Since then, MVX has been tested in other disease populations including heart failure patients, where frailty information was also available. Physical frailty showed a fairly weak correlation with MVX with a 0.2 correlation coefficient. Frailty on top of MVX score adds considerably to mortality prediction, but MVX very powerfully still predicts mortality in the presence of frailty score.
Many studies examine periods longer than five years to gain statistical power from more people having the outcome. The short window of prediction is actually a feature because there are few short-term predictive biomarkers available.
In the EPIC study of older people with 186 different variables measured, small HDL particles were the most powerful at predicting two-year mortality among all variables examined using high-tech methodology to assess causal relations.
Good data now exist in different disease populations showing that MVX replicates and relates to mortality risk regardless of who you are. The next step is to determine whether there are interventions which lower MVX and then conduct studies showing that lowering MVX is connected to reduction in mortality risk.
In a study of 3,000+ people entered between ages 25 and 30 with NMR analysis done when average age was about 30, MVX showed a relationship to mortality even in this young population, though the relationship is weaker than in older populations. Blood samples were taken at intervals more frequent than five years for the first few years, then five years after, with NMR data available at years 10, 15, 20, 25, and 30.
MVX scores are virtually unassociated with age. This was proven by the 30-year-old study showing that MVX can predict mortality even in young people. In the MESA study restricted to healthy 60-year-olds with no self-reported diseases, MVX by quartile had a stepping stone relationship to mortality that was significant but not as strong as in CathGen, with hazard ratios of about 1.5 to 2 per standard deviation adjusted.
The distribution of MVX scores in people aged 30 is nearly identical to that observed in 60-year-olds, with the same wide range including both low and high scores. This finding comes from 30-plus years of follow-up data, making it a study of premature mortality. All participants were healthy at baseline with no pre-existing comorbidities, establishing that the wide distribution occurs in genuinely healthy young adults rather than reflecting existing disease.
The average MVX score in these 30-year-olds was approximately 50-51, with the bottom quartile around 27. Some individuals had scores as high as 75, placing them in the upper quartile despite being young and healthy.
The question of how individuals acquire high or low MVX scores at age 30 introduces the possibility that these scores reflect something present at birth rather than acquired through disease vulnerability. Current research cannot determine whether MVX scores are established early in life, and ongoing work examines younger populations to understand this trajectory.
Available genomic and epigenomic data from studies with MVX measurements provide opportunities to investigate the genetic and epigenetic determinants of these scores.
Published papers suggest associations between MVX and diseases including cardiovascular disease, diabetes, and dementia. However, these associations may be artifacts of how clinical trials combine fatal and non-fatal events into composite endpoints.
Most cardiovascular endpoint trials group fatal and non-fatal myocardial infarctions together as cardiovascular disease events. While this approach makes sense for biomarkers like cholesterol that mechanistically connect to disease development, MVX appears to influence whether individuals die from diseases rather than whether they develop the diseases themselves.
MVX appears to measure metabolic frailty - a susceptibility to dying from whatever disease or event occurs, rather than predicting disease onset. Individuals with high MVX scores at age 30 appear identical to those with low scores, showing no visible signs of frailty, yet metabolically they demonstrate increased susceptibility to mortality.
This distinction separates MVX from traditional biomarkers that predict disease development. MVX influences dying sooner versus later rather than the likelihood of developing the diseases that cause death.
The concept extends to predicting therapeutic intervention outcomes, particularly in oncology. Using Keytruda as an example, patients with identical PD-1 mutations and identical cancers show variable responses to treatment. MVX could potentially serve as a prognostic indicator for intervention success probability in addition to mortality risk.
Surgeons lack objective biomarkers to assess patient resilience or likelihood of surviving or benefiting from surgery. MVX could potentially screen patients for metabolic frailty, addressing gaps in current frailty assessment that rely on malnutrition indicators and physical wasting measures.
Recent research on MASLD (metabolic dysfunction-associated steatotic liver disease, formerly NAFLD) demonstrates MVX adds predictive value beyond FibroScan for predicting subsequent fibrosis in these patients. This application addresses the challenge of clinical trials where events are too rare or too distant, allowing enrichment of study populations for mortality endpoints.
MVX testing requires only 150 microliters of plasma from the same specimen used for NMR lipoprofile, glucose, LPIR, and other measurements. When performed in high-volume settings, the test costs approximately one dollar or less to perform.
The comprehensive metabolic panel analogy illustrates the economics: 14 measurements reimbursed by CMS at $12 despite individual test costs totaling over $60, demonstrating how economies of scale enable profitable testing at low per-test costs.
NMR analysis provides LDL particle number plus triglyceride-rich particle subspecies information that may have differential disease relationships beyond what APOB alone measures. APOB represents a single measurement of all apolipoprotein B on LDL and VLDL particles.
FDA clearance was obtained for APOB measurement quality from NMR spectra using machine learning approaches. The extended lipid panel concept would add APOB without analytic cost, though current Medicare reimbursement pays approximately $20 for APOB and $13 for lipid panels separately.
LPIR was added to NMR lipoprofile as a laboratory-developed test rather than FDA-cleared assay. LDL particle number received FDA clearance after significant difficulty. Laboratory-developed tests require CLIA certification for performance quality but not FDA demonstration of clinical utility.
NMR testing faced commercial challenges due to payer resistance to incremental costs beyond existing lipid panel reimbursement. NMR lipoprofile reimbursement is approximately $30.
In CETP inhibition studies including obicetrapib trials, NMR showed greater LDL particle and LDL cholesterol reductions than APOB reductions. This discrepancy resulted from NMR algorithm difficulties with abnormal HDL particles created by CETP inhibition.
CETP inhibition doubles HDL cholesterol not through increased HDL particle number but through conversion of smaller to larger particles containing 5-10 times more cholesterol per particle. HDL particle number actually decreases slightly overall.
Under CETP inhibition, HDL particles become perilously close in size to small LDL particles, creating potential confusion in NMR analysis. The algorithm used in these studies did not allow for small LDL particles at all, creating artifactual LDL particle reductions that did not align with APOB measurements.
Reanalysis with updated algorithms shows LDL particle results more consistent with APOB reductions, addressing the earlier discrepancy.
Small HDL particles demonstrate powerful inverse relationships with all-cause mortality. CETP inhibitors reduce small HDL particles by 10-20%, potentially counteracting beneficial LDL effects. The biological plausibility stems from proteins associated with small HDL particles involved in antioxidation and anti-inflammation processes.
This raises the possibility that trial outcomes for obicetrapib may be influenced by opposing effects: beneficial LDL reductions versus potentially harmful small HDL particle reductions.
The diagnostic space faces significant hurdles in attracting investor interest compared to therapeutics, largely due to the reimbursement environment in the United States. This structural issue makes it difficult to generate commercial success even with valuable diagnostic tools.
The MVX platform offers potential beyond traditional diagnostic applications. When paired with therapeutic interventions, it could demonstrate substantial cost savings through improved therapeutic selection. Several trials would be needed to validate this approach, but this represents the most promising commercial pathway forward.
A comprehensive vision involves incorporating multiple biomarkers into routine yearly physicals at no incremental cost, including lipid panels along with glucose, LPIR, and glyc MBX measurements. In healthcare systems outside the United States with national healthcare structures, there's greater emphasis on diagnostic efficiency—achieving more information with less work and expense carries real value.
The United States market proved challenging for this efficiency-focused model. After attempting to implement this approach domestically, the decision was made to step away from these commercial challenges to focus remaining time on scientific pursuits rather than continuing to navigate reimbursement barriers.
The discussion concludes with appreciation for the time and insights shared during the conversation.
Keep Peter Attia MD in your library
Save the videos and channels worth coming back to, and find them again in one place.
