Regenerative & personalized
Personalized nutrition
Postprandial variability between individuals, the microbiome's contribution to it, and the evidence gap between a measured difference in glucose response and a recommendation engine.
The empirical core of personalized nutrition is solid and worth stating precisely. Give a large group the identical standardised meal under controlled conditions and their postprandial responses diverge widely — several-fold differences in the glucose excursion, and even wider spread in triglycerides. This has been replicated, notably in the PREDICT studies, which used identical test meals in twins and unrelated participants. The effect is real, it is large, and it means population-average glycaemic tables describe nobody in particular.
Where the variation comes from
Several mechanisms are established. Gastric emptying rate differs between people and largely sets how fast glucose arrives. Insulin sensitivity and beta-cell response set how fast it leaves. Incretin signalling, GLP-1 and GIP, modulates both. Context contributes as much as biology: glucose tolerance falls across the day, so an evening meal produces a higher excursion than the same meal at breakfast; the preceding meal, sleep and recent exercise all shift the curve.
The microbiome’s contribution is genuine but narrower than marketing implies. Non-digestible carbohydrate reaching the colon is fermented to short-chain fatty acids — acetate, propionate, butyrate — which act on gut hormone release and hepatic metabolism, and gut bacteria also transform bile acids, which are themselves signalling molecules. Microbiome composition adds predictive power to models of postprandial response; the influential Weizmann work showed a machine-learning model including microbiome features predicting glucose responses better than carbohydrate counting, with a small blinded crossover showing that diets chosen this way lowered postprandial glucose.
Nutrigenomics is the weakest leg. A handful of variants have clear phenotypes — lactase persistence at the LCT locus, ALDH2 and alcohol, CYP1A2 and caffeine clearance. The common variants sold on consumer panels, such as FTO or MTHFR, explain small fractions of variance and have repeatedly failed to predict who benefits from which diet. The DIETFITS trial tested exactly this and found neither a genotype pattern nor insulin secretion predicted differential weight loss between low-fat and low-carbohydrate assignment.
The gap that is not closed
Between “responses differ” and “therefore this app’s score should govern your shopping” sit several unproven steps.
Postprandial glucose in a person without diabetes is a surrogate, not an outcome. Its association with long-term risk is inferred, not demonstrated as a target to be minimised by food choice. Continuous glucose monitors have real accuracy limits — error against blood reference is typically high single-digit percentages, and two sensors worn simultaneously can disagree by more than the difference between the two foods a user is comparing. Day-to-day within-person variability to the same meal is also substantial, so a single exposure is a noisy measurement being read as a personal fact.
Finally, the commercial algorithms are proprietary and their end-to-end claim — that following the recommendations improves health outcomes rather than the surrogate — is largely untested in independent trials. That is why these products are sold as wellness services rather than under a health-claim authorisation, which in the EU would require substantiation under Regulation (EC) 1924/2006. The biology is real; the inferential chain built on top of it is where the evidence thins.