What a Biomarker Testing Service Can and Cannot Tell You

Oak
biomarkers blood-testing proteomics measurement-error longitudinal-data
A long rail of identical serum vials inside a dark lab instrument, each topped by a blurred halo of blue light forming a drifting band.

A biomarker testing service is worth paying for if you intend to measure the same analytes repeatedly and keep the numbers. It is mostly wasted money if you want a single panel of 75 markers to tell you something. The reason is arithmetic rather than philosophy. For most blood analytes, the biological variation within one person between draws is large compared to the distance between your value and the reference interval edge, so a single measurement carries wide uncertainty about your actual set point. A service that returns structured and exportable results gives you a time series you can reason about. That depends on a fixed method at a fixed lab on a schedule you control. A service that returns a PDF of colored bars gives you one noisy sample and a prompt to buy a supplement.

What follows is how we would choose a service, what we would measure, and how to handle the data so that the second draw is comparable to the first.

What “biomarker” means in practice, and why the category is so uneven

Before comparing services it helps to be precise about what the word covers, because the category spans wildly different levels of maturity. A biomarker is any measurable characteristic used as an indicator of a biological state. That definition covers analytes with fifty years of standardization behind them and analytes that exist only as a research assay in one lab, and the difference matters enormously to how you should read a result. The formal distinction in the literature is between a biomarker, which is a measurand, and a medical test, which is a measurand plus a method, a population, a decision threshold, and evidence that acting on the threshold changes outcomes. Most consumer panels sell you the first and imply the second.1

The gap between the two is where consumer biomarker testing goes wrong. Getting a new analyte into routine clinical chemistry requires reference materials and method comparison studies across platforms. It also requires defined analytical goals derived from biological variation and external quality assessment schemes. Analytes that skip those steps produce results that are not comparable between laboratories or over time.2 When a service reports something like “inflammation score” or a proprietary composite, you generally cannot tell which platform produced it. You also cannot tell what the imprecision is or whether next year’s version of the score is on the same scale.

The Alzheimer’s blood biomarker field is the best-documented example of a measurand maturing into a test in public, and it shows what the finished product looks like. Plasma phosphorylated tau 217 and related markers now discriminate amyloid pathology well enough to be useful in primary care, with reported accuracy substantially exceeding that of clinicians using standard workups.3 Expert groups have gone further and proposed explicit performance floors for such tests. The proposed floors are on the order of 90 percent sensitivity and 85 percent specificity for a triage test used to decide who gets a confirmatory PET scan or lumbar puncture.4 That is what a defined test looks like: a stated use, a stated threshold, and stated minimum performance.

At the same time, careful reviewers have pointed out that strong discrimination in memory-clinic cohorts does not automatically transfer to unselected populations with low prevalence. In those populations positive predictive value falls and incidental findings carry real consequences.5 Both things are true, and holding both at once is the right posture toward any biomarker you buy.

What we would order

With that framing in place, here is the panel we would build. The guiding principle is that we favor a small set of well-standardized analytes measured often over a large set measured once, and each item below comes with the reasoning for its inclusion.

  • Lipids with direct LDL-C or apolipoprotein B, plus lipoprotein(a) once. ApoB is a particle count and avoids the Friedewald estimation error that inflates variability in calculated LDL-C at high triglycerides. Lp(a) is largely genetically determined, so a single good measurement is informative and repeats add little.
  • HbA1c plus fasting glucose plus fasting insulin. HbA1c integrates roughly three months of glycemia and has tight assay standardization. Fasting insulin is poorly standardized across platforms, so keep it at one lab.
  • High-sensitivity CRP, with the understanding that a single value is nearly uninterpretable. Within-person variation is large and any recent infection or injury dominates. Take the median of three draws over months.
  • A broad chemistry and hematology base: complete blood count with differential, comprehensive metabolic panel, liver enzymes, creatinine and cystatin C for two independent estimates of kidney filtration, ferritin with transferrin saturation, TSH with free T4, and 25-hydroxyvitamin D.
  • Urine albumin-to-creatinine ratio, which is cheap and often omitted from consumer panels.

Everything else we treat as exploratory. Broad proteomic panels such as Olink and SomaScan, along with metabolomics, are genuinely informative for research and for tracking trajectories. We use them. Their units are relative, their cross-platform agreement is imperfect, and there is no reference interval you should act on. Read them as a longitudinal shape rather than as a value versus a cutoff.

Analytical and biological variation, concretely

Interpreting any of these numbers means understanding the two error terms that sit between you and your set point. Analytical imprecision is the assay’s own coefficient of variation, typically 2 to 4 percent for HbA1c and lipids on modern platforms and considerably worse for immunoassays of hormones and cytokines. Within-subject biological variation is the day-to-day wobble of your own physiology: small for HbA1c, moderate for cholesterol, and large for triglycerides, CRP, cortisol, and testosterone.

The practical consequence is the reference change value, which is the difference between two measurements that exceeds what error alone would produce. Because it combines both error terms, a 12 percent change in your LDL-C between draws may be noise while a 12 percent change in HbA1c is not. If your service does not publish the assay CV for each analyte, ask. If they will not say, that is itself information about the service.

You can also reduce the error you contribute yourself by controlling the conditions of the draw. Draw fasting 10 to 14 hours, at the same time of morning, and at the same lab on the same platform. Sit for five minutes beforehand and stay off strenuous exercise for 48 hours. Tourniquet time and posture alone move potassium and some proteins measurably. When you change labs you should expect a step change in some analytes, and you should not read it as a change in you.

Getting the data out in a form you can use

Careful measurement is only half the work; the results also have to survive in a form you can analyze years later. Insist on structured export before you pay. The useful formats are a CSV or JSON per draw containing, at minimum, LOINC code (the standard vocabulary that gives each lab test a unique identifier), analyte name, value, units, reference interval, method or platform, and collection timestamp. A FHIR Observation bundle, drawn from the standard healthcare data exchange format, is better still. LOINC coding makes multi-lab merges tractable. A PDF is a dead end. OCR on lab PDFs fails in exactly the ways that matter: units dropped, footnoted flags lost, and ranges misattributed to the wrong row.

We would keep a single long-format table with one row per analyte per draw, a column for lab and method, and a column for pre-analytical conditions. In practice a schema like (draw_id, collected_at, loinc, analyte, value, unit, ref_low, ref_high, lab, platform, fasting_hours, notes) is enough, stored as Parquet or in SQLite, versioned in git alongside the derivation scripts. Normalize units on ingest, because the same analyte will arrive in mg/dL from one lab and mmol/L from another. Ferritin will arrive as ng/mL and µg/L. Those two are numerically identical and still worth recording explicitly. Then fit the simplest thing that respects the noise: a loess or linear trend per analyte with the reference change value drawn as a band, rather than point-to-point comparisons.

If you hand the table to an AI agent for interpretation, give it the method and reference interval columns too. Without them the model will compare values across labs as if they were on one scale, which is the single most common analysis error we see.

Where a clinician is not optional

Data hygiene gets you a clean trajectory, but some findings need a physician regardless of how well you have organized them. Biomarker results become medical decisions at the point where a threshold triggers an action, and that point belongs to a clinician. Some findings need a physician who can order confirmatory testing and interpret prevalence-adjusted predictive values. That includes anything suggesting kidney impairment or a hematologic abnormality. It also includes a markedly elevated ferritin or calcium, a thyroid value well outside range, or a positive result on a cancer or neurodegeneration screening assay.

This applies with particular force to cognitive biomarkers, where implementation guidance stresses that plasma assays should be used within a defined diagnostic pathway rather than as standalone answers.6 Health systems are building those pathways because the triage step changes cost and wait times for confirmatory imaging, which is a system-level calculation rather than a self-directed one.7

Cancer biomarker testing is a separate category from wellness panels and should be treated as such. Tumor molecular profiling to select therapy, and blood-based screening for colorectal or other cancers, sit inside clinical pathways whose modeled cost-effectiveness depends heavily on adherence and follow-up colonoscopy rates.8 Those tests are ordered by oncologists and gastroenterologists for defined indications. Do not substitute a consumer panel for them.

Questions people also ask

Is biomarker testing worth it? Yes, if you will repeat it and keep the raw structured results, because the value is in the trajectory and in having a baseline recorded before anything changes. A one-time 75-marker panel with no export and no follow-up is low value, mostly because single measurements of high-variation analytes cannot distinguish your set point from a bad morning.

How much does a biomarker test cost? A solid standardized blood panel of the kind described above runs roughly $150 to $600 out of pocket. The price depends on the lab and on whether apoB, Lp(a), and cystatin C are included. Consumer subscription services typically charge $400 to $1,000 per year for broader panels. Research-grade proteomics and metabolomics are an order of magnitude more, and specialized assays such as plasma p-tau217 are priced separately and often available only through a clinician.

How reliable are biomarker tests? Reliability is a property of each analyte rather than of the service as a whole. Well-standardized analytes such as HbA1c, creatinine, and lipids are highly reproducible. Immunoassay-based hormones, cytokines, and novel proteomic markers vary meaningfully between platforms and lots. That variation is precisely why routine clinical laboratories require reference materials and external quality assessment before adopting a new marker.2 Even for markers with strong published discrimination, analytical variability and pre-analytical handling remain live implementation problems.6

How do I get my biomarkers tested? You have two routes. A clinician can order panels with an indication and usually get them billed to insurance, or a direct-to-consumer lab service can draw at a patient service center. We would ask two questions before booking: which platform runs each assay, and whether you can export structured per-analyte results with LOINC codes.

Does insurance pay for biomarker testing? Generally yes when a clinician orders a test for a documented indication. Generally no for self-directed screening panels, wellness subscriptions, or research-grade multi-omics. Coverage for newer assays lags their evidence base considerably, and coverage decisions for blood-based neurodegeneration markers are still being written as the evidence for population-level use accumulates.5

Is biomarker testing a good cancer treatment option? Biomarker testing is not a treatment. Tumor molecular profiling can identify which therapies are plausible for a specific cancer, and that testing is ordered and interpreted by an oncologist as part of treatment planning.

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. Andrea R. Horvath, Sarah J. Lord, Andrew StJohn, et al. From biomarkers to medical tests: The changing landscape of test evaluation. Clinica Chimica Acta, 2014. https://doi.org/10.1016/j.cca.2013.09.018 ↩

  2. Catharine Sturgeon, Robert Hill, Glen L. Hortin, et al. Taking a new biomarker into routine use – A perspective from the routine clinical biochemistry laboratory. PROTEOMICS – Clinical Applications, 2010. https://doi.org/10.1002/prca.201000073 ↩ ↩2

  3. Sebastian Palmqvist, Pontus Tideman, Niklas Mattsson-Carlgren, et al. Blood Biomarkers to Detect Alzheimer Disease in Primary Care and Secondary Care. JAMA, 2024. https://doi.org/10.1001/jama.2024.13855 ↩

  4. 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 ↩

  5. Willem A Van Gool, J A Floria Siebrand, Carol Brayne, et al. Evidence gap in blood biomarkers for Alzheimer’s disease. BMJ, 2025. https://doi.org/10.1136/bmj-2025-084781 ↩ ↩2

  6. Michael Schöll, Inge M W Verberk, Marta del Campo, et al. Challenges in the practical implementation of blood biomarkers for Alzheimer’s disease. The Lancet Healthy Longevity, 2024. https://doi.org/10.1016/j.lanhl.2024.07.013 ↩ ↩2

  7. Soeren Mattke, Sang Kyu Cho, Tobias Bittner, et al. Blood‐based biomarkers for Alzheimer’s pathology and the diagnostic process for a disease‐modifying treatment: Projecting the impact on the cost and wait times. Alzheimer’s & Dementia: Diagnosis, Assessment & Disease Monitoring, 2020. https://doi.org/10.1002/dad2.12081 ↩

  8. John M. Inadomi. Cost-Effectiveness of Blood-Based Biomarkers for Colorectal Cancer Screening—An Ounce of Prevention Is Worth a Pound of Cure. JAMA Network Open, 2023. https://doi.org/10.1001/jamanetworkopen.2023.43346 ↩