What an Ultimate Blood Test Should Measure

Oak
biomarkers blood-testing preanalytics longitudinal self-quantification
A studio-lit iridescent wading bird on a stand, its clear throat showing four glowing red fluid chambers in a row.

The best blood test is not the widest panel. It is a small set of analytes chosen because they change decisions, measured three or more times under controlled conditions on the same platform, stored as machine-readable rows with LOINC codes and units. A 56-marker “ultimate performance” panel drawn once, at whatever hour you showed up, on an analyzer you will never use again, produces a PDF full of arrows and almost no signal. The fix is repetition and preanalytic control, plus a few layers that serum chemistry cannot reach: your genome, transcriptome, plasma proteome, and continuous glucose.

Why 56 markers once is worse than 15 markers four times

Reference intervals are central 95% ranges from a reference population. If you measure 56 roughly independent analytes on a healthy person, the probability that at least one falls outside its interval is 1 − 0.95⁵⁶ ≈ 94%. You are nearly guaranteed a red arrow. That arrow is the panel’s product, not its finding.

The deeper problem is biological variation. Each analyte has a within-person coefficient of variation (CV_I) and a between-person one (CV_G), and the ratio determines whether a population reference interval is even meaningful for you. Studies of immunological and hematological markers in healthy adults have quantified these CVs specifically to set analytical quality goals, and for many markers the within-person variation is large enough that a single draw cannot be interpreted against a population range at all.1 For markers with low individuality (CV_I ≪ CV_G), a population range works. For markers with high individuality, only your own prior values work, which means you need a baseline of several draws before any later value is informative.

The practical rule: the reference change value, the difference between two draws that exceeds noise, is roughly 2.77 × √(CV_A² + CV_I²), where CV_A is analytical imprecision. For hs-CRP, where CV_I is large, that means a doubling can be nothing. For albumin, where CV_I is small, a 10% drop is real. Know which analyte you are looking at before you react to a delta.

The panel we would order

Fifteen to twenty analytes, quarterly, one lab, same accession pattern:

  • Lipids as particle counts: ApoB (not just LDL-C) and Lp(a) in nmol/L. Lp(a) is largely set by the LPA locus, so measure it once in your life and never again.
  • Glycemia: HbA1c, fasting glucose, fasting insulin. HbA1c integrates roughly 90 to 120 days of exposure and is biased by anything that shifts red cell lifespan, so read it next to your CGM curve rather than instead of it.
  • Renal: creatinine and cystatin C. Cystatin C is not driven by muscle mass, which matters if you lift.
  • Hepatic: ALT, AST, GGT, albumin, bilirubin.
  • Iron: ferritin, transferrin saturation, and hs-CRP drawn at the same time, because ferritin is an acute-phase reactant and rises with inflammation independent of iron stores.
  • Thyroid: TSH plus free T4. Add free T3 only if TSH is abnormal.
  • Nutrient status measured at the right compartment: 25-OH vitamin D, B12 with methylmalonic acid, and red cell folate rather than serum folate. Serum folate tracks your last few meals. RBC folate reflects the folate incorporated during erythropoiesis, which is why it is the accepted marker of longer-term status.2 Homocysteine sits downstream of both B12 and folate and is worth carrying as an integrator.
  • CBC with differential, and uric acid.

Everything else on the 56-marker menus (most hormone panels in asymptomatic men, broad “vitamin screens”, tumor markers like CA-125 in people without risk) generates follow-up imaging and anxiety at a rate far above its yield. If a result is abnormal, you need a clinician to interpret it in context. We are describing measurement, not care.

Preanalytics decide whether the numbers mean anything

Most variance people attribute to their training block is collection technique.

  • Time of day. Cortisol, testosterone, and iron all have diurnal swings large enough to swamp any intervention. Draw at the same clock time every time, ideally 07:00 to 09:00.
  • Posture. Standing shifts water out of the vascular compartment and concentrates protein-bound analytes by several percent. Sit for 10 minutes before the needle, every time.
  • Tourniquet. Under 60 seconds. Longer causes hemoconcentration and inflates albumin, calcium, and lipids.
  • Hemolysis. Ruptured red cells dump potassium, LDH, and AST into the serum. A hemolyzed potassium of 5.6 mmol/L is a lab error, not hyperkalemia. Ask for the hemolysis index if your lab reports one.
  • Exercise. A hard session raises CK, AST, ALT, and hs-CRP for days. Schedule draws 48 to 72 hours after your last hard effort, or accept that you are measuring recovery.
  • Platform. Ferritin, TSH, and testosterone assays are not harmonized across manufacturers. Switching labs mid-series creates a step change you will misread as biology.

The layers a serum panel cannot reach

Blood chemistry describes current state. Four other layers describe risk, mechanism, and trajectory.

Genome. One 30x whole-genome run, delivered as FASTQ plus an aligned CRAM and a gVCF, is permanent. Align with bwa-mem2 mem -K 100000000 -Y, call with GATK HaplotypeCaller in GVCF mode or DRAGEN, annotate with VEP against ClinVar and gnomAD. Pull pharmacogenes (CYP2C19, CYP2D6, DPYD, SLCO1B1) with a caller like PharmCAT, and compute polygenic scores with plink2 --score. Interpretation of any pathogenic finding belongs with a genetic counselor.

Epigenome. Methylation arrays support clocks that carry information your chemistry panel does not. DNA methylation age of blood predicts all-cause mortality after adjustment for chronological age and standard risk factors.3 DunedinPACE estimates a rate of aging per year rather than an age, which makes it more usable as a repeated within-person measure.4

Proteome. Affinity proteomics (Olink NPX, SomaScan ADAT) puts 1,500 to 7,000 plasma proteins on one draw. Two examples of what that buys: plasma p-tau217 now discriminates Alzheimer’s pathology with accuracy comparable to or better than CSF assays,5 and GFAP has emerged as a readout of astroglial injury across a range of brain and spinal cord conditions.6 These are research and clinical-referral signals, not self-interpretation targets.

Transcriptome. Bulk RNA-seq of whole blood or a sorted fraction gives expression state. Platelet RNA in particular carries diagnostic structure, with published protocols for spliced-RNA classification of blood-based disease signals.7 For a single person, run salmon quant or STAR + featureCounts, then score each timepoint as a within-person z against your own prior samples. Cross-sectional differential expression against a cohort is the wrong frame when n = 1.

Composite indices beat single markers

Physiological dysregulation shows up across systems before it shows up in any one analyte. Multisystem summary scores built from cardiovascular, metabolic, inflammatory, and neuroendocrine markers have been modeled explicitly, and the multi-system formulation carries information the individual components do not.8 If you have four quarterly draws, build your own: z-score each analyte against your own baseline, group by system, and track the system means. That is a two-hour script in R or pandas and it is more useful than any vendor’s composite “wellness score”.

Store it so an agent can use it

Ask for results as HL7 FHIR Observation resources or a CSV with one row per analyte per draw: loinc_code, analyte, value, unit, ref_low, ref_high, collected_at, fasting_bool, lab_id, instrument, hemolysis_index. LOINC matters because it disambiguates serum vs plasma, fasting vs random, and method. Without it, your five-year series is five incompatible PDFs. Keep CGM exports (Dexcom CSV, 5-minute cadence, 288 points/day) in the same store and compute time-in-range, CV, and MAGE with the iglu R package rather than trusting the app’s summary.

Questions people also ask

At what age is a full body checkup best? There is no threshold age. The argument for starting in your twenties or thirties is that a baseline drawn while you are healthy is what makes a later value interpretable, given how much of blood-marker variance is individual rather than population-level.1 Screening decisions tied to age (colonoscopy, mammography) are clinical, and belong with your physician.

What is the most important single test? ApoB, if forced to pick one, because it counts atherogenic particles directly and is not confounded by triglycerides the way calculated LDL-C is. Lp(a) once, because it is genetic and you only need the number.

What are the five main tests in a full-body checkup? CBC, comprehensive metabolic panel, lipid panel, HbA1c, and TSH. That is the conventional core. It says nothing about particle count, inflammation, iron handling, or glucose dynamics, which is why we add ApoB, hs-CRP, ferritin with transferrin saturation, and CGM.

Can a blood test detect disease before symptoms? Sometimes. Multi-analyte tests combining circulating tumor DNA mutations with protein markers have demonstrated detection of several cancer types in otherwise asymptomatic people,9 and blood p-tau217 detects Alzheimer’s pathology years before clinical diagnosis.5 Both are probabilistic and both require clinical workup to act on. A negative result is not an all-clear.

What about telomere length? Interesting as a population variable, weak as a personal one. Associations with health indicators in general-population health checkups are modest and differ by sex.10 Assay imprecision is large enough that year-over-year change in one person is mostly noise.

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. 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 ↩ ↩2

  2. Lynn B Bailey, Patrick J Stover, Helene McNulty, et al. Biomarkers of Nutrition for Development—Folate Review. The Journal of Nutrition, 2015. https://doi.org/10.3945/jn.114.206599 ↩

  3. Riccardo E Marioni, Sonia Shah, Allan F McRae, et al. DNA methylation age of blood predicts all-cause mortality in later life. Genome Biology, 2015. https://doi.org/10.1186/s13059-015-0584-6 ↩

  4. Daniel W Belsky, Avshalom Caspi, David L Corcoran, et al. DunedinPACE, a DNA methylation biomarker of the pace of aging. eLife, 2022. https://doi.org/10.7554/elife.73420 ↩

  5. Nicolas R. Barthélemy, Gemma Salvadó, Suzanne E. Schindler, et al. Highly accurate blood test for Alzheimer’s disease is similar or superior to clinical cerebrospinal fluid tests. Nature Medicine, 2024. https://doi.org/10.1038/s41591-024-02869-z ↩ ↩2

  6. Ahmed Abdelhak, Matteo Foschi, Samir Abu-Rumeileh, et al. Blood GFAP as an emerging biomarker in brain and spinal cord disorders. Nature Reviews Neurology, 2022. https://doi.org/10.1038/s41582-021-00616-3 ↩

  7. Myron G. Best, Sjors G. J. G. In ’t Veld, Nik Sol, et al. RNA sequencing and swarm intelligence–enhanced classification algorithm development for blood-based disease diagnostics using spliced blood platelet RNA. Nature Protocols, 2019. https://doi.org/10.1038/s41596-019-0139-5 ↩

  8. Joshua F. Wiley, Tara L. Gruenewald, Arun S. Karlamangla, et al. Modeling Multisystem Physiological Dysregulation. Psychosomatic Medicine, 2016. https://doi.org/10.1097/psy.0000000000000288 ↩

  9. David Killock. CancerSEEK and destroy — a blood test for early cancer detection. Nature Reviews Clinical Oncology, 2018. https://doi.org/10.1038/nrclinonc.2018.21 ↩

  10. Hiroaki Arima, Shinsaku Inomata, Shirley Victoria Simpson, et al. Sex-specific associations between telomere length and health indicators: a cross-sectional study using Specific Health Checkups in Matsuura City, Japan. Scientific Reports, 2026. https://doi.org/10.1038/s41598-026-70097-1 ↩