https://harmonism.io/wheel-of-harmony/health/monitor/blood-tests--what-to-order-and-why
Second spoke of the Monitor sub-pillar, Wheel of Health. Monitor holds the posture; this article holds the panel. See also: The Diagnostic Instrument (reading the decade a panel produces), The Sovereign Consultation (carrying the result into the room), Sovereign Health.
Marker count is not information. A panel of forty analytes drawn under controlled conditions and repeated annually at the same laboratory outperforms a three-hundred-marker one-off by a wide margin, and the industry sells breadth because breadth is legible, priceable, and easy to photograph. What changes outcomes is a small set of markers that sit on a causal pathway, that have an intervention behind them with outcome data rather than marker-movement data, and that are measured well enough that a change means something. Everything else is a number you paid for. And the sharpest lesson of the past year is that a marker can predict your death with great reliability and still be the wrong thing to aim at.
Start with the case, because it settles the principle faster than the principle can be argued.
Inflammation predicts cardiovascular events. High-sensitivity C-reactive protein has carried that association for three decades, robustly, across populations, independent of cholesterol. If you have atherosclerotic disease and your hs-CRP is above 2 mg/L, your risk is measurably higher than someone whose hs-CRP is low. The association is not in dispute and it never was.
So the field did the obvious thing. ZEUS randomised over 6,300 people with atherosclerotic cardiovascular disease, chronic kidney disease, and hs-CRP at or above 2 mg/L, to monthly ziltivekimab at 15 mg or to placebo. Ziltivekimab is a monoclonal antibody against interleukin-6, which sits directly upstream of hepatic CRP production. The drug did exactly what it was designed to do. Free IL-6 fell. hs-CRP fell.
The hazard ratio for major adverse cardiovascular events was 0.99, with a 95% confidence interval of 0.88 to 1.11. All-cause mortality did not move. Serious infections were more frequent on the active arm, which is what IL-6 blockade costs.
Read the interval rather than the point estimate. It is tight. A trial this size with this event rate had the power to find a modest benefit if one existed, and it found nothing on either side of the line. This is not an underpowered null awaiting a bigger study.
Hold four statements apart, and do not let them blur into each other.
hs-CRP remains prognostic. ZEUS did not touch the association, and a raised hs-CRP still tells you something real about the person in front of you. IL-6 inhibition as a therapeutic target in this population is null. The inflammatory hypothesis is not dead: CANTOS was positive, and colchicine holds regulatory approval on the strength of LoDoCo2 and carries guideline endorsement in Europe and South America. What is now genuinely open is whether any anti-inflammatory strategy earns its infection risk outside specific populations.
Four categories, four different epistemic statuses, one marker. That discipline is the whole method of this article, and hs-CRP is where it was learned the expensive way.
The instruction that does not survive is lower your CRP. It was never a target. It was a thermometer, and a decade of protocols told people to break the thermometer.
A marker earns a place on your panel by passing three tests. Most fail the second. Most of what functional medicine sells fails the third.
Causality. Does the marker sit on a causal pathway to something that kills or disables, or does it merely travel alongside one? Mendelian randomisation is the discriminator, because genetic variants are allocated at conception and cannot be confounded by the diet, the postcode or the exercise habit that confounds every observational cohort. Apolipoprotein B passes. Lipoprotein(a) passes. hs-CRP does not, and ZEUS is the reason we now know that with a randomised trial rather than an argument.
Actionability. Is there an intervention with outcome data behind it — deaths, events, dialysis, dementia — rather than marker-movement data? This is the filter almost everything fails, and the failure is invisible because moving a marker looks exactly like working. A supplement that lowers homocysteine by 30% has done something measurable and, for cardiovascular purposes, has done nothing at all.
Measurement integrity. Is the assay standardised, low-variability, and stable enough across laboratories and across years that a change in the number means a change in you? A great deal of what gets sold as advanced testing fails here quietly. The honest test is the split-double: submit the same sample twice, under a different name and a separate payment, and see whether the two results agree. Applied to the methylation-clock companies, this returns variation of roughly 40% between two readings of one body on one day. That is not a measurement. It is a number with a decimal point.
Every marker in the top two tiers closes on the same three questions, and the first of them governs. What decision does it change? What evidence tier does that claim sit at? And what monitors the intervention afterward, at what interval? If the first question comes up empty, the marker does not belong on the panel, however interesting it is. The tiers below Tier Two answer the first question conditionally — on a prevalence, on an assay, on a hypothesis — and the condition is stated each time rather than assumed.
One more thing the filters do, which is easy to miss. They rank markers within a scope; they do not set the scope. This article is about what comes out of a vein. Blood pressure would outrank most of Tier One on all three filters and is not ranked here at all, because it is a different instrument — it is handled at § What the Tube Cannot See, and treating filter-strength as a licence to widen the subject is how a focused article turns into a survey.
This section comes before the rankings, because a reader who internalises it and orders twenty markers correctly will out-perform a reader who orders three hundred carelessly. These variables move your numbers more than most interventions do.
Fasting duration and time of day. Twelve hours, water only, and go before ten in the morning if hormones are on the panel. Cortisol and testosterone both follow a diurnal curve steep enough that a two-hour difference in draw time can move a value more than a year of protocol.
Posture and tourniquet. Standing raises measured lipids and proteins by several percent relative to lying down, through plasma volume shift alone. A tourniquet left on past a minute does the same locally. Neither is exotic; both are routine, and neither is recorded on your result.
Hydration state. Every concentration-based measurement is a ratio with plasma volume in the denominator. Arriving dehydrated raises the whole panel.
Recent training load. Hard exercise within twenty-four hours elevates creatine kinase, alanine aminotransferase, ferritin and testosterone, in that rough order of sensitivity. A ferritin drawn the morning after a long run reports the run.
Acute illness within two weeks. This one invalidates three things at once, and the reason is a single mechanism: the acute-phase response. Ferritin rises, hs-CRP rises, and lipids fall. A panel drawn while you are fighting something reports an inflammatory state, not a baseline, and it will look like both good news and bad news simultaneously.
Biotin. High-dose biotin — routine in hair and nail supplements, and in many B-complexes — interferes with the streptavidin-biotin chemistry underlying a large share of immunoassays, including thyroid and troponin. Hold it for three days. This has sent people to cardiology.
The same laboratory, every time. Assay platforms differ, and they drift. A trend assembled from three laboratories is a trend of the laboratories. If you change laboratory, you have started a new series and the old one does not continue into it.
The sample-quality flags. Haemolysis and lipaemia indices appear on most reports and are almost never read. Haemolysis invalidates potassium and lactate dehydrogenase. Lipaemia interferes with several photometric assays. A flagged sample is a redraw, not a datapoint.
Control these, and your year-over-year comparison means something. Ignore them, and you have bought an expensive random number generator and will spend the next decade interpreting its output.
A small set of results does not change across a life. You buy them once and they govern decisions for fifty years, which makes their cost per decision the lowest on the panel by a wide margin. Most people never order any of them.
Six are worth the single draw. One is offered with its price attached.
Lipoprotein(a). Genetically determined, minimally responsive to diet, exercise or statins, and causal by Mendelian randomisation for atherosclerotic disease, calcific aortic stenosis and heart failure. The designers of the current outcomes trial put more than 20% of the world’s cardiovascular disease population above the elevated threshold, and that figure travels as background rather than as measured epidemiology, so treat it as an order of magnitude and not a statistic.
What it changes: how hard you push apolipoprotein B downward, since Lp(a) is a risk you cannot lower and ApoB is a risk you can — and cascade screening, because each parent, sibling and child carries a coin-flip chance of the same inheritance and none of them will be tested unless you tell them. Tier: causal. What monitors it: nothing, which is the point.
One qualification, added honestly. A March 2026 preprint reports heterogeneity in Lp(a) trajectories across the menopausal transition, which complicates the flat claim that the number never moves. Preprint status places this at genuinely open. A woman measuring in her thirties may want one confirmation after the transition rather than treating the value as fixed for life.
HFE genotype, for hereditary haemochromatosis. Among the most common inherited disorders in people of Northern European descent. Iron loads silently for decades and then deposits in the liver, pancreas, heart and joints, at which point the damage does not reverse. Caught early, treatment is venesection — the oldest and cheapest intervention in medicine.
What it changes: the interpretation of every ferritin you will ever draw, and cascade screening of siblings, who share the genotype at high probability. Note the ordering honestly: phenotype leads. Transferrin saturation and ferritin are the first-line screen, and the genotype is what you order when the phenotype is suggestive or a first-degree relative is affected. Penetrance is incomplete — carrying the genotype is not the same as loading iron — so a genotype ordered first, in isolation, produces anxiety more reliably than it produces a decision. Tier: causal, with incomplete penetrance stated. What monitors it: ferritin with transferrin saturation, annually, for life.
G6PD status. X-linked, and common enough in Mediterranean, North African, Middle Eastern, sub-Saharan African and South Asian ancestries that it is unreasonable to assume against it. The result gates specific oxidant exposures — primaquine and tafenoquine for malaria, rasburicase, methylene blue, nitrofurantoin, dapsone, high-dose intravenous vitamin C, and fava beans.
What it changes: whether several ordinary prescriptions and one popular biohacking compound are safe for you. This is the clearest case on the panel of a test whose value is entirely in the exposures it forecloses. Tier: causal, with pharmacologic guidance behind it. What monitors it: nothing.
Familial hypercholesterolaemia genetics. Variants in the LDL receptor, in apolipoprotein B, and gain-of-function variants in PCSK9 produce lifelong elevated particle burden from birth rather than from midlife. The condition is substantially underdiagnosed almost everywhere it has been counted, and the reason is structural: a person with FH looks like a person with high cholesterol, and gets managed as one, decades late.
What it changes: the treatment threshold, which moves earlier and lower, and cascade screening of first-degree relatives, which is where most of the value sits. A confirmed variant also removes the perennial argument about whether the number is lifestyle. Tier: causal. What monitors it: ApoB.
Pharmacogenomics, the actionable subset only. A small number of gene–drug pairs carry prescribing guidance from the Clinical Pharmacogenetics Implementation Consortium and, for several, from drug labelling. The highest-yield for a general adult: SLCO1B1 and statin myopathy, which decides whether the most-prescribed drug class in the world will be tolerated; CYP2C19 and clopidogrel, where poor metabolisers get a prodrug that does not activate; DPYD and the fluoropyrimidines, where a deficiency turns a standard chemotherapy dose into a lethal one; TPMT and NUDT15 with the thiopurines, on the same logic.
What it changes: a prescribing decision, usually years before the prescription. The asymmetry is what makes this Tier Zero rather than a curiosity — the cost of knowing is one buccal swab, and the cost of not knowing is discovered during the reaction. Tier: actionable by guideline; the outcome evidence varies by pair and is strongest where the toxicity is severe. What monitors it: nothing.
Coeliac HLA-DQ2 and DQ8, when the question is live. Nearly everyone with coeliac disease carries one of these haplotypes, so a negative result has a very high negative predictive value and closes the question permanently. A positive result establishes almost nothing — the haplotypes are common in the general population — and the diagnosis still requires serology on a gluten-containing diet.
What it changes: it retires a question rather than answering one, which is worth a great deal to anyone who has been circling gluten for years. Order it when the question keeps recurring, not as routine screening. Tier: strong for exclusion, near-useless for inclusion.
And one offered with its price. APOE genotype is real risk information for Alzheimer’s disease with no reliable intervention behind it and a non-trivial psychological cost. It is a choice rather than a checkbox, and the moment to make it is before the sample is drawn rather than when the result arrives. There is no guideline recommending it for asymptomatic people, and that absence is a position rather than an oversight.
What does not belong here, despite being sold as though it does. Direct-to-consumer ancestry-and-wellness genotyping interpreted for “functional variants” — MTHFR most of all. The variants are extremely common, the downstream claims outrun what the genetics supports, and the third filter is where it fails: the interpretation layers sold on top of a raw genotype file are not measurements at all.
The live front on Lp(a). Lp(a)HORIZON, registered as NCT04023552, randomised 8,323 patients with established cardiovascular disease and Lp(a) at or above 70 mg/dL to pelacarsen 80 mg subcutaneously each month or to placebo. The primary endpoint is the composite of cardiovascular death, non-fatal myocardial infarction, non-fatal stroke, and urgent coronary revascularisation requiring hospitalisation, tested in the whole population and again in the subgroup above 90 mg/dL. It is event-driven and closes at 993 adjudicated events. Phase 2 achieved reductions up to 80%. Mendelian estimates suggest an absolute reduction somewhere between 70 and 90 mg/dL is needed to produce a 15 to 20% risk reduction, which is a large required effect and the reason this trial is not a formality. Sponsor guidance has moved at least once, and no date is offered here — a predicted readout is the one thing certain to date this article.
Five, in descending order of what they decide.
One atherogenic particle carries exactly one molecule of apolipoprotein B. So ApoB counts the particles, while every cholesterol measurement weighs their cargo — and cargo per particle varies between people by enough to matter clinically.
The proof runs through discordance analysis, and the method matters because ordinary statistics cannot separate variables this highly correlated. Discordance analysis finds the people in whom the markers disagree and asks which one predicted their outcome. The systematic review reaches an unusually clean verdict: across 15 studies and 593,354 participants, ApoB outperformed LDL cholesterol in 9 of 9. Against non-HDL cholesterol, of the nine studies making that comparison, seven favoured ApoB significantly, one found equivalence, and one favoured non-HDL cholesterol. That single dissenting study is worth naming rather than rounding away, because a review with no exceptions in it usually means the exceptions were not looked for.
In UK Biobank, ApoB beat even LDL particle number, which is the closest competitor it has. Across 41,099 participants with at least ten years of follow-up, 9,663 cardiovascular events and 1,754 coronary events, risk rose significantly at 2% discordance — a hazard ratio of 1.1 for both endpoints — climbing to 1.4 for events and 2.5 for coronary disease at 30% discordance. The asymmetry is the finding. Where ApoB exceeded particle number, risk rose; where particle number exceeded ApoB, it did not.
European guidelines concluded in 2019 that ApoB is the more accurate marker, and cheaper, and more precisely measured than calculated LDL cholesterol. Seven years on, most European check-ups still do not order it.
The sharpest consequence is the one clinicians resist. Judging whether a statin, ezetimibe or PCSK9 inhibitor is doing its job by watching LDL cholesterol is an interpretive error about what those trials demonstrated. The benefit tracked particle reduction. The reported endpoint was cholesterol because cholesterol was what the laboratories measured.
What it changes: whether you treat, how hard, and whether current therapy is adequate. Tier: causal. What monitors it: ApoB itself, at eight to twelve weeks after any change in therapy or diet, then annually.
A note on thresholds, held deliberately loose. This article prints no target number. Every published cutpoint is a population-level convenience, the right target depends on your Lp(a), your calcium score, your family history and your age, and a reader who takes a single number from an article and steers by it has reproduced the error the article is about.
Triglycerides come free in every lipid panel ever drawn, and they are read as a lifestyle scold rather than as the signal they are.
Two things make them worth a proper entry. First, they are the most responsive marker on the whole panel — dietary carbohydrate, alcohol, and insulin resistance move them within weeks, which makes them the fastest feedback loop available for judging whether a change is working. Second, they let you calculate remnant cholesterol from a panel you already have: total cholesterol minus HDL cholesterol minus LDL cholesterol. That residue is the cholesterol carried in triglyceride-rich lipoproteins, and the Copenhagen population studies have made a sustained case that it carries risk information beyond LDL-C. It costs nothing and almost nobody computes it.
Mark the boundary precisely, because this is where the field itself got confused. Triglyceride-rich lipoproteins carry risk; lowering triglycerides pharmacologically has repeatedly failed to deliver. The fibrate outcome trials on top of statin therapy are the record, and PROMINENT — the most recent and best-designed of them — stopped for futility despite moving the lipid numbers as designed. This is the ZEUS pattern in a second field, and it is the reason triglycerides sit here as a signal rather than higher as a target.
The free proxy worth knowing: the ratio of triglycerides to HDL cholesterol tracks insulin sensitivity well enough to be useful for anyone who cannot get insulin ordered, which is most people in most systems.
What it changes: it is the fastest read on whether a dietary change is working, and remnant cholesterol refines cardiovascular risk at zero marginal cost. Tier: the lipoprotein is causal; the number as a drug target has failed. What monitors it: itself, at four to twelve weeks.
Four numbers, three of which most people already have, and the missing one is the whole point.
Insulin resistance precedes dysglycemia by years and sometimes by more than a decade. Glycated haemoglobin reports the average of the previous three months, and by the time it moves the compensating capacity has already been spent. Fasting glucose moves later still — it is the last thing to break, because the body will raise insulin indefinitely to hold glucose normal. Fasting insulin catches the compensation while it is still working, which is the entire interventional window, and it is the one of the four almost never ordered.
HOMA-IR is arithmetic on fasting glucose and fasting insulin rather than a separate assay, so it costs nothing beyond the two tests and gives a single number that moves coherently.
Ketones belong here rather than in a tier of their own. Beta-hydroxybutyrate measured in blood is the reliable form — urine strips report acetoacetate and become unreliable precisely when a person adapts, which is when they most want the reading, and breath acetone tracks the blood value only approximately. What a blood ketone reading gives you is confirmation that a fasting or carbohydrate-restricted protocol is doing metabolically what it claims, on the day you are doing it. What it does not give you is an outcome: no trial has tied a ketone level to a hard endpoint, and a number pursued for its own sake is the article’s own error in a new costume. Measure it to verify a protocol, not to optimise a figure. The protocols themselves are at The Fasting Principle and Fasting Protocols.
What it changes: whether dietary and movement intervention starts now or after a diagnosis. Tier: causal for the pathway; the specific cutpoints are conventional. What monitors it: the same four at twelve weeks after a dietary change.
The most under-ordered high-yield test in medicine, and the omission this article most wants to correct.
Kidney disease is silent. Albumin appears in urine years before the estimated filtration rate falls, so a normal creatinine tells you nothing reassuring about a kidney that is already leaking. The reason this belongs in Tier One rather than in a prognostic tier is that the finding now routes to treatment. CREDENCE randomised 4,401 people with type 2 diabetes and urine albumin above 300 mg/g to canagliflozin, and returned a hazard ratio of 0.70 (95% CI 0.59–0.82) on a composite of end-stage kidney disease, doubling of creatinine, and renal or cardiovascular death. DAPA-CKD randomised 4,304 people with albumin between 200 and 5,000 mg/g, with and without diabetes, and returned 0.61 (95% CI 0.51–0.72). Both stopped early.
State the mechanism precisely, because it is easy to overclaim. In these trials albuminuria was an entry criterion, not a surrogate endpoint. Nobody has demonstrated that lowering your albumin-to-creatinine ratio improves your outcome. What the trials demonstrated is that people with albuminuria benefit from these drugs, which makes the measurement the thing that identifies you as such a person. EMPA-KIDNEY, at 6,609 participants and a hazard ratio of 0.72, did not require albuminuria at all in its low-filtration stratum, and so does not support the narrower claim.
Two cautions. A qualitative dipstick reporting protein is not an albumin-to-creatinine ratio, and a great many European check-ups carry the former while calling it kidney screening. And creatinine-based filtration estimates overstate function in anyone with low muscle mass, which is why cystatin C belongs beside it and becomes the better estimate with age.
What it changes: whether a drug class with outcome evidence is indicated. Tier: causal for the pathway, outcome-proven for the intervention in the albuminuric population. What monitors it: the ratio and the filtration estimate, every six to twelve months.
Never ferritin alone. Ferritin is an acute-phase reactant, which means inflammation raises it and a single value cannot distinguish a well-stocked iron store from an inflamed one. Transferrin saturation is the discriminator, and the pair costs almost nothing.
Two diseases hang on it, running in opposite directions. Hereditary haemochromatosis is completely treatable if caught before iron deposits and untreatable afterward, because deposition does not reverse — this pair is the phenotype screen that the HFE genotype in Tier Zero sits behind. Iron deficiency is the most common nutritional deficiency in the world and the most frequently missed cause of fatigue in menstruating women, in whom a ferritin technically within range is routinely reported as normal.
What it changes: venesection in one direction; repletion and a search for the bleeding source in the other. Tier: causal and outcome-proven for both. What monitors it: the same pair at three months.
These markers are worth measuring and dangerous to aim at. Each requires you to hold the four categories apart.
hs-CRP. Treated at the head of this article, and placed here rather than in Tier One for exactly the reason ZEUS established. Measure it, because it carries prognostic information and because it tells you when the rest of the panel is uninterpretable. Do not set a target and chase it.
Homocysteine, which splits four ways. This is the model case for how a contested marker should be handled, and the handling is more useful than the marker.
The Cochrane review pooled 15 trials and 71,422 participants. Homocysteine-lowering with B vitamins produced a relative risk of 1.02 (95% CI 0.95–1.10) for myocardial infarction and 1.01 (0.96–1.06) for all-cause mortality. Both null. Stroke returned 0.90 (0.82–0.99) — a modest reduction whose upper bound touches the line, with a number needed to treat of 143 over 5.4 years, and only in the comparison that combined B vitamins with antihypertensive therapy.
So the cardiovascular target failed. Then VITACOG found something else entirely. It randomised 271 adults over 70 with mild cognitive impairment to folic acid 0.8 mg, B12 0.5 mg and B6 20 mg daily, or placebo, for 24 months, with the rate of whole-brain atrophy on serial volumetric MRI as its stated primary endpoint. Atrophy ran at 0.76% per year (95% CI 0.63–0.90) on treatment against 1.08% (0.94–1.22) on placebo, p = 0.001. In the subgroup entering with homocysteine above 13 µmol/L, atrophy was 53% lower. A later post-hoc analysis found the benefit concentrated in those with adequate baseline omega-3 status: above 590 µmol/L of plasma long-chain omega-3, atrophy slowed by 40%; below it, the vitamins did nothing at all.
Four lines, held apart. Failed as a cardiovascular target. Alive as a neurocognitive one, in an elevated subgroup. Conditional on omega-3 repletion, per a post-hoc subgroup interaction within a single trial, which is weaker evidence than the headline result and is marked as such. And open as to whether any of it generalises to people without cognitive impairment.
Measure it. Know precisely which of those four claims your number supports.
Thyroid: TSH with free T4 and thyroid peroxidase antibodies. TSH alone is a screen rather than an assessment, and anti-TPO identifies the autoimmune subgroup and predicts progression to overt disease, which a single TSH cannot. Note the boundary honestly: treating subclinical hypothyroidism in older adults is contested, and the corpus holds no settled position on it. Thyroid architecture beyond the panel entry is out of scope here by design.
GGT, ALT and uric acid. Gamma-glutamyl transferase is the most underrated marker in routine chemistry. Its reputation is as an alcohol flag; its actual behaviour is as a sensitive index of oxidative stress and hepatic xenobiotic load, and it predicts incident metabolic disease well beyond that reputation. It is already on every basic panel you have ever had. Almost nobody reads it.
Omega-3 index. Cheap, standardised, with a strong mortality association — and, per the VITACOG post-hoc, possibly a gate on whether an unrelated intervention works at all. That last property is rare enough to justify the test on its own.
Widespread deficiency changes the calculation. Where a nutrient is genuinely depleted across a large fraction of the population, testing an asymptomatic person is not a fishing expedition — the prior probability is high enough that the test is answering a real question. This tier holds the nutrients where that is true, and it is also where the third filter does its hardest work, because several of the most-discussed nutrients have no assay worth ordering.
25-hydroxy vitamin D, paired with parathyroid hormone. Deficiency is common enough across northern latitudes, indoor life and darker skin that population testing is defensible. Correcting a genuine deficiency is worth doing. Supplementing the already-replete has failed repeatedly on hard endpoints, VITAL being the largest demonstration. Parathyroid hormone is what distinguishes those two situations, and it is almost never ordered alongside — a rising PTH against a low-normal vitamin D says the body is compensating and the deficiency is real, where the same vitamin D with a normal PTH often says it is not.
Magnesium — and the assay problem is the lesson. Deficiency is common, the physiology is central, and serum magnesium is close to useless as an index of status, because the overwhelming majority of body magnesium sits in bone and inside cells while the serum fraction is held tightly regulated at the expense of the stores. A normal serum magnesium is compatible with substantial depletion. Red-cell magnesium is better and is what to ask for; it is also less standardised between laboratories than the markers in Tier One, which means it obeys the same-laboratory rule more strictly than most of the panel.
This is the clearest worked example of the third filter on the whole page. A marker can be causally important, the intervention can be cheap and safe, and the measurement can still fail — and when it does, the honest move is to treat the assay as weak evidence rather than to pretend it is strong. Magnesium repletion is one of the few cases where an empirical trial of the intervention is more informative than the test.
B12 with methylmalonic acid, and folate. Serum B12 alone misses functional deficiency, and the miss is not rare. Methylmalonic acid accumulates when B12 is functionally insufficient at the cellular level, which catches people whose serum B12 reads normal while their metabolism reports otherwise. Order the pair, particularly with any plant-predominant diet, any metformin use, any proton-pump inhibitor use, and past sixty, where absorption falls with gastric acid.
Iron. Covered at Tier One, and it belongs conceptually here too — it is the most common deficiency in the world.
And the ones whose assays are not good enough to screen with: zinc, copper, selenium, iodine. Each is genuinely important. Each has an assay that fails filter three for population screening in different ways — plasma zinc is heavily influenced by inflammation, albumin and time of day; copper needs ceruloplasmin beside it to mean anything and the pair matters mostly for the zinc-to-copper balance in anyone supplementing zinc; selenium status is regional and reflects soil more than intake; and iodine is properly assessed on urinary excretion at population scale rather than on a serum value in an individual. Order these against a question — a supplementation regimen, a restricted diet, a regional exposure, a symptom picture — rather than as routine coverage.
The general rule this tier teaches: a common deficiency justifies screening only where the assay can detect it. Prevalence sets the prior; measurement integrity decides whether the test can move it.
The tests below are good tests when a question comes first, and expensive noise when it does not. Sex hormones with sex hormone-binding globulin. Cortisol and DHEA-S. Coeliac serology, on a gluten-containing diet, or it reads falsely negative. Autoimmune panels. Heavy metals and toxicant load.
The failure mode has a mechanism rather than a moral. Run enough tests on a healthy body and abnormal results are guaranteed, because a reference range is defined so that a fixed proportion of healthy people fall outside it. Order twenty independent tests on a well person and you should expect one or two flags from arithmetic alone. Each flag then generates follow-up, and follow-up carries its own scans, biopsies, months and risks. This is how people acquire diagnoses they did not have — not through bad tests, but through good tests ordered without a question in front of them.
The discipline is simply to have the question first. What symptom, what exposure, what family history, what protocol am I verifying? A named question makes any of these tests worth ordering. Its absence is what turns a panel into a hazard.
High-plex proteomics measures thousands of proteins from a few dozen microlitres and has no individual reference intervals, which makes it a research instrument rather than a clinical one. Methylation clocks are classified as wellness products rather than medical devices, and the split-double problem described earlier is theirs. Untargeted metabolomics is in the same position.
Blood-based Alzheimer’s markers have moved, and the movement is narrower than the coverage suggests. Two phosphorylated-tau 217 assays now carry FDA clearance: a plasma ratio test cleared in May 2025 for adults over 50 presenting to specialised care with cognitive symptoms, and a single-biomarker test cleared in August 2026 for people 55 and over with signs or complaints of cognitive decline, in primary as well as specialty care. The second carries explicit labelling that it is not a stand-alone test and has not been established for predicting the development of dementia or for monitoring therapy. The Alzheimer’s Association’s 2025 guideline scopes blood biomarkers to patients with objective cognitive impairment, permitting them as a triage test above 90% sensitivity and 75% specificity and as a substitute for amyloid imaging or spinal fluid above 90% on both.
No test in this class is cleared or guideline-endorsed for screening people without symptoms. If you have no cognitive complaint, this is not yet a test you can act on.
Multi-cancer early detection needs its own paragraph, and the paragraph is not the one the vendors write. The NHS-Galleri trial screened roughly 140,000 adults aged 50 to 77 across three annual rounds. Its primary endpoint was the incidence of stage III and IV cancers across twelve prespecified types, and the result was an incidence rate ratio of 1.03 (95% CI 0.92–1.14), p = 0.63 — a three per cent increase, which is to say nothing at all. The trial missed.
It also found things. Stage IV diagnoses fell, at a rate ratio of 0.86 (95% CI 0.744–0.998). Screen-detected cancers numbered 1,173 against 290 under standard care alone, roughly fourfold. Emergency-setting diagnoses fell by a figure that different official sources report as either 20 or 25 per cent, which is why no figure is printed here.
Now compare the sponsor’s own headline against the trial’s own primary endpoint. The press release leads with the stage IV reduction, the fourfold detection rate, and increased early-stage detection of deadly cancers. It does not lead with the miss. Nothing in it is false and none of it is fraud. This is what a company does with a trial that produced interesting secondary findings and failed its primary one, and recognising the shape of it is a transferable skill — considerably more transferable than any single fact in this article.
This section is not a list of tests to skip. Almost everything in it arrives free inside a standard lipid or metabolic panel, and you should read it when it does. What fails here is a narrower and more specific thing: using these numbers as the variable a decision turns on, or as the target an intervention aims at. The distinction is the whole article in miniature, and collapsing it produces exactly the reflex — don’t test that — that the three filters exist to replace.
Total cholesterol as a decision variable. It sums particles that do opposite things and hands you one number that averages them. Read it as arithmetic input to remnant cholesterol; do not steer by it.
HDL cholesterol as a therapeutic target. Cholesteryl ester transfer protein inhibitors raised it substantially and delivered nothing, which is as clean a refutation as this field produces. But HDL still carries information at the extremes: a very low HDL is a real metabolic signal, and a very high HDL is not the good news it was sold as. Measure it, include it in the triglyceride ratio and the remnant calculation, and stop there.
LDL cholesterol as the metric for judging whether therapy is adequate. This is the most consequential routine error on the list, and it is worth being exact about what is being claimed. Measure LDL-C — it comes in the same panel and it is the entry point to the discordance question. What fails is treating it as the adequacy metric when ApoB is available, because the trials that justified the therapy tracked particle burden and reported cholesterol. Where ApoB genuinely cannot be obtained, non-HDL cholesterol — total minus HDL — is the better of the two remaining options and costs nothing extra.
Tumour markers in asymptomatic screening. They appear on several large commercial panels. In a healthy population they generate false alarms faster than the follow-up can resolve them, and the follow-up involves scans, biopsies and months.
IgG food-sensitivity panels. IgG to a food indicates exposure. It has been shown to track what a person eats, which is why the results feel uncannily accurate and mean nothing.
Most cortisol-based adrenal-fatigue protocols. The underlying observations are often real. The interpretive framework built on them is not.
The panel changes with age. What changes is not the diseases a decade carries but the value of a measurement, which depends on what that measurement can still alter.
Twenties and thirties. The once-ever set, which is cheapest to buy here because it pays out over the longest remaining span: lipoprotein(a), G6PD where ancestry makes it plausible, and the pharmacogenomic subset. Then the annual core — apolipoprotein B, triglycerides, fasting insulin, transferrin saturation with ferritin. The argument for measuring this early has nothing to do with the likelihood of finding something wrong. You are establishing the trajectory before the slope changes, and an Lp(a) drawn at twenty-five sets the plan for the following fifty years at the cost of one blood draw.
Forties. Add the albumin-to-creatinine ratio and the full thyroid set. Add a coronary artery calcium score, which is imaging rather than blood and which the panel is incoherent without, because plaque becomes measurable in this decade and a calcium score of zero means something quite different from a calcium score of two hundred at identical ApoB.
Fifties. For women, the menopause-anchored sequence below, which should have started a decade earlier. For men, prostate-specific antigen with its overdiagnosis problem stated rather than buried.
Sixty and beyond. An anaemia workup rather than a haemoglobin. B12 with methylmalonic acid. Albumin as a frailty index. And cystatin C in place of creatinine for estimating kidney function, because creatinine-based estimates flatter a body that has lost muscle, and reporting good kidney function to a sarcopenic person is a systematic error rather than a random one.
This is the section that most changes what a reader does, and almost no general article on blood testing carries it.
The menopause transition is the dominant timing fact in a woman’s lipid history rather than a background variable within it, and it determines whether every subsequent measurement can be interpreted at all.
The Study of Women’s Health Across the Nation followed 1,054 women through the final menstrual period and asked a specific question: do midlife changes in cardiovascular risk factors fit a model of chronological aging, or a model of ovarian aging anchored to the transition itself? The answer separated the risk factors cleanly. Blood pressure, glucose, insulin, fibrinogen and C-reactive protein followed the linear pattern of ordinary aging. Total cholesterol, LDL cholesterol and apolipoprotein B did not. Those three showed substantial increases inside the one-year interval spanning the final menstrual period, consistent with a menopause-induced change rather than the passage of time, and the pattern held across ethnic groups.
The instruction that falls out of this is concrete, and it is the most actionable sentence in the article.
Draw the baseline in the late thirties or early forties, before the transition, not after symptoms start. A postmenopausal ApoB of a given value means one thing in a woman whose premenopausal value was low and something entirely different in a woman whose premenopausal value was already high, and without the earlier draw those two women are indistinguishable. Most women are measured for the first time after the change has already happened, and the step is then invisible — absorbed into a single number that looks like a lifetime baseline and is not one.
Two further items belong in a woman’s panel and are routinely missed. Ferritin with transferrin saturation across the whole reproductive span, because iron deficiency with a technically normal ferritin is the single most commonly missed cause of fatigue in menstruating women. And thyroid autoimmunity, which is several times more prevalent in women and which anti-TPO identifies before TSH moves.
One question is genuinely open and is presented as such. Whether sex-neutral ApoB thresholds over-predict risk in postmenopausal women is a live and important question, and this article states it without a number attached, because the analysis currently advancing it could not be independently verified. That is a real gap rather than a rhetorical hedge, and it is named so that a reader encountering a confident sex-specific cutpoint elsewhere knows what it rests on.
For men, three notes. Total testosterone with sex hormone-binding globulin, since most panels report total alone and the calculated free value is arithmetic rather than an assay. Haemochromatosis, which expresses earlier and more severely in men because they lack the monthly loss that masks it in women. And prostate-specific antigen, which deserves an honest overdiagnosis conversation rather than reflexive inclusion or reflexive omission.
Rank the panel by what it costs against what it decides, and a pattern appears that ought to be embarrassing to the field.
Apolipoprotein B, the urine albumin-to-creatinine ratio, transferrin saturation and fasting insulin sit at the top. All four are inexpensive. All four are standardised. All four route directly to an intervention with outcome evidence behind it. And all four are ordered rarely, while a great deal of money moves through tests that fail the second filter entirely.
Neither conspiracy nor stupidity explains it. A cheap test with a clear decision attached is a poor product, and a three-hundred-marker panel with a beautiful dashboard is an excellent one. The incentive runs toward breadth because breadth is what can be sold, and nothing in the structure of the market rewards the four tests above.
Which is the whole argument in one observation. The panel that changes your life costs less than the panel that impresses you.
Four things sit outside the vein, and one of them outranks most of this page.
Blood pressure. On all three filters it beats nearly every marker in Tier One: causality established beyond argument, interventions with half a century of mortality data, and a measurement cheap enough to repeat daily. It is not a blood test, which is why it is here rather than in the rankings — decision yield ranks markers within a scope; it does not set the scope. Measure in both arms at least once, since a persistent difference above roughly 15 mmHg is itself a vascular finding, then measure at home, seated and rested, across a week, and trust that series over any single clinic reading. The practice belongs to Monitor § The Continuous Signal.
A coronary artery calcium score reports plaque that has already formed, which no lipid value can tell you — and it is the measurement that converts an ambiguous ApoB into a decision.
A DEXA scan reports bone density and body composition, and composition is what weight was always a poor proxy for.
The daily continuous signal — sleep, resting heart rate, heart rate variability read as a within-person trend against your own baseline, continuous glucose — belongs to Monitor rather than to this article, because a worn instrument is part of the posture rather than part of the panel.
A bloods-only protocol is incomplete, and saying so strengthens the argument rather than weakening it. The same principle applies throughout: measure what changes a decision, wherever the measurement happens to live.
What this comes to, concretely.
Once, ever. Lipoprotein(a). G6PD status, where ancestry makes it plausible. HFE genotype, where the phenotype or a first-degree relative points at it. Familial hypercholesterolaemia genetics, where the lipid picture or the family history warrants it. The actionable pharmacogenomic subset. Coeliac HLA typing, when the question keeps recurring. APOE only on a deliberate choice.
Annually, in a settled adult, drawn under the conditions in the noise-floor section, at one laboratory.
Lipids and particles. Apolipoprotein B, plus the standard lipid panel it arrives with, which yields total cholesterol, LDL-C, HDL-C and triglycerides, and from those four, remnant cholesterol by subtraction.
Metabolic. Fasting glucose, fasting insulin, HbA1c, and HOMA-IR computed from the first two.
Kidney. Urine albumin-to-creatinine ratio, with creatinine and cystatin C.
Liver, blood and purine. Full blood count with differential; liver enzymes including GGT; uric acid.
Iron. Ferritin with transferrin saturation.
Inflammation. hs-CRP, read as information rather than as a target.
Nutrients. 25-hydroxy vitamin D with parathyroid hormone; red-cell magnesium, remembering what the assay is worth; B12 with methylmalonic acid, and folate; omega-3 index.
Thyroid. TSH with free T4 and anti-TPO.
Methylation. Homocysteine, filed under the neurocognitive claim rather than the cardiovascular one.
That is roughly twenty-five determinations, and most of them arrive bundled — the lipid panel is one order, the metabolic panel is one order, the full blood count is one order. It fits on one page, it is cheap in most systems, and it carries more decision weight than any three-hundred-marker product currently sold.
Alongside it, not in it. Blood pressure at home across a week. Ketones, when a fasting or carbohydrate-restricted protocol is running and you want confirmation it is doing what it claims.
What is deliberately absent. Zinc, copper, selenium and iodine, unless a question calls them; sex hormones and cortisol, unless a question calls them; tumour markers; IgG food panels; anything from Tier Five. Their absence is the argument.
At twelve weeks after any deliberate change — a therapy, a diet, a fasting protocol, a supplement stack — retest only the markers that change should have moved, and only those. A full panel repeated at twelve weeks mostly measures noise, and noise at twelve-week intervals is how people conclude that something worked.
Every six to twelve months where a specific problem is under management.
The rhythm matters more than the breadth. Twenty markers drawn the same way every year for ten years is a decade of signal. Three hundred markers drawn once is a photograph of one morning, taken under conditions nobody recorded.
None of this is optimisation, and reading it as optimisation is the failure mode closest at hand.
Monitor is the fractal of Presence applied to the body — attention turned inward, continuously, with instruments. The instruments serve the attention. They do not manufacture it, and a panel handed to someone who cannot feel their own fatigue has changed nothing except the location of the file. The interior listening comes first, always, and everything above is downstream of it.
What the panel adds is reach. It reaches the dysregulations that direct sensation cannot report — the ApoB that produces no symptom for forty years, the albumin in the urine that arrives long before anything hurts, the iron accumulating quietly in an organ. This is alignment with Logos at the biological register: the order is already operating in the body whether or not it is observed, and observation is how a person comes into accord with what is already the case.
The discipline the article asks for is narrower than it looks. Before any test, one question. What would I do differently depending on the answer? If the honest answer is nothing, the test is entertainment, and you will pay for it twice — once at the laboratory and again in the months spent interpreting a number that was never going to change anything.
Ask the question. Order the twenty. Draw them the same way, at the same place, every year.
Then read the decade.
See also: Monitor, The Diagnostic Instrument, Monitor in the First Year, The Sovereign Consultation, Sovereign Health, Wheel of Health, Root Cause of Disease, Cancer — The Harmonism Protocol, Logos, Sovereignty.