Chapter 35 — Key Takeaways

One page.

⚠️ Carried from Chapter 34: this chapter is about products that generate individual rules about food and encourage continuous self-monitoring. ⚠️ For some readers that is not neutral. §35.12.

⚠️ THE STRUCTURE TO LEARN

⚠️ Theo's advertisement made three claims:

⚠️ 1. Your body responds to food differently from everyone else's — TRUE. ⚠️ 2. We can measure how — PARTLY TRUE, and it depends enormously on which test. ⚠️ 3. Eating to that measurement will make you healthier — ESSENTIALLY NOT DEMONSTRATED, and it is where all the money is.

⚠️ A real finding, a partly-real measurement, and an unevidenced outcome claim stacked on top. ⚠️ Every product in this chapter has that shape. The skill is seeing the join.

§35.1 — ⚠️ The Genuinely Impressive Finding, Stated First

⚠️ Weizmann Institute, Segal and Elinav, Cell 2015. Continuous monitoring across a large cohort and many thousands of real meals.

⚠️ Individual glucose responses to the SAME food differ SUBSTANTIALLY between people, are reasonably stable within a person, and are partly predictable from microbiome, blood markers and meal composition — better than carbohydrate content alone predicts.

⚠️ Real work, replicated in its core observation, and it properly complicates the glycaemic index. The GI was always a population average; this showed how much variation the average conceals.

⚠️ §35.2 — THE GAP

⚠️ Step 4 does all the work and is the least examined. In diabetes, glycaemic control is an established target with hard outcome evidence (Ch 26). In non-diabetic people it is unclear.

⚠️ This is Chapter 2's BETA-CAROTENE TEMPLATE with a subscription attached — a measurable intermediate travels with an outcome, and an intervention on the intermediate is sold before anyone tests whether moving it moves the outcome.

§35.3 — Genetic Tests

⚠️ ❌ "Your variant says low-carb or low-fat" · 🟠 MTHFR · 🟠 ACTN3 · 🟡 caffeine metabolism · ✅ lactase persistence and alcohol flushing — and you already know both.

⚠️ THE PATTERN: where the genetics is solid, you either already know or it is a medical diagnosis. Where the product tells you something new, the genetics is weak.

⚠️ AND THE SAME SAMPLE RETURNS DIFFERENT ADVICE — different panels, different effect-size assumptions, no agreed standard. A test that gives different answers to the same DNA is not measuring what it claims to.

⚠️ §35.3b — DIETFITS

⚠️ Stanford, Gardner, JAMA 2018. ⚠️ PRE-SPECIFIED the genotype hypothesis · randomized · one year · adherence supported · and BOTH arms ate a good diet.

⚠️ 1. Average weight change similar. ⚠️ 2. Individual variation within each arm ENORMOUS. ⚠️ 3. The genotype pattern did not predict who did better. ⚠️ 4. Nor did insulin secretion.

⚠️ Read 2 and 3 together: individual variation is real and LARGE — the genetic test simply did not explain it.

⚠️ A commercially attractive hypothesis, specified in advance, falsified by the people best placed to find it if true. Products making that claim are sold AFTER the test, not before it.

§35.4–35.6 — The Rest

⚠️ Microbiome tests: ⚠️ no reference standard (the comparison group is the company's own customers) · method-dependent · a snapshot · and the advice converges on eat more plants, fibre, variety and fermented food — ⚠️ which is good advice you'd have got free (Ch 27).

⚠️ ❌ IgG panels (Ch 28) · ❌ hair mineral analysis · ❌ live blood analysis · ❌ blood-type diets · 🟠 broad DTC micronutrient panels in asymptomatic people (Ch 13).

⚠️ The blood-type diet was tested directly. Some diets produced better markers. It had nothing to do with blood type — the better diets had more plants and less processed food.

⚠️ The cleanest available demonstration: the personalization did nothing; the dietary quality did everything.

⚠️ §35.5 — CGMs Without Diabetes: 🟡

⚠️ Three honest positions, all held at once: ⚠️ it shows you something real · there is no outcome evidence in non-diabetic people · and the interpretation problem is severe.

⚠️ Three rules if you use one: a FIXED period · compare MEALS not FOODS · and decide in advance you will not eliminate a food on the basis of a curve.

⚠️ §35.7 — WHAT PERSONALIZATION ACTUALLY HAS EVIDENCE FOR

⚠️ ✅ Diagnosed allergy · coeliac disease · intolerance · diagnosed disease · medication interactions · life stage · athletic demand · documented deficiency.

⚠️ 🟢 AND PREFERENCE, CULTURE, BUDGET AND SCHEDULE — the big one.

⚠️ Chapter 10's threshold was that adherence beats composition. So the most consequential personalization available is matching the diet to what you will actually eat.

⚠️ That IS personalized nutrition. It requires no test, produces larger effects than anything sold, and cannot be a subscription — which is much of why it is not what the word has come to mean.

§35.8–35.9 — Responders and n-of-1

⚠️ Apparent "responders" in a trial = measurement error + regression to the mean + differential adherence + genuine individual response. ⚠️ Only a REPLICATE CROSSOVER separates them.

⚠️ "Some people respond differently" is a hypothesis until someone shows the same person responds the same way twice. Most products have not shown that.

⚠️ n-of-1: 🟢 for short-latency, measurable, high-signal questions · ❌ for anything taking years. ⚠️ One variable · outcome defined first · ABAB not A-then-B · long enough blocks · blind if you can · prediction written down · and DECIDE IN ADVANCE WHAT WOULD MAKE YOU STOP — the rule nobody follows, and what separates an experiment from a justification.

⚠️ §35.10 — WHY THEY CONVERGE

⚠️ Collect the outputs — glucose algorithms, microbiome reports, genetic panels, apps — and they say:

⚠️ more plants · more fibre · more variety · fewer refined carbohydrates alone · protein and fat alongside carbohydrate · less ultra-processed food · move after meals · sleep · moderate portions.

⚠️ THREE REASONS: ⚠️ the pattern really is that robust · sensible algorithms HARD-CODE conventional constraints to stop themselves recommending cheese over lentils · and the data does not yet support more than marginal personalization.

⚠️ The most sophisticated personalization available, applied honestly, produces recommendations that look remarkably like the boring ones.

⚠️ That is not a failure of personalization. It is information about how much of nutrition is shared — and it is what Chapter 37 is built on.

⚠️ §35.10b — Theo bought two products in one month. One said reduce carbohydrate; the other said increase fibre-rich carbohydrate. Both said cheese was excellent. Both sold him supplements.

⚠️ Yolanda: "Look at what they disagree about. That's the part you paid for."

§35.11–35.11b — Fairness

⚠️ What would change my mind: a hard outcome · compared against GOOD conventional advice, not against nothing · durable past a year · independently replicated · individual response shown twice · and a test that replicates.

⚠️ PRECISION NUTRITION as a research programme is serious and will produce answers. ⚠️ What to watch: objective metabolomic biomarkers of intake, and STRATIFIED rather than individualized advice. ⚠️ Both would improve population guidance rather than replace it.

🧾 §35.13 — The Cost

⚠️ Genetic test $100–400 · microbiome $150–400 · CGM $70–200/month · subscription $30–100/month · IgG panel $150–500 · recommended supplements $300–1,200/yr. ⚠️ An enthusiastic first year: $1,500–4,000.

⚠️ Chapter 32 §32.13 costed a nutritionally excellent week for one adult at a small fraction of a CGM subscription.

⚠️ The person spending $2,000 a year discovering which foods suit them could have spent it on food.

One Thing to Remember

⚠️ Your responses genuinely differ from other people's — and after every test on the market has been run, the advice still converges on the boring answer.