What is Oxford CEBM Level 5 evidence?

Level 5 is the lowest rung of the Oxford Centre for Evidence-Based Medicine hierarchy — case reports, expert opinion, and mechanistic reasoning. No controls, no blinding, no statistical power. Project Peakspan is a single-subject case report and therefore structurally sits at Level 5 forever. The data is real; the claims do not generalise.

Why publish biomarker drops half-yearly rather than quarterly?

Most clinical biomarkers need a full cell-turnover cycle (roughly 12 weeks) to reflect a real intervention change. Retesting on a 3-month clock frequently measures noise rather than signal. Half-yearly drops give interventions time to express in the data while keeping cadence tight enough to correct course. Exceptions are made for acute or symptomatic issues.

What counts as an intervention in an N=1 experiment?

Any deliberate, logged change to diet, training, supplementation, pharmaceuticals, or lifestyle that is maintained for at least 12 weeks with a pre-declared target and stop criterion. Every intervention is recorded alongside the biomarkers it is meant to move, and null results are published with the same detail as positive ones.

Why should I not follow the Peakspan protocols?

Because it is a single-subject case report. An intervention that moved a marker for Niko Hems may do nothing (or harm) for someone with different genetics, history, or comorbidities. Pharmaceutical-level interventions require a clinician in the loop. Anything described here is a data point, not a recommendation.

Where does the biomarker data come from?

Primarily from the YEARS preventive medicine clinic in Berlin (Dr. Alexandru Ardelean) with lab work run through Bioscientia. The 2025 stool panels are from Bioscientia Ingelheim. The 2024 US draws are from Quest Diagnostics. The earliest German baseline (June 2024) is from aware.app via MDI Limbach Berlin. Wearables (Apple Watch, Eight Sleep) supply continuous data.

Methodology — Project Peakspan

Evidence levels · 1 to 5.

Every intervention and claim on this site carries an evidence weight. The one I use is the Oxford Centre for Evidence-Based Medicine (OCEBM) hierarchy — 1 at the top, 5 at the bottom. It's what clinicians use to weigh studies against each other.

Peakspan itself sits at Level 5. This site is a single-subject case report. That is the lowest rung of the hierarchy. Nothing I publish here constitutes evidence in the sense that higher-level research does. I use this scale to rate the inputs I rely on (guidelines, trials, mechanistic claims), not to elevate the outputs of this project. The outputs stay Level 5 forever — that's structural, not a target to climb.

Systematic reviews of RCTs.

Meta-analyses and Cochrane reviews of randomised controlled trials. Strongest evidence available. Anchors most of my lipid and cardio decisions.

Individual RCTs.

Well-designed, adequately powered randomised trials. Most direct support for specific interventions (e.g. Ezetimib on ApoB, zinc-L-carnosine on mucosal healing).

Cohort / case-control.

Non-randomised controlled studies with large populations and follow-up. Where I lean when RCTs don't exist (e.g. grip strength as mortality marker).

Case series · low-quality cohort.

Observational without controls or with known bias risk. Hypothesis-generating. I cite with explicit caveats.

Case report · mechanistic reasoning · expert opinion.

Single subjects and arguments from first principles. No controls, no blinding, no statistical power. This is where Peakspan lives. The data is real, the claims are not.

Peakspan is here

Two things follow from sitting at Level 5:

Any lifestyle or supplement protocol I describe is a data point, not a recommendation. If I want to argue that an intervention is generally worth trying, I borrow the argument from Levels 1–3 and cite it inline. My experience adds flavour; it never moves the evidence bar.
Calibrate trust accordingly. Drug decisions (Ezetimib, B-complex) reference Level 1–2 evidence. The 2025 gut protocol was held together partly by Level 2 (zinc-L-carnosine, curcumin) and partly by Level 5 (L-glutamine, the specific stack order) — which is exactly why I can say the stack worked for me but not claim the stack works.

Cadence · half-yearly, with one annual deep dive.

Too-frequent measurement is noise. Too-infrequent measurement is hope. The compromise I've settled on:

Annual: full YEARS check-up — complete lab panel, CPET for VO₂max, body-comp, HRV profile, grip, jump, audio/visual, skin scan. This is the one where the whole picture comes together. January each year.
Half-yearly drops (H1 & H2): two public drops per calendar year. H1 wraps the annual data; H2, published around July, is a mid-year retest on whatever moved, plus any new interventions on a ~6-month clock. Published within 30 days of the relevant draw.
Continuous: body weight, resting HR, HRV, sleep stages, workout load — wearable-tracked daily, summarised in each drop.
Ad hoc: when symptoms change (e.g. the Sept 2025 gut retest after the protocol), outside the schedule.

One hard rule: no drop is skipped because the numbers are bad. A missed drop would be the loudest possible signal that the ledger isn't actually open.

Intervention rules.

01
Single-variable where possible — and annotated where not.
The gut protocol in 2025 was a stack. I could not tell you which component drove calprotectin from 434 to 9. That's a confound. I say so. Where I can isolate (e.g. Ezetimib starting alone in 2026), I do.
02
Minimum 12 weeks before retest.
Most clinical biomarkers take a full cell-turnover cycle to reflect a real change. Retesting sooner usually measures noise, which is part of why the public drop cadence is half-yearly, not quarterly. Exception: acute inflammation or symptomatic problems, where I retest earlier and say so.
03
Pre-declared target + stop criterion.
Every intervention gets a specified target (e.g. ApoB ≤70, homocysteine <10) and a stop criterion (no movement after two retests → reassess). No moving goalposts.
04
Null results get the same word count as wins.
If an intervention doesn't move a marker, I write about it with the same detail. Publication bias is the most common way individual data goes wrong.
05
Physician in the loop.
Dr. Alexandru Ardelean (YEARS) is my primary. Prescription-level interventions (e.g. Ezetimib) are his decision, not mine. Lifestyle and OTC supplement changes I own, but I'm not freelancing on pharmacology.
06
No new intervention in the month before a YEARS check-up.
Otherwise I pollute the comparison that's supposed to anchor the drop.

Data sources.

Everything below the dashboard tiles comes from one of these. If a number has no citation next to it, it's from one of these. Raw PDFs available on request: niko@nikohems.de.

Primary clinic · annual

YEARS, Berlin

Dr. Alexandru Ardelean, internal medicine. Full annual check-up: CPET, plethysmography, HRV, body-comp, skin scan, audio, vision, grip, jump, cognitive battery, and full labs via Bioscientia.

Half-yearly labs

Bioscientia MVZ · Berlin + Ingelheim

Standing orders through YEARS. Ingelheim runs the stool / microbiome panels (calprotectin, α1-antitrypsin, SCFAs, diversity). Berlin runs the blood work.

US draws (2024)

Quest Diagnostics

Three draws Oct–Dec 2024 while living in California. LC-MS testosterone, 25-OH vitamin D, hs-CRP, full metabolic. Useful for the trans-continental baseline bridging June 2024 (aware.app) to Feb 2025 (YEARS).

Early baseline

aware.app · MDI Limbach Berlin

June 2024 baseline draw, pre-Berkeley. Useful context for the 2024 → 2026 drift, but a different assay family than Bioscientia; treated carefully in direct comparisons.

Wearable · continuous

Apple Watch · Eight Sleep

Resting HR, HRV (rMSSD), sleep stages, training load. Not clinical-grade, but the trend is honest, and the morning HRV reading drives daily training decisions.

Gym & training log

Hevy · manual

Every session, every set, every RPE, since mid-2025. Provides the lift-side picture the clinic tests can't capture.

What's in the panel — and what isn't.

Tracked half-yearly or annually

Lipid panel: total, LDL-C, HDL-C, non-HDL-C, triglycerides, ApoB, Lp(a), remnant cholesterol
Inflammation: hs-CRP, NT-proBNP, ferritin (contextualised against haematology)
Gut: calprotectin, α1-antitrypsin, sIgA, microbiome diversity (annual), histamine
Metabolic: fasting glucose, HbA1c, fasting insulin, HOMA-IR, OGTT (annual)
Thyroid & endocrine: TSH, fT3, fT4, TPO-Ak, Tg-Ak, IGF-1, cortisol, DHEA-S, total & free testosterone, SHBG, oestradiol
One-carbon + vitamins: homocysteine, active B12 (holotranscobalamin), folate, 25-OH vitamin D, magnesium, zinc, selenium, Omega-3 Index, Vitamin B6 (PLP)
Renal: creatinine, cystatin C, eGFR, ACR
Fitness: CPET (VO₂max, VT1/VT2), grip R/L, countermovement jump, body composition (BIA + skinfolds), spirometry / plethysmography
Neurocognitive battery (YEARS annual)
HRV profile (clinic + wearable)

Not (yet) tracked

CGM · planned for a future drop, not standing
Epigenetic age clocks · intentionally excluded until between-lab reproducibility improves
Continuous BP · on the backlog
Polysomnography · the Eight Sleep data is the current proxy

Honest limitations.

A public-facing health experiment carries some built-in biases. The ones I can name:

Publication bias, in reverse. When a wellness space full of silver-bullet claims meets a single person with recovery data, there's a pull to over-dramatise the wins. I try to weigh wins and non-movers equally. Every drop has a "what didn't move" section.
Observer effects. Knowing a retest is coming changes behaviour. I try to minimise it by keeping the retest window short and the intervention window long.
Single-subject. Everything in this experiment is n=1. A number moving for me says nothing about what will happen for you.
Assay drift. Different labs, different reference ranges, different analytic methods — especially between US (Quest, mass spec) and Germany (Bioscientia). I note the lab alongside the value; direct comparisons within one lab only.
Confounds. My diet changed. My sleep changed. My stress changed. My training changed. Any of those can move most of the markers here. Ascribing a change to a single pill or protocol is frequently wrong and I try to say so.

Evidence levels · 1 to 5.

Cadence · half-yearly, with one annual deep dive.

Intervention rules.

Single-variable where possible — and annotated where not.

Minimum 12 weeks before retest.

Pre-declared target + stop criterion.

Null results get the same word count as wins.

Physician in the loop.

No new intervention in the month before a YEARS check-up.