The Optimal Blood Test Is the One You Repeat Under the Same Conditions

Oak
biomarkers blood-tests reference-ranges preanalytics longitudinal
Five identical iridescent winged reptiles in the same pose on matching black plinths, lit by one hard light, receding into darkness.

There is no single optimal blood test, and the marketed “optimal ranges” sold alongside large panels are usually softer evidence than the number formatting suggests. The approach we would take is different: draw a compact, outcome-linked panel under tightly controlled conditions, repeat it on a fixed schedule, and interpret each result against your own prior values rather than against a population interval. A reasonable annual core is apolipoprotein B, HbA1c, fasting glucose and insulin, a complete blood count with differential, a comprehensive metabolic panel with cystatin C-based estimated glomerular filtration rate, high-sensitivity CRP, a full iron panel with ferritin, TSH, 25-hydroxyvitamin D, and urine albumin-to-creatinine ratio, with lipoprotein(a) measured once in your life. What turns that list into information is the repetition and the control of preanalytical conditions, which is where most self-directed testing falls apart.

What “optimal” means, and where the term is doing real work

A standard reference interval is the central 95% of results from a reference population screened to be free of obvious disease. It is a statement about who else looks like your sample, not about your risk. So 2.5% of healthy people fall outside any given interval by construction, and an interval built from a population with a high prevalence of subclinical metabolic disease will happily call an unhealthy value normal.

“Optimal range” is an attempt to fix that by anchoring to outcome data instead of population percentiles. For a few analytes this is well grounded: apolipoprotein B and LDL cholesterol have monotonic, dose-dependent relationships with atherosclerotic events across large genetic and interventional datasets, and HbA1c maps onto microvascular risk. For most of the analytes on a 59-marker panel, the “optimal” band is an interpolation from small observational studies, sometimes a single cohort, often with confounding that no one has adjusted away. Treating a homocysteine or a free T3 “optimal” window with the same confidence as an ApoB target is a category error. The first-principles question to ask of any marker is whether its analytical performance and its link to the outcome are strong enough to change a decision at your prior probability, which is the framework used in diagnostic biomarker selection generally 1.

Why your own baseline beats any published range

Every analyte has two variances: within-person variation around your own homeostatic set point (CVi) and between-person variation across the population (CVg). When CVi is small relative to CVg, your personal set point sits somewhere inside a wide population interval, and you can move a long way from your own normal while staying comfortably “in range.” Studies of biological variation quantify these components precisely so that analytical quality goals can be derived from them, and the same numbers let you decide whether a population range is informative for you at all 2.

The practical tool is the reference change value, the difference between two serial measurements that exceeds what analytical noise and normal biological fluctuation can explain:

RCV = 1.96 * sqrt(2) * sqrt(CVa^2 + CVi^2)     # two-sided, 95%

For LDL cholesterol with an analytical CV near 3% and a within-person CV near 6%, RCV is about 19%. A move from 100 to 110 mg/dL is noise. For high-sensitivity CRP, where within-person variation routinely exceeds 40%, a doubling between two draws means very little, which is why a single CRP is close to useless and three draws averaged are worth something. Compute RCV for each analyte you track, store it beside the value, and stop reacting to changes smaller than it.

The second statistical trap is multiplicity. If you order 40 independent analytes each with a 95% reference interval, the probability that at least one comes back flagged is 1 - 0.95^40, roughly 87%. Large panels manufacture abnormalities. That is an argument for fewer, better-chosen markers and for repeat testing before acting on any single flag.

Preanalytics decide half the result

The largest source of spurious variation in self-ordered bloodwork is not the assay. It is what happened in the twenty-four hours before the draw and the thirty minutes after it. Anticoagulant choice alone shifts the measured biochemical profile substantially: EDTA, heparin, citrate, and clotted serum give systematically different values across hundreds of small molecules, so a serum result and a plasma result are not interchangeable 3. Time from venipuncture to centrifugation, storage temperature, and freeze-thaw cycles introduce further drift, which is why well-run biomarker programs protocolize collection tube, processing delay, aliquot volume, and reporting of every deviation 4.

Concretely, the rules we follow for a serial panel are these. Draw at the same time of day, ideally 07:00 to 09:00, because cortisol, testosterone, and serum iron all have strong diurnal swings. Fast 12 hours with water allowed when fasting insulin and triglycerides are on the panel. Sit quietly for 10 minutes before the draw, since moving from supine to upright concentrates protein-bound analytes by roughly 5 to 10%. Avoid hard exercise for 48 hours, which otherwise inflates creatine kinase, ALT, and AST. Postpone the draw if you have had a febrile illness or a vaccination in the prior two weeks, because CRP, ferritin, and the white count will all move. Use the same laboratory and ask which platform runs each assay: ferritin, testosterone, and free T4 immunoassays are not harmonized across vendors, and switching labs mid-series creates a step change you will misread as biology. Request the hemolysis index with any potassium, LDH, or AST result.

A concrete protocol for the first year

Establish the baseline before you start interpreting trends. Draw the core panel three times over six weeks under identical conditions. Take the mean as your set point and the observed spread as an empirical check on the published CVi. After that, once a year is enough for most analytes, with a second draw at three months for anything you changed on purpose.

Store the results as data, not PDFs. A minimal schema that survives lab changes:

date,time,analyte,loinc,value,units,method,lab,fasting_hours,hemolysis_index,cv_a,cv_i
2026-03-04,07:42,Apolipoprotein B,1884-6,78,mg/dL,immunoturbidimetric,LabA,12,0,0.03,0.07

LOINC codes matter because they pin down specimen type and method, which is exactly what gets lost when you retype a number off a report. With cv_a and cv_i in the table you can compute RCV per analyte in three lines and flag only changes that exceed it. Note that lipoprotein(a) is reported either as mass (mg/dL) or particle concentration (nmol/L) and the two do not convert reliably, so record the units and keep the assay constant.

What we would not buy

Untargeted tumor marker panels in asymptomatic people are the clearest waste. CA-125, CEA, and CA 19-9 have low specificity and, at the prevalence of undiagnosed cancer in a healthy adult, produce far more false positives than true ones, each of which leads to imaging and biopsy. The research effort to find blood-based markers with usable sensitivity at early stage is active and genuinely hard 5, and even promising serum protein panels for colorectal carcinoma remain at the discovery and validation stage rather than the screening stage 6. Multiplex mass-spectrometry proteomic panels are similar: powerful for discovery, but requiring formal analytical validation of precision, linearity, and interference before any individual result should be read as clinical 7. If a panel arrives with a “biological age” or an “optimal” band and no stated CVa, no method, and no citation, treat the number as decorative.

Anything outside your reference interval, and anything that crosses the RCV in a direction you did not intend, belongs in front of a physician who can take a history and examine you. Interpretation of a molecular profile is a measurement exercise. Diagnosis is not.

Questions people also ask

What bloodwork should you get every year? The core we described above, drawn under fixed conditions: ApoB, HbA1c, fasting glucose and insulin, CBC with differential, CMP with cystatin C eGFR, hs-CRP, ferritin with iron and transferrin saturation, TSH, 25-OH vitamin D, and urine albumin-to-creatinine ratio. Lipoprotein(a) is a one-time measurement because it is largely genetically determined.

What is the difference between normal and optimal lab results? Normal means inside the central 95% of a reference population. Optimal means associated with lower risk of a specific outcome in some study. The first is a well-defined statistical statement, the second varies enormously in evidentiary quality depending on the analyte.

Which blood test is best for overall health? No single test covers it. If forced to pick one number for long-run cardiovascular risk it would be ApoB, and for metabolic status HbA1c paired with fasting insulin, but the value comes from tracking a small set over years rather than from any one draw.

Why do my results change between labs? Different platforms, calibrators, and specimen types. Ferritin, free T4, and testosterone immunoassays are poorly harmonized, and serum versus plasma differences alone shift many analytes 3. Keep the lab and method constant across a series.

How do I know whether a change is real? Compute the reference change value from the analytical and within-person coefficients of variation and treat smaller differences as noise 2. For high-variability analytes like CRP, average several draws instead of interpreting one.

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. Samantha A. Byrnes, Bernhard H. Weigl. Selecting analytical biomarkers for diagnostic applications: a first principles approach. Expert Review of Molecular Diagnostics, 2017. https://doi.org/10.1080/14737159.2018.1412258 ↩

  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 ↩ ↩2

  3. Adam D. Kennedy, Lisa Ford, Bryan Wittmann, et al. Global biochemical analysis of plasma, serum and whole blood collected using various anticoagulant additives. PLOS ONE, 2021. https://doi.org/10.1371/journal.pone.0249797 ↩ ↩2

  4. Xuemei Zeng, Yijun Chen, Anuradha Sehrawat, et al. Alzheimer blood biomarkers: practical guidelines for study design, sample collection, processing, biobanking, measurement and result reporting. Molecular Neurodegeneration, 2024. https://doi.org/10.1186/s13024-024-00711-1 ↩

  5. Samir M. Hanash, Christina S. Baik, Olli Kallioniemi. Emerging molecular biomarkers—blood-based strategies to detect and monitor cancer. Nature Reviews Clinical Oncology, 2011. https://doi.org/10.1038/nrclinonc.2010.220 ↩

  6. 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 ↩

  7. Emily S. Boja, Thomas E. Fehniger, Mark S. Baker, et al. Analytical Validation Considerations of Multiplex Mass-Spectrometry-Based Proteomic Platforms for Measuring Protein Biomarkers. Journal of Proteome Research, 2014. https://doi.org/10.1021/pr500753r ↩