Chapter 35 — Further Reading

⚠️ A field where the primary literature is unusually accessible and unusually worth reading directly, because the gap between what the papers say and what the products claim is visible to anyone who opens them.

⚠️ Read the papers, then read the marketing. That comparison is the chapter's whole lesson.

1. ⚠️ The glucose-response work — read this first

Zeevi, Korem, Segal, Elinav et al., "Personalized Nutrition by Prediction of Glycemic Responses," Cell (2015).

⚠️ The founding paper of the sector and a genuinely good piece of work. Read it properly.

⚠️ What to look for:

⚠️ The size of the between-person variation in response to identical foods — this is the real finding. ⚠️ The model's inputs and how much each contributes. ⚠️ And the validation and intervention sections, which are much smaller and shorter than the headline suggests.

⚠️ Then notice what the paper claims as its outcome: lower postprandial glucose. ⚠️ Not weight, not disease, not any hard endpoint. §35.2's step 4 is visible in the paper's own results section.

Subsequent independent replications and extensions, ⚠️ including work in different populations that confirms the core observation of substantial between-person variation. ⚠️ The observation replicates; the clinical claim has not been tested at the same standard.

The PREDICT studies (Berry, Spector, Segata et al.) — ⚠️ large twin-and-cohort work on postprandial responses to standardized meals. ⚠️ Genuinely informative on the sources of variation, including the finding that genetics explains less of it than people expect. ⚠️ Read the funding and affiliation statements alongside the results — this is the literature §35.14 means when it says a substantial part of the field is produced by the companies selling the product.

2. ⚠️ DIETFITS, and how to kill a hypothesis honestly

Gardner et al., "Effect of Low-Fat vs Low-Carbohydrate Diet on 12-Month Weight Loss… and the Interaction With Genotype Pattern or Insulin Secretion," JAMA (2018).

⚠️ Read the title carefully — the interaction hypotheses are in it, because they were pre-specified.

⚠️ What to look for:

⚠️ The pre-specification, and the fact that the genotype pattern was defined before the data. ⚠️ The dietary instructions to BOTH arms — maximize vegetables, minimize added sugar and refined flour — which is why the trial answers a narrower question than it is often reported as answering. ⚠️ And the individual-variation figure, which is the most instructive thing in the paper.

Gardner's own commentary and talks — ⚠️ unusually candid about having expected the genotype hypothesis to work. ⚠️ Worth reading as a model of how to report a null result you did not want.

⚠️ §35.3b's point stands: had the pattern been sought after the data came in, something would almost certainly have been found. Chapter 3's multiple-comparisons material is the reason.

3. ⚠️ The reproducibility problem in consumer testing

Journalistic and academic investigations sending identical samples to multiple DTC companies — ⚠️ and in some cases the same sample twice to one company.

⚠️ Consumer organizations in several countries have run these. ⚠️ The findings are consistent: materially different dietary recommendations from the same biological material.

Professional-body position statements on nutrigenomic testing — ⚠️ from dietetic and genetics associations. ⚠️ These are generally more sceptical than the marketing and more measured than a debunking, which makes them useful.

Regulatory positions on DTC health claims — ⚠️ worth reading for what companies are and are not permitted to claim in your jurisdiction, which is frequently narrower than what the advertising implies.

4. Microbiome testing

⚠️ Chapter 27's further reading covers the science. What to add here is specific to the products:

Critical appraisals of consumer microbiome testing by microbiome researchers — ⚠️ frequently written by people who are enthusiastic about the science and sceptical about the products, which is the most credible combination available.

Work on methodological variability — ⚠️ sequencing approach, sample handling, and bioinformatic pipeline effects on results. ⚠️ Chapter 27 §27.3b's "why two labs disagree" material.

5. ⚠️ CGMs in people without diabetes

Studies characterizing CGM patterns in non-diabetic populations — ⚠️ read these to see how poorly established "normal" is, which is the basis of §35.5's third interpretation error.

Any trial of CGM-guided eating in non-diabetic people — ⚠️ and note what the outcomes are. ⚠️ At the time of writing, the outcomes are glucose metrics and short-term behaviour, not disease.

⚠️ For contrast, the CGM literature in DIABETES — ⚠️ where the outcome evidence is strong and the technology is genuinely valuable. ⚠️ Reading the two side by side is the clearest way to see what has and has not been transferred.

6. ⚠️ The methodology — the most useful section here

Literature on the replicate crossover design and on distinguishing true individual response from apparent response.

⚠️ This is the most transferable material on the page and applies far beyond nutrition. ⚠️ Once you have it, you will see unjustified responder claims everywhere — in exercise science, in psychology, and in medicine.

Work on regression to the mean in health interventions — ⚠️ the mechanism behind a great deal of apparent individual response, and behind a great deal of apparent treatment effect generally.

Guidance on n-of-1 trial design — ⚠️ there is a formal methodological literature, used in clinical practice for genuinely individual treatment questions. ⚠️ Read it to see how much structure a proper n-of-1 requires, and how far the casual version falls short.

7. Precision nutrition as a research programme

⚠️ Major national precision-nutrition research initiatives — ⚠️ their published protocols and design papers are freely available and are worth reading for the contrast in §35.11b.

⚠️ Look at: cohort size and diversity, what is being collected, what the primary outcomes are, and how explicit the investigators are about what is not yet known.

⚠️ The contrast with a product website is instructive and is the fastest way to internalize the distinction between the research programme and the consumer sector.

Reviews of dietary-intake biomarkers — ⚠️ metabolomic and proteomic signatures of what people have actually eaten. ⚠️ If this works, it fixes the single largest methodological weakness in nutrition science (Chapter 3), and it would improve population research more than it would enable individual prescriptions.

8. ⚠️ Read with caution

⚠️ Books and podcasts by founders of personalized-nutrition companies. ⚠️ Some contain genuinely good science and all have a commercial interest in the conclusion. ⚠️ Chapter 3's framework for funded science applies — not dismissal, but attention to what is claimed beyond the data.

⚠️ Any source telling you that "everyone is different" therefore population advice is useless. ⚠️ §35.10 is the answer: the variation is real AND it sits on top of a large shared component, and the second half is not a rhetorical concession.

⚠️ Anything promising to tell you your "metabolic type," "nutritional type," or a numerical score with no established clinical meaning.

⚠️ And per §35.12: if a monitoring product has stopped being informative for you, the further reading you need is Chapter 34 §34.14, not this page.

9. Where this goes in the book

⚠️ One recommendation, if you read nothing else

Read the Zeevi 2015 paper's results section, and then read the homepage of any company citing it.

⚠️ Fifteen minutes, both free, and the distance between them is this chapter.