The Average Patient Doesn’t Exist: N-of-1 Experiments for Personalized Health

The Air Force measured 4,063 pilots across ten body dimensions. Not one of them was average on all ten.

Zero.

That report was supposed to help engineers design a better cockpit around the “average” pilot. Instead, it proved the average pilot was a person… who didn’t exist.

This same broken assumption governs much of modern health optimization too.

The strange part? Population research is useful. It surfaces broad patterns, flags real risks, and tests whether something is safe enough to study in the first place. What it can’t tell you?

Which dose, which schedule, which training style, which food, which supplement, or which recovery protocol gets the best result in the body you’re walking around in.

I’ve built my own health the opposite way. I test things on myself before I trust them, because the study population is never me. That’s the whole idea behind an N-of-1 experiment, and it’s the closest thing we have to science that answers your question.

Before we get into the how, here’s the short version of why one-size-fits-all keeps letting you down.

Why Does the “Average” Patient Keep Failing?

The pilot story highlights a major design flaw. If no pilot matched the average across ten dimensions, then equipment built around that made-up center point was unlikely to fit any real person well.

The engineers measured bodies for a practical reason, and arrived at a practical conclusion. Design for the range of real people, and forget the imaginary middle (Gilbert S. Daniels’ 1952 Air Force report, “The Average Man?”).

Health data carries the same trap.

The average blends all three paths into a single line and throws away the part you care about, which is where you’d land.

Roger J. Williams gave this idea its name in his 1956 book Biochemical Individuality: The Basis for the Genetotrophic Concept. In a 1957 survey, he cataloged apparent 5X, 10X, or larger ranges across organ sizes, blood enzymes, hormone excretion, nutrient needs, irritation thresholds, and taste.

So, essentially, across the dimensions of humanity.

The map of bioindividuality was never just genes, or food, or even supplements. In a 1957 essay, Williams made the abstraction concrete…

The more dimensions you measure, the faster the “average person” disappears (Williams, “Individuality and Education”).

Williams wasn’t arguing against science.

On the contrary. He wrote, “Generalizations are the lifeblood of science but spurious generalizations can only poison it.” Population research sets a starting prior. Your own repeated measurements update it.

How Different Can Two Healthy Bodies Really Be?

Using a controlled, within-person method, researchers exposed 55 healthy adults aged 18 to 30 to less than 1 lux and then 10 to 2,000 lux, for five hours in the evening. The estimated intensity needed to suppress melatonin by half ranged from 6 lux to 350 lux (Phillips et al., 2019, PNAS).

That’s a more than 50-fold spread. A room bright enough to move one person’s circadian system may hardly register for another.

Strength training tells the same story.

In 585 adults, a study used MRI and one-repetition-maximum testing before and after a standardized 12-week program for the non-dominant arm (Hubal et al., resistance-training response).

One person lost 2% of biceps cross-sectional area; another gained 59%. One person’s curl strength didn’t budge; another jumped 250%.

This isn’t just due to varying genetics.

A study of four pairs of identical twins showed they shared a matching baseline mix of B-cells. Yet, when given the same vaccine, each twin used completely unique B-cell clones to attack it.

Instead of saying “nature wins” or “nurture wins,” the study shows that both genetics and random chance work together at the exact same time. (Wang et al., 2015, PNAS)

Drug metabolism drifts too.

Your current medications, your inflammation, your age, and other exposures push an enzyme’s real activity away from what your genetic baseline. Researchers call it phenoconversion (Klomp et al., 2020, CYP phenoconversion systematic review).

Merely asking “what works?” in’t nearly as helpful as “what works for me, under which conditions, and how do I know?”

What Can Population Research Tell You About Yourself?

Population research can tell you what’s plausible, what’s risky, and what’s worth testing. It can catch a signal that would be invisible in a single person, and it can stop you from mistaking one great story for a reliable effect. Helpful, and also not the whole job.

Population research:

• Can’t hand you your own dose-response curve
• Can’t say whether your afternoon workout helps your sleep or wrecks it
• Can’t tell you whether a supplement lifts your energy once you account for your training load, your travel, your stress, your cycle, an illness, or a medication change
• Can’t predict whether you’ll be a non-responder or hyper-responder to a biohacking or longevity intervention

Even lab reference ranges are drifting toward personal context. Using a longitudinal modeling method on 2.1 billion measurements from 2.8 million adults, a 2021 study found that age and sex explained under 10% of the healthy variation in 89 of 92 common lab tests (personalized laboratory reference intervals, Nature Medicine).

That same method found that models built from each patient’s own history explained 60% of the variation for 17 tests and more than 36% for half of the tests.

Your prior biological responses carries information that age-and-sex buckets miss.

Here’s the takeaway…

A single population-wide cutoff can hide a real change from your own baseline, or set off alarm bells over a value that has always been normal for you.

That’s the gap an N-of-1 experiment closes. It’s a planned study in one person. You:

1. Define a question
2. Set a baseline
3. Change an intervention
4. Measure the outcome again and again
5. Compare the periods

Rather than chasing perfect certainty, you’re after a better decision than a generic recommendation or a single noisy reading.

Why Does More Tracking Still Produce Bad Answers?

More data isn’t the same as better evidence. Change your diet, add three supplements, start a new training block, fly across four time zones, implement cold plunging, and strap on a new wearable in the same week, and your results might look thrilling (while telling you nothing). You changed seven things. You can’t credit any one of them.

Personalization without design is just a fancier guess.

A credible experiment needs a question narrow enough to answer. It needs:

1. A baseline long enough to show you what “normal” even looks like for you
2. An outcome that matters
3. A time window that fits your biology
4. A plan for what you’ll do whether the results come back good, bad, or muddy

Most importantly, it has to watch for the stuff you weren’t looking for.

Better energy that costs you your sleep is a step towards burnout. Bigger biceps at the expense of diminished grey matter? Also not a worthwhile trade.

Even the label “nonresponder” can be wrong.

In a randomized crossover study, 21 adults completed both three-week endurance and sprint protocols. Every participant improved on at least one of three outcomes. A nonresponder may be responding to the wrong stimulus or the wrong metric (Bonafiglia et al., 2016, randomized crossover training study).

The AHRQ user guide to N-of-1 trials makes the same argument. These studies earn their keep when effects vary a lot between people, when the outcome can be measured over and over, and when the intervention is safe to start and stop under the right guidance (AHRQ N-of-1 trials guide).

What Makes an N-of-1 Experiment Scientific?

A scientific personal experiment has four things going for it: a testable question/hypothesis, repeated measurements, attempted isolation of signal from noise, and a final verdict.

It has a question you wrote down first. “Does morning light help?” is weak. “Does a 10-minute outdoor walk within an hour of waking reduce my bedtime variability over two weeks?” is testable.

It uses repeated measurements. One great night, or one ugly lab result, is an anecdote. A pattern that holds across comparable stretches of time is evidence.

On the High Performance Longevity Podcast, Dr. Jay Wiles gave a clean operating rule: treat every wearable data point as an N-of-1 experiment. Form a hypothesis, log the context, test it, retest it, and reformulate instead of chasing absolute numbers.

It pulls the intervention apart from the noise where it can. That can mean alternating conditions, holding your routine steady, or at least logging any context you can’t control (travel, a cold, a bad week) so you can read around it later.

Finally, it updates its own verdict. Grade the evidence as experimental, under test, or confirmed. No finding (a “null result”) earns its place too. It can save you the money, the attention, and the hope you’d otherwise pour into something counterproductive.

The stakes aren’t always trivial.

In the 12-week PREPARE trial, 6,944 adults were enrolled for 12-gene-guided prescribing or standard care. Using this randomized crossover method, the all-patient analysis found clinically relevant adverse drug reactions in 21.5% of the genotype-guided group versus 28.6% of the standard-care group, an odds ratio of 0.70. (PREPARE pharmacogenomics trial)

In other words, patients with prescriptions were customized to their genetics were 30% less likely to experience a harmful side effect compared to those who received standard treatment.

To be clear about self-experimentation…

N-of-1 work sits next to medical care, and your doctor still has say. Medication changes, worrying symptoms, and anything with a serious downside belong with a qualified clinician.

Why I Built Outliyr Around One Person at a Time

Outliyr is the first health optimization platform to give you the holistic view of your health and help you figure out what actually works and what might backfire.

Most health tools hoard fragments. One app owns your sleep. Another owns your labs. A spreadsheet owns your supplements. Your notes hold the context that explains why any of the numbers moved.

The answer is scattered across the seams between them, which is to say it’s nowhere.

I built Outliyr to close those seams. It brings together wearables, lab results, genetic and epigenetic tests, biological age, brain assessments, voice scan, facial analysis, supplements, protocols, journal context, and how you feel. You can ask one question across the whole picture instead of five apps (See the Outliyr method).

The Outliyr loop is simple:

1. Log what matters
2. Run a structured experiment
3. Get back an answer that can tell a real signal from a coincidence.

It tracks your baseline window, your target metric, the other outcomes, and the possible side effects, so a single impressive number doesn’t distract from the totality.

The Personal Scientist’s causal-inference backend is built to enforce those evidence gates, with tested estimators and reference fixtures. It gives your own experiments a clinical-grade structure you can inspect, without overwhelming you with stats and machine learning.

Outliyr is wearable-agnostic, meaning it’ll work perfectly with (almost) every wearable you use. I built the app to also be useful for those that can’t stand wearables.

Best of all, to the early birds that sign up now, the vast majority of the platform is free forever. Get started today.

What Else Should You Know About N-of-1 Experiments?

Build Health Advice for the Person Who Exists

The average patient (let alone, human) does not exist. You quite literally are special.

Population studies provide a starting hypothesis. Bioindividuality is why you test it. A well-built N-of-1 experiment closes the distance between “this tends to work” and “this reliably works for me.”

For example, some people notice that their sleep dramatically improves from supplementing with magnesium. It helps them fall asleep faster, and they sleep deeper. Others have a “paradoxical” response and feel stimulated. Completely opposite responses to an essential nutrient.

So start with one question that would move your next decision. Pick the outcome you care about. Track a baseline long enough to trust it.

Change one meaningful thing. Then review the result with enough humility to run it again before you believe it.

Give it a few real cycles and something shifts. You stop outsourcing your body to the median and start reading your own data like the primary source it is.

If you want the bigger picture on measuring the whole person instead of chasing one marker, I get into it on the BioHarmony conversation.

If you want one place to hold the whole process, create a free Outliyr account. Bring the data you already have, run your first experiment, and start collecting answers that belong to the person who’s here.

Know someone who’s been let down by “average” advice? Send this their way.