Which Biomarker Tests Are Worth Running

Oak
biomarkers blood-tests apob longitudinal lab-data
Night marsh full of clear seed pods on stalks, only about sixteen glowing amber, their older glowing husks trailing in the water.

The short answer: about fifteen to twenty blood markers carry most of the decision-relevant signal for a healthy adult, and the “64 biomarkers” and “100+ biomarkers” panels sold direct-to-consumer are mostly those fifteen plus a long tail of analytes with wide biological variation, poor reference intervals, or no action attached. The markers worth paying for are apolipoprotein B, lipoprotein(a) once in a lifetime, a full lipid panel, HbA1c with fasting insulin and glucose, ALT/AST/GGT, hs-CRP, creatinine and cystatin C with an eGFR from both, a CBC with differential, ferritin plus transferrin saturation, TSH, 25-hydroxyvitamin D, uric acid, homocysteine, and albumin. Everything else on a big panel is either a repeat of one of those in a less useful unit, or a number you will not act on.

The harder answer is that a single draw of any panel tells you much less than you think, because the variance you measure is analytical variance plus within-person biological variance plus the actual change you care about. Panel breadth does not fix that. Repeat measurement does.

Why panel size is the wrong axis

Vendors compete on marker count because count is easy to advertise. The useful question is: for each marker, what is the probability that this result changes what I do in the next twelve months?

For apoB, high. ApoB counts atherogenic particles directly, one apoB per LDL, IDL, VLDL, and Lp(a) particle. Discordance between apoB and LDL-C is common in people with high triglycerides or metabolic syndrome, so an LDL-C of 100 mg/dL can sit on top of an apoB anywhere from roughly 70 to 110 mg/dL. Non-HDL-C tracks apoB better than LDL-C does but still misses particle-size effects. Diet composition moves this: serum omega-6 fatty acid levels, particularly linoleic acid, associate inversely with apoB and other atherogenic lipoprotein measures.1 ApoB is a standardized immunoturbidimetric assay, runs about $15-30, and does not require fasting.

For, say, a broad “wellness” panel’s serum copper or a random T3, low. You will get a flag, you will look it up, and nothing will follow.

Ask for the CV. Every analyte has an analytical coefficient of variation (CV_A) from the instrument and a within-subject biological CV (CV_I). A study of immunological blood biomarkers in healthy individuals quantified this variation and derived quality goals for the assays on that basis.2 If CV_I is 30% and your two results differ by 20%, nothing happened. The reference change value (RCV) is roughly 2.77 × sqrt(CV_A² + CV_I²) for a two-sided 95% call. For hs-CRP, where CV_I commonly runs above 40%, the RCV easily exceeds 100%, which is why a single hs-CRP of 3.1 mg/L means little and why the standard advice is to take the median of two or three draws several weeks apart, excluding any draw within two weeks of an infection or hard training block.

The list, with what each one is for

Cardiovascular. ApoB is the primary particle measure. Lp(a) is genetically set, measure it once in mass units (mg/dL) or preferably nmol/L, and never again unless you start something known to move it. Standard lipid panel for triglycerides and HDL-C, which feed the triglyceride-to-HDL ratio as a crude insulin-resistance proxy. Ask for apoB in mg/dL and record the assay method, because switching labs mid-series introduces a step change you will mistake for a trend.

Glycemic. HbA1c reflects roughly 90 days of average glucose but is distorted by red-cell lifespan, so read it alongside your CBC. If MCV is high or reticulocytes are elevated, HbA1c underestimates. Fasting insulin plus fasting glucose gives you HOMA-IR. Fasting insulin assays are poorly standardized between labs, so hold the lab constant. Continuous glucose data, if you have it, dominates all of these for detecting postprandial excursions that a fasting draw cannot see.

Liver and kidney. ALT is the more specific of the transaminases for hepatocyte content. GGT adds signal on alcohol and on oxidative stress. For kidney, run creatinine and cystatin C together: creatinine is muscle-mass dependent and will read falsely low in a small, sedentary person and falsely high in someone lifting heavily, while cystatin C is not. A large gap between the two eGFRs is itself informative.

Inflammation and iron. hs-CRP, interpreted as a median of repeats. Ferritin is an acute-phase reactant, so read it with CRP and with transferrin saturation. Ferritin of 300 ng/mL with CRP of 6 mg/L is an inflammation signal, not an iron-loading signal.

Thyroid, vitamin D, uric acid, homocysteine, albumin. TSH first, free T4 and free T3 only if TSH is out of range. 25-OH-D once or twice a year with a seasonal note. Uric acid tracks fructose and alcohol intake and renal handling. Homocysteine is sensitive to B12, folate, and B6 status and to MTHFR genotype.

Interpretation of any out-of-range result belongs with a clinician. We are describing what to measure, not what to do about it.

Where multi-marker panels genuinely add something

Panels beat single markers when the markers are combined into a model rather than read individually. This is the systems-biology argument: disease signatures live in patterns across proteins, not in one analyte crossing a threshold.3 Colorectal cancer work is a concrete example, with serum protein panels and multi-gene blood signatures showing detection performance no single marker reaches.45

The Alzheimer’s blood biomarker field is the cleanest current illustration of what a good test looks like and what it costs to get there. Plasma p-tau217 and Aβ42/40 assays now perform well enough that the Global CEO Initiative published explicit performance thresholds for clinical use: a triage test should achieve roughly 90% sensitivity with 85% specificity against amyloid PET or CSF, and a confirmatory test around 90% on both.6 Those numbers came from a decade of assay standardization and cohort validation.7 Even so, the population-level use case is unsettled, because positive predictive value falls sharply in asymptomatic, low-prevalence groups, and a positive amyloid result in a person without symptoms does not carry a clear next step.8 If you order one of these, do it through a clinician who will discuss what a result means before you see the number.

How to run it

Get the raw data, not the dashboard. Most consumer platforms hand you a PDF with colored bars. Request the HL7 or the lab’s CSV export, or at minimum transcribe into a single file with columns: date, analyte, value, unit, lab, assay_method, fasting_hours, ref_low, ref_high. Units bite you: Lp(a) in mg/dL versus nmol/L is roughly a 2.0-2.4x factor and is not a fixed conversion, insulin in µIU/mL versus pmol/L is 6.0x, glucose mg/dL to mmol/L is 0.0555.

Protocol that removes most of the noise:

  • Same lab, same assay, every time.
  • Draw between 7 and 9 a.m., 12-14 hours fasted, water allowed.
  • No alcohol for 72 hours, no hard training for 48 hours before a draw (ALT, AST, CK, CRP, and ferritin all move).
  • Sit quietly for 5 minutes before venipuncture, and avoid prolonged tourniquet time, which concentrates protein-bound analytes by several percent.
  • Baseline is two draws two to four weeks apart, not one.
  • Then quarterly, or six weeks after any deliberate change.

For analysis, keep it simple. A long-format CSV in pandas, a per-analyte plot with the reference interval shaded, and an RCV band around each prior value. Flag a change only when it exceeds the RCV. Use published CV_I values where you have them, and estimate your own once you have five or more draws of a stable marker. Sleep quality is a covariate worth logging alongside, since sleep fragmentation has been linked to arterial stiffness measures.9

Questions people also ask

What is Peter Attia’s recommended ApoB level? He has publicly argued for targets well below standard population reference ranges, in the vicinity of 60 mg/dL or lower for people optimizing long-term cardiovascular risk, which is more aggressive than most guidelines. Whether that target fits you depends on your overall risk profile and is a conversation for a lipid-literate physician.

Are biomarker tests worth it? The fifteen-to-twenty marker core is worth it if you repeat it on a schedule and hold the lab constant. A one-time 100-marker panel is mostly worth it as a baseline, and its main risk is incidental flags that lead to follow-up testing with low pre-test probability.

What is the best at-home biomarker test? At-home fingerstick collection is fine for HbA1c, lipids, and hs-CRP, and poor for anything requiring proper serum separation or cold chain. If you want apoB, insulin, and cystatin C in the same series, use a venous draw at a patient service center. Dried blood spot results are not interchangeable with venous results, so do not mix them in one time series.

What biomarkers come with 64-marker panels? Typically a CMP (14), CBC with differential (about 15), lipid panel (4-6), thyroid (2-4), iron studies (4), several vitamins, hs-CRP, HbA1c, insulin, and sex hormones. Count the actual distinct decisions, not the analytes.

Oak builds longitudinal molecular profiles of individuals: whole-genome sequencing, RNA sequencing, proteomics, blood biomarkers, and continuous glucose data, integrated into one model of you. Build your profile.

Footnotes

  1. Kevin Maki, Carol Kirkpatrick, Meredith Wilcox, et al. Associations of serum omega-6 polyunsaturated fatty acids with apolipoprotein B and atherogenic lipoprotein lipids. Journal of Clinical Lipidology, 2025. https://doi.org/10.1016/j.jacl.2025.04.086 ↩

  2. Najib Aziz, Roger Detels, Joshua J. Quint, et al. Biological variation of immunological blood biomarkers in healthy individuals and quality goals for biomarker tests. BMC Immunology, 2019. https://doi.org/10.1186/s12865-019-0313-0 ↩

  3. Bradley J. Smith, Licia C. Silva-Costa, Daniel Martins-de-Souza. Human disease biomarker panels through systems biology. Biophysical Reviews, 2021. https://doi.org/10.1007/s12551-021-00849-y ↩

  4. Mark Han, Choong Tsek Liew, Hong Wei Zhang, et al. Novel Blood-Based, Five-Gene Biomarker Set for the Detection of Colorectal Cancer. Clinical Cancer Research, 2008. https://doi.org/10.1158/1078-0432.ccr-07-1801 ↩

  5. Melanie M. Ivancic, Bryant W. Megna, Yuriy Sverchkov, et al. Noninvasive Detection of Colorectal Carcinomas Using Serum Protein Biomarkers. Journal of Surgical Research, 2020. https://doi.org/10.1016/j.jss.2019.08.004 ↩

  6. Suzanne E. Schindler, Douglas Galasko, Ana C. Pereira, et al. Acceptable performance of blood biomarker tests of amyloid pathology — recommendations from the Global CEO Initiative on Alzheimer’s Disease. Nature Reviews Neurology, 2024. https://doi.org/10.1038/s41582-024-00977-5 ↩

  7. Harald Hampel, Sid E. O’Bryant, José L. Molinuevo, et al. Blood-based biomarkers for Alzheimer disease: mapping the road to the clinic. Nature Reviews Neurology, 2018. https://doi.org/10.1038/s41582-018-0079-7 ↩

  8. Michelle M. Mielke, Nicole R. Fowler. Alzheimer disease blood biomarkers: considerations for population-level use. Nature Reviews Neurology, 2024. https://doi.org/10.1038/s41582-024-00989-1 ↩

  9. R. Vallat, V. Shah, S. Redline, et al. Broken sleep predicts hardened blood vessels. Sleep Medicine, 2019. https://doi.org/10.1016/j.sleep.2019.11.1103 ↩