Biomarkers & Diagnostics: What to Measure and What to Trust

A biomarker earns trust only when changing it changes the hard endpoint. The sharpest, most repeated lesson in this substrate is the surrogate trap: a marker can shift reliably while the hard endpoint does not, or moves the wrong way. Niacin raised HDL and lowered LDL yet produced no cardiovascular benefit and net harm [1]. So the honest map of diagnostics sorts into two piles: markers whose level is a validated prognostic read, and the places where reading a change in the marker is a mistake.

ApoB is the biomarker that earns the most trust. Atherosclerotic disease is caused by cumulative exposure to apoB-containing lipoproteins, established by Mendelian randomization and trials, not mere association [2]. ApoB, the count of atherogenic particles, is the best single lipid risk measure, superior to LDL-C, non-HDL, or particle size [3]. Lowering LDL with statins cuts events proportionally and dose-dependently [4]. Lp(a) is a separate, genetically-fixed, causal axis: genetically improved Lp(a) raises MI and ASCVD risk independent of LDL [5], the argument for measuring it once, since a person with "normal" cholesterol can still carry substantial inherited risk.

Blood pressure is the cleanest causal, dose-dependent lever. Each 10 mmHg lower systolic BP reduces major CV events ~20%, stroke ~27%, heart failure ~28%, and all-cause mortality ~13% [6]. The benefit holds in primary and secondary prevention with no reliable baseline threshold [7]. Targeting <120 vs <140 cut events ~25% and mortality ~27% in high-risk adults, at the cost of more hypotension, syncope, and acute kidney injury [8].

Imaging: calcium and ankle pressure are trustworthy; the carotid is a trap at the wrong read. Coronary artery calcium predicts coronary events across ethnicities, rising steeply with burden [9], and adding it to standard risk factors reclassifies intermediate-risk people [10]. A CAC of zero marks a low near-term risk state (~1% 10-year mortality), but not an absolute free pass: within CAC=0, smokers and diabetics retained meaningful risk [11]. A low ankle-brachial index (≤0.90) independently predicts cardiovascular and all-cause mortality, roughly doubling 10-year mortality within every Framingham category [12]. Carotid intima-media thickness is where the distinction bites hardest. Baseline CIMT is prognostic, per 1-SD higher, MI RR 1.26 and stroke RR 1.32 [13]. But change in CIMT is not a valid surrogate. In an individual-participant meta-analysis (n≈36,984), CIMT progression did not predict events [14]. Across 41 RCTs, treatment-induced CIMT regression showed no relationship with event reduction [15]. A trial that "reversed CIMT" has not, on that basis, shown benefit.

Fitness markers are the strongest modifiable predictors, and one is causal-grade. Cardiorespiratory fitness (VO2max) inversely predicts all-cause mortality with no observed ceiling. Elite versus low fitness runs roughly 5× lower mortality, and low fitness tracks as badly as smoking, diabetes, or hypertension [16], with about 13% lower mortality per 1-MET of fitness [17]. Fitness also passes the causation test most markers fail: raising it with training reverses cardiac aging in an RCT [18].

Body composition: measure muscle, but trust strength over mass. Appendicular lean mass index (ALMI) is the foundational quantitative definition of sarcopenia [19], with weakness-anchored cutpoints [20], and current consensus makes low strength the primary criterion and low mass only confirmation [21]. The reason is a surrogate correction: in the same cohort, muscle strength predicted mortality while muscle mass did not [22], and ALMI was not independently associated with mortality once the analysis accounted for function [23], consistent with grip strength out-predicting systolic BP for mortality [24]. Bone mineral density is a parallel case: each 1-SD lower BMD raises fracture risk ~1.5× [25], a hip fracture carries large sustained excess mortality [26], and antiresorptive drugs that raise BMD do cut fractures (denosumab, alendronate) [27, 28]. But "BMD up → fractures down" does not generalize to every lever: exercise raised BMD yet the Cochrane review found no significant fracture reduction [29].

Status markers: read them as risk signals, not dials to turn. A low omega-3 index tracks higher cardiovascular risk and lower survival observationally. A 17-cohort meta found ~13–17% lower all-cause mortality in the highest omega-3 quintile [30], and a Framingham analysis put the top-versus-bottom gap at 4.74 years of life expectancy, with low omega-3 status comparable to smoking [31]; low index also tracks sudden cardiac death [32]. These are observational, they cannot rule out that high-omega-3 people differ in other health behaviors. And the omega-3 story carries a genuine primary harm signal: supplementation raises atrial fibrillation risk dose-dependently (+12% near 1 g/day, ~+50% at 1.8–4 g/day) [33]. Finally, the terminal niacin/NAD metabolite 4PY causally drives vascular inflammation in mice and tracks 3-year MACE risk in humans independent of standard risk factors [1], a reminder that a "boost NAD+" strategy has a measurable downside marker.

Epigenetic clocks are the best available biomarkers of biological aging, and the clearest cautionary tale. As predictors they are real: GrimAge is the strongest mortality clock [34], PhenoAge beats calendar-age clocks [35], DunedinPACE captures aging rate [36], and the pan-tissue Horvath clock tracks chronological age [37]. But whether lowering a clock lengthens life is unproven. First-generation clocks also carry enough replicate noise to make a single consumer reading nearly meaningless [38], so this ships as contested below, never as "intervention X reversed aging."

Where the Evidence Is Contested

Epigenetic clocks such as GrimAge [34], PhenoAge [35], and DunedinPACE [36] predict mortality and healthspan outcomes. Their readings also shift in response to interventions like caloric restriction and omega-3 supplementation. To their proponents, this responsiveness to known health-promoting interventions is exactly what a useful surrogate marker of aging should show.

Critics push back on whether any of that constitutes evidence of causation rather than correlation. Even the inventors of these clocks concede that the causal link remains unproven. First-generation clocks carry replication noise of up to 8 to 9 years [38]. That is a wide margin of error for a tool being considered as a clinical endpoint. Kaeberlein has argued that functionally grounded measures such as grip strength and VO2max are better-validated than epigenetic scores for tracking what aging does to a person. The debate stays open for a simple reason. Predicting an outcome and measuring the biological process that drives that outcome are not the same thing, and the field has not yet found a way to tell them apart.

References

1. Ferrell et al. 2024, Nature Medicine — niacin 4PY & CVD · PubMed 38374343
2. Ference et al. 2017, Eur Heart J — LDL causes ASCVD (EAS consensus) · PubMed 28444290
3. Marston et al. 2022, JAMA Cardiol — apoB primary risk driver · PubMed 34773460
4. CTT/Baigent 2010, Lancet — statin LDL-lowering meta-analysis · PubMed 21067804
5. Kamstrup et al. 2009, JAMA — genetic Lp(a) & MI (MR) · PubMed 19509380
6. Ettehad 2016, Lancet — BP lowering per 10 mmHg SBP (613,815 participants) · PubMed 26724178
7. BPLTTC 2021, Lancet — BP lowering per 5 mmHg, IPD meta (344,716) · PubMed 33933205
8. SPRINT 2015, NEJM — intensive BP <120 cuts events + mortality (HR 0.75/0.73) · PubMed 26551272
9. Detrano 2008 MESA, NEJM — CAC predicts coronary events across ethnicities · PubMed 18367736
10. Polonsky 2010 MESA, JAMA — CAC reclassification NRI 0.25 · PubMed 20424251
11. Blaha 2009, JACC Img — CAC=0 ~1% 10-yr mortality (but smokers/diabetics retain risk) · PubMed 19520338
12. ABI Collaboration/Fowkes 2008, JAMA — low ABI → mortality (n=48,294) · PubMed 18612117
13. Lorenz 2007, Circulation — baseline CIMT predicts MI/stroke · PubMed 17242284
14. Lorenz 2012 PROG-IMT, Lancet — CIMT progression does NOT predict events (refuted surrogate) · PubMed 22541275
15. Costanzo 2010, JACC — drug-induced CIMT regression doesn't track outcomes (refuted surrogate) · PubMed 21126642
16. Mandsager et al. 2018, JAMA Netw Open — CRF & mortality · PubMed 30646252
17. Kodama et al. 2009, JAMA — CRF & mortality meta-analysis · PubMed 19454641
18. Howden/Levine et al. 2018, Circulation — exercise reverses cardiac aging (RCT) · PubMed 29311053
19. Baumgartner 1998, Am J Epidemiol — original ALMI sarcopenia cutpoint · PubMed 9554417
20. Cawthon 2014 FNIH, J Gerontol A — weakness-anchored ALM cutpoints · PubMed 24737559
21. Cruz-Jentoft 2019 EWGSOP2, Age Ageing — strength-first sarcopenia consensus · PubMed 31081853
22. Newman 2006 Health ABC, J Gerontol A — strength (not mass) predicts mortality · PubMed 16456196
23. Sim 2019, JAMDA — ALMI not associated with mortality once function considered · PubMed 30527277
24. Leong et al. 2015, Lancet (PURE) — grip strength & mortality · PubMed 25982160
25. Marshall 1996, BMJ — BMD predicts fracture (RR ~1.5/SD; ~2.3–2.6 site-matched) · PubMed 8634613
26. Haentjens 2010, Ann Intern Med — hip fracture → large excess mortality · PubMed 20231569
27. Cummings 2009 FREEDOM, NEJM — denosumab: vertebral −68%, hip −40% · PubMed 19671655
28. Black 1996 FIT, Lancet — alendronate: vertebral −47–55%, hip −51% · PubMed 8950879
29. Howe 2011, Cochrane — exercise raises BMD but fracture benefit not significant · PubMed 21735380
30. Harris WS, Tintle NL et al. (FORCE Consortium) — 17-cohort omega-3 index + ACM meta (Nature Commun. 2021; n=42,466; HR ~0.87 = 13–17% lower ACM) · PubMed 33888689
31. McBurney MI, Tintle NL, Vasan RS, Sala-Vila A, Harris WS (senior) — Framingham omega-3 index + longevity (AJCN 2021; n=2,240; ~4.74-yr life… · PubMed 34134132
32. Harris & von Schacky — omega-3 index / SCD · PubMed 17324586
33. Gencer 2021 (Circulation) — omega-3 & AFib meta
34. Lu et al. 2019, Aging — DNAm GrimAge · PubMed 30669119
35. Levine et al. 2018, Aging — DNAm PhenoAge · PubMed 29676998
36. Belsky et al. 2022, eLife — DunedinPACE · PubMed 35029144
37. Horvath 2013, Genome Biol — pan-tissue DNAm clock · PubMed 24138928
38. Higgins-Chen et al. 2022, Nat Aging — clock reliability (PC-clocks) · PubMed 36277076