What Is the Huberman Blood Test, and What Should You Order?

There is no such thing as a “Huberman blood test.” Andrew Huberman is a neuroscientist at Stanford who runs a podcast and has had a sponsorship relationship with Function Health. Function Health is a consumer lab-testing company that resells a large panel of roughly 100-160 assays, depending on the tier, drawn at Quest patient service centers. The panel people associate with the podcast is a fairly conventional preventive-cardiology and endocrine workup with a long tail of extras attached. It covers lipids with ApoB and Lp(a), fasting glucose and insulin, HbA1c and hs-CRP. It also includes a full thyroid panel, sex hormones, ferritin and iron studies. The rest is liver and kidney chemistries, homocysteine, and a handful of vitamins. None of it is proprietary. You can order the same assays through any lab that will accept a requisition, and in many states you can order them yourself without going through a physician.

The more useful question is which of those numbers change your understanding of your own physiology, which are noise, and how to store the results so that the fifth draw is worth more than the first. This article works through all three.

The subset that carries most of the signal

Large panels spread your attention across many analytes of unequal value. If you were ordering ten things instead of 160, these are the ones we would choose, with a note on why each earns its place and what to watch for in its interpretation.

• ApoB. Every atherogenic lipoprotein particle carries exactly one ApoB molecule, so the measurement is a direct particle count. LDL-C, by contrast, estimates the mass of cholesterol carried inside those particles. That is a proxy for particle count and a poor one in people with high triglycerides or metabolic syndrome. Ask for the immunoturbidimetric assay, which is standard at Quest and Labcorp under CPT 82172. Get Lp(a) once in your life, reported in molar units (nmol/L) rather than mass units, since it is roughly 90% genetically determined and does not need repeating.
• Fasting insulin plus fasting glucose. Glucose on its own is a late signal, because insulin starts rising years before fasting glucose moves. Compute HOMA-IR = (glucose mg/dL × insulin µIU/mL) / 405 yourself, and record which insulin assay was used. Insulin immunoassays are not harmonized across platforms, so absolute values are not comparable between labs.
• HbA1c, keeping in mind that it reflects both average glucose and red cell lifespan. Anemia, hemoglobin variants, or a high reticulocyte count will shift it. If your A1c and your continuous glucose monitor disagree, the monitor is usually closer to the truth about your glucose, and the A1c is telling you something about your red cells instead.
• hs-CRP. Use the high-sensitivity assay or skip the test entirely. Discard any value above about 10 mg/L as an acute-phase response to something transient, and redraw in two weeks.
• Ferritin with transferrin saturation. Ferritin on its own is uninterpretable because it behaves as an acute-phase reactant. Pair it with iron, TIBC, and CRP.
• ALT and GGT. Both are inexpensive, and together they say more about hepatic fat and alcohol exposure than either does alone.
• Creatinine and cystatin C. Creatinine-based eGFR is confounded by muscle mass, which matters if you lift. Cystatin C is unaffected by it. Order both and look at the gap between them.
• TSH with free T4. In people without thyroid disease, free T3 and reverse T3 generate more confusion than insight.

The long tail on consumer panels mostly produces false positives. That tail means most vitamins, most hormones in people with no symptoms, and tumor markers. Order 160 roughly independent tests with 95% reference intervals and you should expect about eight flagged values in a perfectly healthy person, purely from the arithmetic of reference ranges. This is the central failure mode of large panels: the flags are the expected output of the test design rather than findings about you.

Where the cited literature lands

Some of the inflammatory and immune markers that appear on extended panels do have real associations in the published literature. Knowing the size and type of those associations is what keeps you from over-reading a single number on your own report.

IL-6, for instance, covaries inversely with cognitive performance in middle-aged community volunteers. That is an epidemiological association across a population rather than a readout you can apply to your own draw.¹ Immune aberrations are measurable in Alzheimer’s patients relative to controls, but separation between groups does not make any of these markers a screening test for an individual.² Monocyte phenotyping has been proposed as a window on neuroinflammation, on the grounds that circulating monocytes share a lineage with microglia. That proposal remains at the research stage.³ Adipokines such as leptin and adiponectin are mechanistically tied to obesity biology and are plausible therapeutic targets. That is a different claim from being useful for tracking yourself month to month.⁴ Plasma gelsolin falls in acute inflammation and sepsis and has been studied as a diagnostic and therapeutic target, again in a clinical rather than a wellness context.⁵

One further point is worth holding in mind when you read your own results: fasting glucose responds to psychological stress. In a cohort of US Chinese immigrants, increases in perceived stress were associated with higher fasting glucose.⁶ A bad week before your draw will show up in your numbers.

How to get the data, and in what format

If the goal is a series of measurements that becomes more informative over time, then optimize for data you own in a machine-readable form. It should be drawn under controlled conditions and repeated. That breaks down into five practical decisions.

Ordering. You have three broad routes, in rough order of cost. Direct-to-consumer requisition services bill you for Quest or Labcorp assays at close to cost. Consumer panel companies such as Function or InsideTracker add interpretation and a subscription on top of the same underlying assays. A physician’s order through insurance is usually cheapest when the tests are indicated, and free when they are covered. We would use a direct requisition service and pick the assays individually.

Standardizing the draw. Variation you control swamps the variation you care about, so hold the conditions fixed. Fast for 12-14 hours, water only. Draw at the same time of morning every time, ideally before 9am. Avoid hard training in the prior 48 hours, since it moves ALT, AST, CK, and CRP, and avoid alcohol for 72 hours, which affects GGT and triglycerides. Sit for five minutes before the venipuncture and keep tourniquet time short, because prolonged tourniquet time concentrates plasma proteins and can shift lipids by several percent. If you are tracking sex hormones, fix the point in your cycle as well.

Getting it out. Most portals export PDFs, which are documents rather than data. Quest and Labcorp both support access through patient APIs using HL7 FHIR, a standard format for exchanging health records, and many consumer panels will export CSV. If you are stuck with PDFs, parse them once and never again:

pdftotext -layout results_2026_03.pdf - \
  | rg -o '^(ApoB|Lp\(a\)|hs-CRP|Insulin|Glucose|HbA1c)\s+([\d.]+)\s+(\S+)' \
        -r '$1,$2,$3'

Store one tidy row per (date, analyte, value, unit, lab, assay_method, LOINC). The LOINC code, which is the standard identifier for a specific lab measurement, and the identity of the performing lab both matter more than people expect. Switching labs mid-series can introduce a step change larger than any real biological trend, particularly for insulin, testosterone, vitamin D, and any ELISA-based assay. If you must switch, run one draw split across both labs so you can measure the offset.

Deciding whether a change is real. Every analyte carries an analytical CV, meaning assay imprecision. It also carries a within-subject biological CV, meaning your own day-to-day variation. The reference change value that combines them is roughly RCV = 2.77 × sqrt(CV_a² + CV_i²). For hs-CRP, where within-subject CV is large, two values can differ by well over 50% with nothing having changed. For albumin or sodium, a 5% move is substantial. Look up both CVs for your analytes before interpreting any delta, and prefer a regression across four or more points to a comparison between two draws.

Where a clinician is required. Several situations call for a physician rather than a spreadsheet. Those include anything flagged outside the reference range on a repeat draw and any Lp(a) above the commonly used 125 nmol/L threshold. They also include ferritin above roughly 400 ng/mL alongside high transferrin saturation, a persistent unexplained abnormal CBC, or anything you would be tempted to act on. Interpreting an individual result in the context of your history is the practice of medicine. Our work is measurement rather than diagnosis, and neither a podcast nor an AI agent replaces the person who can order a confirmatory test and examine you.

What blood alone cannot tell you

Even a well-run blood series has hard limits, and it helps to be clear about where they fall. Serum chemistry is a low-dimensional snapshot of a high-dimensional system. It says almost nothing about which variants you carry, what your tissues are transcribing, or how your glucose behaves across a day.

Other layers fill in those gaps. Whole-genome sequencing gives you Lp(a) risk alleles, ApoE status, and hemochromatosis variants. It also gives you pharmacogenes once and permanently. RNA sequencing of whole blood gives you expression state, which is dynamic in ways serum proteins are not. Continuous glucose data gives you thousands of points per week against the four numbers a fasting panel provides.

The published work points in the same direction. Machine learning models built on longitudinal clinical and biomarker data for heart failure prevention consistently outperform single-timepoint risk scores, which is the same lesson at a different scale: density and repetition beat one comprehensive snapshot.⁷ The pattern recurs in cardiac aging research, where model-derived age estimates track molecular senescence markers better than chronological age does.⁸

Questions people also ask

Which blood test company is Andrew Huberman affiliated with? Function Health has been a Huberman Lab podcast sponsor. The relationship is advertising rather than a scientific endorsement of any specific panel design.

Is Function Health a legitimate health provider? It is a legitimate ordering and presentation layer. The assays run at Quest, a CLIA-certified reference lab, so the numbers are real. What you pay for beyond the cost of the assays is the interface, the physician sign-off on the requisition, and the summaries.

How credible is Huberman Lab? Huberman is a working neuroscientist whose primary literature is real, including work in Nature on partial epigenetic reprogramming restoring vision in mice.⁹ The podcast is a different product: long-form summaries of other people’s research, sometimes extrapolated well beyond what the underlying studies support. Treat episode claims as pointers to papers, then read the papers.

Is there a blood test that can check for everything? No. Blood proteins and metabolites are a narrow projection of your biology. Large panels increase false positives faster than they increase findings, and by construction they miss entire categories, including structural disease, most cancers at early stage, and genotype.

What are the most important blood tests for longevity? For cardiovascular risk specifically, ApoB and Lp(a) are the two with the clearest mechanistic and genetic support. For metabolic health, fasting insulin combined with a CGM. Everything else is secondary.

Woolf Software 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. Anna L. Marsland, Karen L. Petersen, Rama Sathanoori, et al. Interleukin-6 Covaries Inversely With Cognitive Performance Among Middle-Aged Community Volunteers. Psychosomatic Medicine, 2006. https://doi.org/10.1097/01.psy.0000238451.22174.92 ↩

2. Konstantinos Bonotis, Eleni Krikki, Vasiliki Holeva, et al. Systemic immune aberrations in Alzheimer’s disease patients. Journal of Neuroimmunology, 2008. https://doi.org/10.1016/j.jneuroim.2007.10.020 ↩

3. Gerd Schmitz, Kerstin Leuthauser-Jaschinski, Evelyn Orso. Are Circulating Monocytes as Microglia Orthologues Appropriate Biomarker Targets for Neuronal Diseases? (Supplementry Table). Central Nervous System Agents in Medicinal Chemistry, 2009. https://doi.org/10.2174/187152409789630424 ↩

4. Marleen Würfel, Matthias Blüher, Michael Stumvoll, et al. Adipokines as Clinically Relevant Therapeutic Targets in Obesity. Biomedicines, 2023. https://doi.org/10.3390/biomedicines11051427 ↩

5. Ewelina Piktel, Ilya Levental, Bonita Durnaś, et al. Plasma Gelsolin: Indicator of Inflammation and Its Potential as a Diagnostic Tool and Therapeutic Target. International Journal of Molecular Sciences, 2018. https://doi.org/10.3390/ijms19092516 ↩

6. Carolyn Y Fang, Ajay Rao, Elizabeth A Handorf, et al. Increases in Psychological Stress Are Associated With Higher Fasting Glucose in US Chinese Immigrants. Annals of Behavioral Medicine, 2024. https://doi.org/10.1093/abm/kaae056 ↩

7. Arsalan Hamid, Matthew W. Segar, Biykem Bozkurt, et al. Machine learning in the prevention of heart failure. Heart Failure Reviews, 2024. https://doi.org/10.1007/s10741-024-10448-0 ↩

8. Dhivya Vadhana Meenakshi-Siddharthan, Christopher Livia, Timothy E. Peterson, et al. Artificial Intelligence–Derived Electrocardiogram Assessment of Cardiac Age and Molecular Markers of Senescence in Heart Failure. Mayo Clinic Proceedings, 2023. https://doi.org/10.1016/j.mayocp.2022.10.026 ↩

9. Andrew D. Huberman. Sight restored by turning back the epigenetic clock. Nature, 2020. https://doi.org/10.1038/d41586-020-03119-1 ↩