402 ‒ NMR blood analysis: how mortality risk and more can be assessed from a single blood sample

Episode Overview

• Dr. Peter Attia (who discloses his clinical practice uses oral glucose tolerance tests and NMR testing, including the MVX score over the last 4 months) interviews Dr. Jim Otvos, a biochemist who spent 20 years at the University of Wisconsin-Milwaukee studying the active site of zinc metalloenzymes using Nuclear Magnetic Resonance (NMR) spectroscopy before moving to North Carolina State University in 1990.
• The discussion traces the accidental birth of clinical NMR lipidology—beginning when Otvos debunked a 1986 New England Journal of Medicine paper claiming NMR could diagnose cancer—to his filing of the original NMR lipoprotein analysis patent, receipt of $100,000 from Siemens Medical Systems and two NIH grants, and founding of the university spin-off LipoScience in the mid-1990s (acquired by Labcorp in 2014, where Otvos's research group was retained until he recently left to focus on science).
• A central focus is dismantling conventional lipidology myths: explaining why lipoprotein particle concentration (LDL-P and ApoB) rather than cholesterol mass (LDL-C) or particle size ("small dense" vs. "large fluffy" LDL) drives cardiovascular risk, and why LDL-P is critical for managing patients on lipid-lowering therapy.
• The episode explores how a single 30-second NMR scan of 150 microliters ($\mu\text{L}$) of plasma can detect insulin resistance years before glucose rises (via the LP-IR and DRI scores) and measure steady-state systemic inflammation (via GlycA) much more reliably than high-sensitivity C-reactive protein (hs-CRP).
• Finally, the conversation examines the newly developed Metabolic Vulnerability Index (MVX), an age-independent biomarker of "metabolic frailty" that predicts short- and long-term all-cause mortality across both sick and healthy populations from age 30 to 60+.

Key Concepts

• The Accidental Discovery of NMR Lipoprotein Analysis: In 1986, a New England Journal of Medicine paper claimed a narrow NMR signal halfway up its peak in blood plasma diagnosed cancer. Testing 6 leftover healthy hospital samples, Otvos found 3 had narrow signals—all from women who had just given birth (pregnancy was a known false positive). Using a $100,000 grant from Siemens Medical Systems, Otvos’s team separated VLDL, LDL, and HDL and published a paper in Clinical Chemistry (1990 or 1991) proving isolated lipoprotein signals did not differ in cancer; rather, the narrow peak was caused by higher triglycerides and lower HDL-C, a common lipid profile in cancer patients.
• How NMR Measures Lipoproteins vs. Standard Chemical Assays: Standard clinical auto-analyzers measure lipid mass in milligrams per deciliter (mg/dL). For triglycerides, they use a lipase to hydrolyze fatty acids from glycerol and add a reagent to create a color change proportional to glycerol (assuming a 3:1 fatty acid-to-glycerol ratio, which yields false-positive high triglycerides in rare patients with high free glycerol). Historically, labs precipitated out VLDL and LDL to measure HDL-C, and avoided ultracentrifugation by estimating LDL-C via the Friedewald formula: $\text{Total Cholesterol} - \text{HDL-C} - (\text{Triglycerides} / 5)$, where $\text{TG}/5$ estimates VLDL-C (Otvos helped develop a more accurate NIH equation used by Labcorp). In contrast, NMR requires no chemical consumables and takes 30 seconds to measure the terminal methyl groups on fatty acid chains, reporting particle concentration (LDL-P) in nanomoles per liter (nmol/L)—where 1 mole is $6.02 \times 10^{23}$ particles, equating to roughly $10^{16}$ particles in the blood. Due to physical-chemical properties that Otvos hedges are governed by "complex equations that I don't even understand very well," larger particles resonate at slightly lower frequencies (to the left: VLDL, then LDL, then HDL) and smaller particles at higher frequencies.
• Particle Number (LDL-P) Trumps Particle Size: At UC Berkeley/Donner Laboratory, Ron Krauss used gradient gel electrophoresis (which took a couple of days to run) to show that people with predominantly small dense LDL (Pattern B) had a 3-fold higher cardiovascular risk at a given LDL-C than those with large buoyant LDL (Pattern A). Otvos tested ~45 of Krauss's samples and proved NMR could differentiate Pattern A and B in 30 seconds. However, in the Framingham and MESA cohorts, Otvos’s team proved smaller LDL particles contain less cholesterol per particle, meaning at a given LDL-C, Pattern B patients simply have more particles (higher LDL-P). When stratified by LDL-P, particle size adds zero cardiovascular risk. Furthermore, drugs that make LDL and HDL particles larger, raise HDL-C, and lower triglycerides (niacin, fibrates, CETP inhibitors) failed in outcome trials, and patients with homozygous Familial Hypercholesterolemia (FH) have very high LDL-C and LDL-P composed of large LDL particles (e.g., LDL-P > 2,000 nmol/L, ~80th percentile) and die at age 30 to 35.
• LDL-C vs. LDL-P Discordance and Risk Management: In a 2007 Framingham paper and a 2011 MESA paper converting mg/dL (LDL-C) and nmol/L (LDL-P) into percentiles, about 80% of the general population was concordant (or when divided into thirds of ~33% each, concordance was defined as within $\pm 12$ percentile units). Whenever LDL-C and LDL-P were discordant, Kaplan-Meier cumulative incidence curves showed cardiovascular events tracked with LDL-P, not LDL-C. A ~2012/2013 study showed a monotonic increasing relationship between Metabolic Syndrome criteria met (0 to 5: high blood pressure, truncal girth/obesity, fasting glucose, HDL-C, triglycerides) and discordance, making patients with 5 of 5 criteria virtually guaranteed to be discordant. While adding LDL-P or ApoB to initial risk assessment equations (which include Total Cholesterol, HDL-C, diabetes, smoking, and hypertension) does not statistically improve initial risk prediction because low HDL-C acts as an indirect associative proxy for high LDL-P (though Otvos notes "we still don't really understand" if HDL brings extra risk assessment), LDL-P and ApoB are superior for risk management. Statins and other drugs lower LDL-C disproportionately more than LDL-P/ApoB because triglycerides replace cholesterol esters in the particle core, leaving many patients with artificially low LDL-C but persistently high particle numbers.
• Early Detection of Diabetes via LP-IR and DRI: By the time fasting glucose crosses 100 mg/dL (prediabetes) on its way to 126 mg/dL (diabetes), a patient has already lost ~50% of their pancreatic beta-cell function. The Lipoprotein Insulin Resistance (LP-IR) score combines 6 VLDL, LDL, and HDL size and subclass concentration parameters into a 0–100 score. In an unpublished analysis of the Diabetes Prevention Program (DPP), LP-IR predicted incident diabetes better than fasting insulin, decreased with lifestyle intervention (and less so with metformin), and the reduction (delta) in LP-IR strongly predicted reduced diabetes risk. Adding branched-chain amino acids (BCAAs) to LP-IR created the Diabetes Risk Index (DRI), available at Labcorp.
• The Economics and Commercial Bottleneck of the Vantera Analyzer: LipoScience developed the Vantera NMR analyzer (costing ~$400,000 to $500,000 to manufacture) and received FDA clearance for the analyzer and LDL-P in 2011, placing machines in Labcorp, Mayo Clinic, Cleveland Clinic, Scripps Clinic, and ARUP. When Labcorp acquired LipoScience in 2014, they recalled the analyzers from external labs to keep NMR proprietary, halting global in vitro diagnostic (IVD) translation; Otvos warns these first-and-only-generation machines will eventually cease to function. Because NMR uses no chemical reagents, running a 150-$\mu\text{L}$ plasma sample in high volume costs $1 or less and outputs a standard lipid panel, LDL-P, glucose, LP-IR, GlycA, ApoB (FDA cleared via machine learning/AI at zero incremental analytic cost), and MVX. By comparison, CMS reimburses ~$12 for a 14-marker Comprehensive Metabolic Panel (~$60+ if billed individually), ~$13 for a lipid panel, ~$20 for ApoB, and ~$30+ for an NMR LipoProfile.
• GlycA vs. hs-CRP for Systemic Inflammation: GlycA is an NMR signal arising from N-acetylglucosamine (carbohydrate/glycan) decorations on acute-phase proteins and correlates with IL-6 and systemic inflammation. While hs-CRP can spike 1,000-fold during an acute infection and requires averaging 2 to 3 tests due to day-to-day volatility, GlycA only rises ~2-fold during acute infection (Otvos admits he cannot explain the mechanistic "because" behind this muted response, only the observation) and outperforms CRP in mortality prediction models.
• The Metabolic Vulnerability Index (MVX) and "Metabolic Frailty": Discovered top-down via epidemiological data mining ("irrational discovery" that Otvos hedges "raises more questions than answers right now" regarding causality, modifiability, and lack of intervention trials), MVX is a 0–100 score combining two 0–100 sub-scores: the Inflammatory Vulnerability Index (IVX: small HDL-P and GlycA) and the Metabolic Malnutrition Index (MMX: citrate and the 3 BCAAs—leucine, isoleucine, and valine). Otvos doubts separating IVX and MMX will guide different therapies because they represent a single malnutrition-inflammation syndrome. Counterintuitively, while high BCAAs predict obesity, insulin resistance, and diabetes, low BCAAs (and high citrate) reflect muscle wasting, sarcopenia, and cachexia, predicting mortality. Otvos argues MVX measures "metabolic frailty"—the likelihood of dying once you get a disease, rather than the likelihood of developing the disease (noting that links to incident disease in trials are artifacts of composite endpoints combining fatal and non-fatal events).
• Clinical Validation of MVX Across Age and Disease:
• Cardiac Catheterization Cohorts: In Duke University’s CATHGEN biorepository (7,000 patients, average age ~60, plasma stored at $-80^\circ\text{C}$), 5-year mortality was 17%, with 60% of deaths from non-cardiac causes. MVX strongly predicted all-cause mortality across all subgroups (with or without obesity, diabetes, heart failure, prior MI, coronary occlusion, or kidney disease), a finding replicated in a Salt Lake City, Utah cath population published in 2023.
• Older Adults, Heart Failure, and Liver Disease: In a Minnesota heart failure cohort, physical frailty correlated weakly with MVX ($r = 0.2$), and both independently predicted mortality. In a 2022 EPESE study of older adults evaluating 186 variables via causal-inference epidemiology, small HDL-P was the #1 predictor of 2-year mortality. Additionally, a paper published two days before recording showed MVX predicted progression to fibrosis in MASLD (formerly NAFLD) patients on top of FibroScan.
• Healthy 60-Year-Olds and 30-Year-Olds: A recent MESA paper on healthy participants (average age 60, free of chronic disease) showed MVX had an adjusted hazard ratio of ~1.5 to 2.0 per standard deviation (a 2- to 3-fold mortality difference between top and bottom quartiles). In an unpublished study (draft about to be submitted) of 3,000+ healthy 25- to 30-year-olds followed for 35 years (with NMR at baseline and years 10, 15, 20, 25, and 30), the baseline MVX distribution in 30-year-olds was identical to 60-year-olds in MESA (top quartile average ~50–51; bottom quartile ~27), proving MVX is independent of age yet strongly predicts premature 30-year mortality.
• CETP Inhibitors, Obicetrapib, and NMR Algorithm Artifacts: In the Phase 3 BROADWAY and BROOKLYN trials of the CETP inhibitor obicetrapib, LDL-P and LDL-C appeared to drop much more than ApoB. Otvos explains this was an artifact of an older NMR algorithm: CETP inhibitors double HDL-C (to 120–150 mg/dL) by converting small HDL into giant, cholesterol-rich HDL particles that overlap in size with small LDL. To prevent the computer from misclassifying giant HDL as small LDL, the old algorithm barred small LDL from the model, artificially exaggerating the LDL-P drop. Re-analysis with a newer algorithm (used over the last 5 years) showed LDL-P reductions matching ApoB. However, CETP inhibitors also reduce protective small HDL-P by 10% to 20%, leading Otvos to offer "pure speculation" that obicetrapib might underperform in clinical outcome trials.

Quotes

• At 0:04:19 - "Sure enough, half of the signals were narrow and half were broad. Did half of these people have cancer? No, these were—these three people were women who had just given birth. So pregnancy was one false positive that was given in this New England Journal paper." - Otvos explaining how testing 6 hospital samples immediately exposed flaws in the 1986 NEJM cancer test claim.
• At 0:06:43 - "So this was nothing to do with cancer per se, it was to do with the lipoproteins, the lipid levels of people with cancer. Twenty years before that, people had published that people with cancer on the average have higher triglycerides and lower HDL cholesterol." - Otvos revealing the true lipid-driven cause behind the NMR signal attributed to cancer.
• At 0:09:17 - "There were no NMR machines in clinical laboratories. In fact, there are no NMR machines to this day in clinical laboratories." - Otvos highlighting the historical and ongoing absence of standard NMR spectrometers in routine clinical labs outside of specialized analyzers.
• At 0:26:45 - "You're reporting the concentration of the package, not the lipid molecules in the package." - Otvos clarifying the fundamental difference between NMR particle counting (LDL-P in nmol/L) and chemical cholesterol mass assays (LDL-C in mg/dL).
• At 0:28:51 - "So a smaller LDL particle means by definition that it's carrying less cholesterol, less lipid per particle than a large LDL particle." - Otvos explaining the physical basis for why patients with small LDL have higher particle numbers at any given LDL-C level.
• At 0:30:51 - "It's quite straightforward at that point to ask the question: at a given level of LDL particles, if you stratify people according to LDL-P, and then say some people have small LDL, some people have large, is there any difference in their risk? And the answer is no, not a bit." - Otvos debunking the myth that small dense LDL is independently more atherogenic once particle number is controlled for.
• At 0:34:37 - "People with FH have very high LDL cholesterol levels, they also have very high LDL particle levels, but those particles are large, they're not small. And there's, you know, these are people that die when they're 30, 35 years old when they're homozygous FH. So this idea that somehow fluffy, large LDL particles are not to worry about or benign is completely fallacious." - Otvos using homozygous Familial Hypercholesterolemia to prove that large Pattern A LDL particles are lethal when present in high numbers.
• At 0:41:21 - "And so that's what the discordance analysis allowed us to do. Whenever there's a discrepancy in one direction or another, the cardiovascular risk tracks with the particle number, not the cholesterol." - Otvos summarizing the core finding of the Framingham and MESA discordance studies.
• At 0:48:48 - "It's very clear that LDL-P or ApoB is a better biomarker because a lot of people who achieve very low LDL cholesterol have not achieved equally very low LDL particles or ApoB, and they therefore, with visibility to that, would be candidates for more aggressive LDL lowering." - Otvos explaining why particle number is essential for managing patients on lipid-lowering therapy.
• At 0:50:13 - "The LDL particles are lower, but the LDL cholesterol is even lower because not only has the particle number gone down, but the cholesterol in the particles independently has gone down." - Otvos describing the mechanism by which statins create discordance between LDL-C and LDL-P.
• At 0:56:20 - "If you don't know that you're insulin resistant, you're just—you're waiting for the easily measured glucose to become elevated. And now, once that happens, you've lost about 50% of your beta-cell function." - Otvos emphasizing the clinical danger of relying on fasting glucose to detect early metabolic dysfunction.
• At 0:58:13 - "We know that insulin resistance is the cigarette to diabetes as cancer. Do we want to wait until we actually see the glucose rise, which by the way is the biomarker that defines diabetes, when in reality by the time that's happening, there's potentially already cellular damage at the level of the beta cell in the pancreas and it's basically running out of steam?" - Attia framing the necessity of measuring insulin resistance directly before pancreatic damage occurs.
• At 1:00:43 - "People imagine that hyperglycemia, prediabetes, is a risk factor for diabetes. It's not a risk factor, it is the disease, it's just in a less manifest form. So why not address the cause?" - Otvos reframing prediabetes as already-established disease rather than an early warning sign.
• At 1:14:47 - "What people don't know is that at some point in the future, maybe sooner rather than later, these Vantera analyzers, first generation, only generation, are going to cease to function." - Otvos warning about the aging hardware infrastructure of clinical NMR analyzers following Labcorp's acquisition.
• At 1:30:10 - "So it turns out, rather miraculously, that how big this signal is, is a very useful measure of your steady state, of your systemic inflammation level. It's a very stable parameter because it's the integration of lots of different things." - Otvos explaining why GlycA provides a superior, stable readout of chronic systemic inflammation compared to volatile hs-CRP measurements.
• At 1:36:50 - "So the presumption was that these are people that die of cardiac causes. But less than half of them died from cardiac issues; 60% died of non-cardiac, cardiovascular causes." - Otvos discussing the surprising mortality breakdown in Duke University's CATHGEN study of 7,000 cardiac catheterization patients.
• At 1:39:32 - "So high branched-chain amino acids speak to insulin resistance, obesity, diabetes risk. Low branched-chain amino acid levels speak to mortality risk." - Otvos contrasting the bidirectional risk profile of BCAAs in metabolic disease versus muscle wasting and mortality.
• At 1:43:37 - "Out of those 186 things that were looked at, small HDL particles were the most powerful at predicting two-year mortality in these people." - Otvos citing the 2022 EPESE causal-inference study of older adults.
• At 1:54:36 - "MVX scores are virtually unassociated with age." - Otvos highlighting the counterintuitive finding that 30-year-olds and 60-year-olds share the exact same distribution of MVX mortality scores.
• At 2:02:08 - "This is something that has to do with dying, not getting the diseases that cause the death." - Otvos defining MVX as a measure of overall metabolic frailty and survival resilience rather than disease incidence.

Takeaways

• Manage cardiovascular risk and titration of lipid-lowering therapy using lipoprotein particle number (LDL-P or ApoB) rather than LDL-C, as statins lower cholesterol mass disproportionately more than atherogenic particle counts.
• Do not rely on having "large, fluffy, buoyant" (Pattern A) LDL particles as protection against atherosclerosis; if your total LDL particle number (LDL-P) is elevated, your cardiovascular risk remains high regardless of particle size.
• Screen for discordance between LDL-C and LDL-P/ApoB especially if you meet any of the 5 criteria for Metabolic Syndrome (high blood pressure, increased truncal girth/obesity, elevated fasting glucose, low HDL-C, or high triglycerides), as meeting more criteria monotonically increases the probability that your LDL-C understates your true particle risk.
• Do not wait for fasting glucose to reach prediabetic thresholds (100 mg/dL) to intervene on metabolic health, because roughly 50% of pancreatic beta-cell function is already lost by that stage.
• Use NMR-derived Lipoprotein Insulin Resistance (LP-IR) or the Diabetes Risk Index (DRI, which adds branched-chain amino acids) to detect insulin resistance years before glucose rises, and track the reduction (delta) in LP-IR to verify that lifestyle interventions or metformin are successfully lowering your diabetes risk.
• Consider ordering an NMR LipoProfile through Labcorp to obtain multiple high-value cardiometabolic biomarkers—including a standard lipid panel, LDL-P, ApoB, LP-IR/DRI, GlycA, and MVX—from a single blood draw.
• Use GlycA alongside or in place of hs-CRP to assess steady-state systemic inflammation; if using hs-CRP, average 2 to 3 separate measurements to account for its extreme day-to-day volatility (up to 1,000-fold spikes during acute infections, compared to ~2-fold for GlycA).
• Interpret branched-chain amino acid (BCAA) levels in the context of the specific clinical question: treat elevated BCAAs as a marker of obesity, insulin resistance, and future diabetes risk, but recognize low BCAAs (paired with high citrate in the MMX sub-score) as a warning sign of sarcopenia, muscle wasting, cachexia, and elevated mortality risk.
• Utilize the Metabolic Vulnerability Index (MVX) to assess overall "metabolic frailty" and all-cause mortality resilience, recognizing that even young, apparently healthy 25- to 30-year-olds can harbor high-risk MVX scores (e.g., top quartile ~50–51 vs. bottom quartile ~27) that predict premature death over a 30- to 35-year horizon.
• In patients with Metabolic Dysfunction-Associated Steatotic Liver Disease (MASLD, formerly NAFLD), consider adding MVX testing alongside FibroScan to improve prediction of progression to liver fibrosis.
• If evaluating rare patients with unexplained, severely elevated triglycerides on standard chemical lab panels, check for high free glycerol, which causes false positives on standard lipase-based colorimetric assays but is avoided by NMR spectroscopy.
• When interpreting clinical trial data or patient labs involving CETP inhibitors (such as obicetrapib, which doubles HDL-C to 120–150 mg/dL), ensure modern NMR deconvolution algorithms (implemented within the last 5 years) are used so giant HDL particles do not artificially distort small LDL and total LDL-P counts, and rely on ApoB to confirm true atherogenic particle reduction.