Otvos says NMR plus machine learning can output ApoB and lipid panel cheaply
""you can actually feed NMR data to a machine learning algorithm AI and train it to produce accurate lipid panel and APOB information. So we published this four or five years ago that you can actually use the NMR spectrum to produce an extended lipid panel APOB plus lipids.""
About this episode
Peter Attia interviews Jim Otvos, a physical chemist who developed NMR-based blood analysis that underpins advanced lipoprotein testing (LDL-P/HDL-P). Otvos recounts how a disputed 1980s New England Journal of Medicine cancer claim led his lab to identify reproducible NMR signals from lipoprotein particles, enabling rapid particle quantification and later subclass sizing. He argues that apparent risk from “small, dense” LDL is largely explained by higher LDL particle numbers at a given LDL-cholesterol level, warning against the common “large, fluffy LDL is safe” interpretation.
The discussion then centres on discordance between LDL-C and LDL-P, links to insulin resistance, and how NMR-derived composite scores such as LPIR and the Diabetes Risk Index may help identify metabolic risk earlier than glucose-based criteria. Otvos also describes “glyc”, an NMR-derived inflammation marker tied to glycosylated acute-phase proteins, which he says is more stable than CRP.
A major focus is MVX (metabolic vulnerability index), a composite score (including small HDL-P, glyc, citrate and branched-chain amino acids) that Otvos says strongly predicts all-cause mortality across cohorts, including claims of prediction in a young-adult cohort followed for decades. Finally, he gives an insider account of Liposcience’s acquisition by LabCorp, saying it halted plans to deploy FDA-cleared Vanta NMR analysers broadly and raising concerns about the ageing analyser fleet and long-term continuity of NMR testing.
Key takeaways
- Otvos says LDL particle number (LDL-P/ApoB) better reflects LDL-related risk management than LDL-C when the two are discordant.
- He argues LDL size adds little once LDL-P is accounted for, challenging the idea that ‘large, fluffy’ LDL is benign.
- Otvos describes NMR ‘glyc’ as a composite systemic inflammation marker that is less day-to-day variable than CRP and correlates with IL-6.
- LPIR and related NMR metrics are presented as tools for earlier assessment of insulin resistance and future diabetes risk, with Otvos citing unpublished DPP analyses.
- MVX is described as a composite NMR-derived score that predicts all-cause mortality in multiple cohorts; Otvos previews work in a young-adult cohort with long follow-up.
- He says Liposcience’s 2014 acquisition by LabCorp ended efforts to distribute Vanta NMR analysers widely, and warns the existing analyser fleet may be nearing end of life.
- Otvos claims NMR spectra can be processed with machine learning to output an extended lipid panel including ApoB without added analytic cost.