Individual variation in health and disease. What modern genetics and traditional medical systems have in common

Summary

Two people eat the same meal and their blood glucose rises by different amounts. Two patients take the same drug and one is helped while the other is not. A disease risk score built in one population loses accuracy when it is applied to another. Questions like these are central to precision medicine, and they are the questions that shaped my own research.

The idea behind them is older than the technology. Persian medicine, Ayurveda and Chinese medicine each built a system for classifying people before treating them. Mizaj, Prakriti and syndrome differentiation are different answers to one observation, that people differ in how they fall ill and in how they respond to food and treatment. This essay asks what modern genetics and nutrition science can say about that observation, and whether any part of these older classifications describes variation that we can now measure.

It does not argue that the older systems were right, or that Mizaj is a genotype in disguise. It argues something narrower. The problem that precision medicine is trying to solve is not new. What is new is that we can finally test it.

Introduction

Modern medicine learns from populations. The randomised trial, the genome-wide association study, the cohort and the meta-analysis all work by pooling many people so that a pattern too weak to see in one person becomes visible in thousands. This is the method behind most of what medicine knows.

It also produces a paradox that every clinician meets. The average patient in a trial does not exist. Each average is built from individuals who differ in ancestry, genetics, diet, environment, microbiome, life history and physiology, and a treatment effect estimated across all of them is a statement about the group, not about the person in the consulting room.

Precision medicine grew out of this recognition. The United States Food and Drug Administration describes it as an approach to prevention and treatment that takes into account differences in people’s genes, environments and lifestyles, with the aim of giving the right treatment to the right patient at the right time [1]. Stated this way the aim is simple. Reaching it is not, because it requires knowing which differences matter, and for whom.

That raises the question this essay follows. Long before a genome could be read, physicians in several civilisations were already sorting patients into types and adjusting their advice accordingly. Were they, in their own vocabulary, trying to practise individualised medicine?

Part I. The problem with the average patient

A trial may find that a drug lowers systolic blood pressure by 15 mmHg on average. Inside that average some people improve a great deal, some a little, some not at all, and a few are harmed. The average describes the population. The physician treats the individual.

This is not an abstract concern. Pharmacogenomics gives the clearest cases. Variants in the genes that metabolise drugs change how fast a medicine is cleared and how likely it is to cause harm, and for a growing list of drug and gene pairs the evidence is strong enough that the dose is adjusted before the first prescription [2]. The FDA keeps a table of such associations for exactly this purpose [3].

So the question is not whether people differ in their response to treatment. They do. The question is what produces the difference, how much of it can be measured in advance, and whether older attempts to classify people captured any part of it.

Part II. Where the differences come from

Genetics

Any two human genomes agree at more than 99.9 percent of positions, but a genome is three billion positions long, so a typical person still carries four to five million sites at which they differ from the reference sequence [4]. Most of that variation does nothing measurable. Some of it shifts disease risk, immune response, physiology and the handling of drugs and nutrients, and fifty years of twin studies put the heritability of the average human trait at close to half [5].

Insulin resistance illustrates how much structure a single clinical label can hide. When we ran genome-wide studies on 21 insulin sensitivity indices, measured in the fasting state and after a glucose load, the indices separated into groups with distinct genetic architectures, and the genetic risk scores built from them associated with different cardiometabolic traits [6, 7]. Two people with the same diagnosis can carry different biology behind it.

Genetics, though, is only part of the story, and for many outcomes not the largest part.

Environment

People develop within an environment, and the environment leaves a measurable mark. The Global Burden of Disease study attributes a large share of deaths worldwide to exposures that are not inherited, among them air pollution, unsafe water, occupational hazards and temperature [8]. Genetically similar people living in different places can have very different health.

Behaviour

What people do with their bodies changes their biology. The clearest demonstration is the Diabetes Prevention Program, a randomised trial in adults at high risk of type 2 diabetes, in which a lifestyle programme of diet and physical activity reduced the incidence of diabetes by 58 percent over about three years, compared with 31 percent for metformin [9]. Sleep, smoking, food and physical activity carry as much of the story as any genome, and unlike a genome they can change.

Microbiome

Each person also carries trillions of microbes, mainly in the gut, and their composition differs widely between people. In a cohort of over a thousand healthy adults, microbiome composition was shaped far more by environment and diet than by host genetics [10]. This matters for the argument because the microbiome is part of what makes two people respond differently to the same food, as Part III shows.

Genes, environment, behaviour and microbes do not act separately. Each person is one particular combination of all four, and the outcome of interest, a glucose response, a drug effect, a disease that appears or does not, is the result of that combination.

Part III. The same meal, different biology

The strongest evidence that individual responses differ comes not from genetics but from nutrition. Dietary guidelines have long assumed that a given food has a broadly predictable effect on the body. Large studies over the last decade have tested that assumption directly, by feeding people identical meals and measuring what happened in their blood.

Ten thousand people, two weeks of glucose

The largest test to date comes from the Human Phenotype Project in Israel, which follows more than 10,000 adults with two weeks of continuous glucose monitoring alongside genomics, gut microbiome, imaging and diet records [11]. The post-meal glucose response to the same food varies widely between people, and a food that raises glucose sharply in one person can barely move it in another. In 2026 a model trained on more than ten million glucose readings from 10,812 of these adults learned representations of a person’s glucose dynamics that transferred to 19 external cohorts in five countries and predicted individual glycaemic responses to food [12]. The variability is real at scale, and it can now be modelled.

PREDICT 1, 2020

The PREDICT 1 study in the United Kingdom took 1,002 adults, most of them twins, and gave them standardised test meals in a clinic and at home [14]. After identical meals, the between-person coefficient of variation was 103 percent for blood triglycerides, 68 percent for glucose and 59 percent for insulin. Genetics explained little of this, under 10 percent of the variance for the glucose response and under 1 percent for triglycerides. Person-specific factors such as the gut microbiome, meal timing and sleep explained more. Even identical twins responded differently to the same food.

A caution on measurement

One recent study cuts the other way and belongs here. In two inpatient feeding studies at the US National Institutes of Health, 30 adults without diabetes were given identical meals about a week apart under fully controlled conditions [13]. The same person’s glucose response to the same meal was only weakly reproducible, with intraclass correlations between 0.17 and 0.28, and the day to day variation within a person was as large as the variation between different meals. Between-person differences are real, but a single measurement of one meal is not a reliable estimate of a person’s response. Personal advice needs repeated measurements, aggregated, before it can be trusted.

Does acting on it help?

Variation on its own does not prove that personal advice works. In 2024 the same group reported a randomised trial of 347 adults in the United States, comparing an 18 week programme of personalised food scores against standard dietary guidance [15]. Triglycerides fell modestly more in the personalised arm, by 0.13 mmol/L, and body weight, waist circumference and HbA1c improved among those who followed the programme closely. LDL cholesterol, blood pressure, fasting glucose and insulin did not differ between arms. The trial was funded by the company selling the programme, which is a reason for caution, and the effects were small. It is the first controlled evidence that measuring the individual response can change the outcome, and it shows how far the field still is from large effects.

Taken together, these studies establish two things. Biological individuality in the response to food is real, large and measurable. And the part of it that comes from genes is small compared with the part that comes from the microbiome, behaviour and physiology. Both matter for what follows.

Part IV. Population genetics and the portability problem

A second limit comes from genetics itself. Most of what we know about the genetics of common disease was learned in people of European ancestry, and the tools built from that knowledge travel less well than the knowledge itself.

Polygenic scores

A polygenic score adds up the small effects of many variants into one estimate of a person’s genetic risk. For coronary disease, type 2 diabetes and several cancers, people in the top few percent of the score carry risk comparable to a rare monogenic mutation, and the scores are now being prepared for clinical use [16].

The difficulty is where the scores were trained. In 2019 Martin and colleagues showed that scores built from European ancestry studies were several times less accurate in people of African ancestry, and warned that clinical use in that state would widen health disparities rather than narrow them [17]. That warning has held. A 2024 evaluation across 14 conditions found that the best European trained score lost about half its performance in South Asian and East Asian ancestry groups and about 60 percent in African ancestry groups, and that scores selected for each ancestry recovered part of the loss [18]. The methods for transferring scores across populations are improving, but they cannot replace discovery data from the populations themselves [19, 20].

Ancestry is a continuum

The problem is not confined to labelled groups. In 2023 Ding and colleagues showed that the accuracy of a polygenic score falls smoothly with a person’s genetic distance from the training sample, even inside a population usually treated as homogeneous [21]. Across 84 traits the correlation between genetic distance and accuracy was close to minus one. Every person is at some distance from the reference, and the score is a little less right for each of them in turn.

This does not mean the genetics is wrong. It means the estimate was made in one context and applied in another, and the transfer costs accuracy. The lesson is the same one as in Part III. A measurement made on a population describes that population; applying it to a person requires knowing how far that person is from the group the measurement came from.

Part V. The older answers

Long before a genome could be read, physicians in several civilisations had noticed that people differ, and each built a system for sorting patients into types before deciding how to treat them. The systems disagree about mechanism. They agree on the premise, that the same illness in two different constitutions calls for two different responses. Three of them are still practised at scale and still generate research.

Persian medicine and Mizaj

In Persian medicine the central concept is Mizaj, usually translated as temperament. A person’s Mizaj is judged from physique, body composition, skin, sleep, appetite, tolerance of heat and cold, physical and psychological responsiveness, and it determines diagnosis, diet, prevention and treatment. Contemporary authors present it explicitly as a personalised approach [22].

The empirical literature is thin. A 2020 review of the English language studies on Mizaj found 32 papers in total, of which 14 were reviews that largely restated classical texts, and the human studies used no common method for assessing temperament [23]. Since then there have been small association studies, for example between temperament and polymorphisms in antioxidant genes in a few hundred people, without replication [24]. Even the 2026 review that argues most strongly for integrating Persian medicine into modern personalised care concedes that the evidence base remains nascent [22].

What science can say is limited and worth stating plainly. Persian medicine developed a systematic, individualised framework, and standard instruments for measuring Mizaj now exist. Whether the categories correspond to anything measurable in biology has not been tested at scale.

Ayurveda and Prakriti

Ayurveda classifies people by Prakriti, a constitution assigned at birth from physical, physiological and psychological traits and grouped under three doshas, Vata, Pitta and Kapha. This is the tradition where the question has been pursued furthest, under the name Ayurgenomics [25].

The best known study screened 3,416 men, selected 262 with unambiguous Prakriti, and ran a genome-wide scan. It reported 52 variants that differed between the three groups after correction for ancestry, and one gene, PGM1, involved in glucose metabolism, that the authors linked to the classical description of Pitta [26]. The sample was small by the standards of modern genetics, the variants did not reach conventional genome-wide significance, and the finding has not been independently replicated. More recent work is of the same scale, for example a 2026 case control study of coronary disease variants stratified by Prakriti in an Indian cohort [27]. Prakriti may be a useful phenotypic framework. That it has a genetic basis remains a hypothesis.

Chinese medicine and syndrome differentiation

Traditional Chinese medicine classifies not the person but the presentation. Syndrome differentiation, or Zheng, groups a patient’s signs into a pattern, and two patients with the same biomedical diagnosis receive different treatment if their patterns differ. In form this is stratified medicine. Research programmes have applied metabolomics and other omics to ask whether Zheng patterns have biochemical correlates [28].

The evidence for the treatments themselves is weaker than the volume of publications suggests. An analysis of all Cochrane systematic reviews of traditional Chinese medicine up to 2020 found that the certainty of evidence was mostly low or very low, and that half the reviews were themselves of low methodological quality [29]. That is a statement about trials of interventions, not about whether the classification captures real variation, but it is a reason to keep the two questions apart.

Part VI. One question in several languages

Set side by side, the older and newer vocabularies line up more closely than either side usually admits.

Traditional systems	Precision medicine
Constitution, temperament	Phenotype
Mizaj, Prakriti	Multidimensional phenotype, subtype
Syndrome differentiation	Clinical stratification
Individualised diet	Precision nutrition
Assessment by observation over time	Longitudinal deep phenotyping
Treatment adjusted to the type	Stratified intervention

This is a comparison of concepts, not evidence that the concepts are biologically equivalent. A temperament assigned by a physician from appearance and habit and a subtype assigned by a model from glucose curves and genotypes may coincide, or may not; the point is only that both are attempts to answer the same question. Why do people with the same illness respond differently, and what should be done about it?

Discussion

Five points follow from the evidence reviewed above.

Human variation is real and large. It is visible in the genome, in the response to the same meal, in the effect of the same drug, and in the accuracy of the same risk score applied to different people. This is no longer in question.

Population science is not the problem. Every result cited in this essay came from a large study of many people. Without population data there is nothing to personalise. The task is to read the variation inside the population estimate, not to discard the estimate.

The older systems recognised the variation. Persian medicine, Ayurveda and Chinese medicine each built a classification of people or presentations and adjusted treatment to it. They did so by observation, over centuries, without instruments, and the fact that three unrelated traditions arrived at the same premise is itself worth noting.

Recognition is not validation. A classification can be internally consistent, widely used and old, and still fail to correspond to anything in biology. The studies that have tried to test the correspondence are small, unreplicated and, for the treatments, of low certainty. The honest position is that the question is open.

Precision medicine is itself unfinished. Its scores travel poorly across ancestries, its measurements are noisier than the marketing implies, its trials show small effects, and it has barely begun to integrate the environment. It is not a finished standard against which older systems can be judged; it is a younger attempt at the same problem.

What the evidence supports

Established. People differ substantially in their response to food. Genetic variation contributes to disease risk and drug response. Environment and behaviour change outcomes on the scale of the strongest drugs. Prediction accuracy falls with genetic distance from the population a tool was built in.

Emerging. Precision nutrition guided by measured responses, with small effects in one industry funded trial. Foundation models of continuous glucose data. Microbiome informed personalisation. Genomic correlates of Prakriti, from one unreplicated study.

Not established. That Mizaj or Prakriti corresponds to a genotype or to any measured biological signature. That traditional constitutional systems have been validated. That traditional treatments outperform, or match, evidence based ones. That every person requires a wholly unique treatment rather than membership of a well defined subgroup.

What would settle it

The question can be tested, and the design is not exotic.

Measure the classification properly. Standardised instruments for Mizaj and Prakriti exist. Apply them in a cohort that already has genotypes, metabolomics, microbiome, continuous glucose monitoring and anthropometry, and ask whether the traditional categories predict anything the modern measurements do not, or align with subtypes the modern measurements find on their own.

Test it across populations. Persian medicine, Ayurveda and Chinese medicine arose in particular populations. Whether their categories describe something universal or something local is an empirical question, and the answer bears directly on the portability problem of Part IV.

Use the tools that already exist. Unsupervised clustering of deep phenotypes is routine in metabolic research; it is how insulin resistance subtypes were found. Running the same methods with a traditional classification as one input, and seeing whether it adds information, would cost little and answer much.

Keep the two questions apart. Whether a classification captures real variation, and whether the treatments attached to it work, are different questions with different study designs. Conflating them has damaged the literature in both directions.

Conclusion

The question in this essay is not whether traditional medicine was right and modern medicine wrong, nor whether genomics can translate ancient concepts into DNA. It is narrower. Across civilisations and centuries, physicians met the same observation, that people differ in how they fall ill and how they respond, and each tradition built a language for it. Mizaj, Prakriti and syndrome differentiation are one set of answers. Genetics, physiology, the microbiome and the exposome are another. They are not describing the same mechanisms, and they may not be describing the same groupings. They are answering the same question.

For the first time, the tools exist to test whether the older answers contain anything the newer ones have missed. Most likely some do and most do not. Finding out would settle a very old argument on evidence rather than on loyalty, and it would help with the tension that every clinician still faces, between what is known about populations and what is true of the person in front of them.

The technology of personalised medicine is new. The problem of the individual is not.

References

1. US Food and Drug Administration. Precision medicine. fda.gov/medical-devices/in-vitro-diagnostics/precision-medicine
2. Relling MV, Evans WE. Pharmacogenomics in the clinic. Nature 2015;526:343-350. doi:10.1038/nature15817
3. US Food and Drug Administration. Table of pharmacogenetic associations. fda.gov/medical-devices/precision-medicine/table-pharmacogenetic-associations
4. 1000 Genomes Project Consortium. A global reference for human genetic variation. Nature 2015;526:68-74. doi:10.1038/nature15393
5. Polderman TJC, Benyamin B, de Leeuw CA, et al. Meta-analysis of the heritability of human traits based on fifty years of twin studies. Nature Genetics 2015;47:702-709. doi:10.1038/ng.3285
6. Suleman S, Madsen AL, Ängquist L, et al. Genetic underpinnings of fasting and oral glucose-stimulated based insulin sensitivity indices. Journal of Clinical Endocrinology and Metabolism 2024;109:2754-2763. doi:10.1210/clinem/dgae275
7. Suleman S, Huang Y, Jensen RT, et al. Adult-based genetic risk scores for insulin resistance associate with cardiometabolic traits in children and adolescents. Journal of Clinical Endocrinology and Metabolism 2025;110:2645-2654. doi:10.1210/clinem/dgae882
8. GBD 2019 Risk Factors Collaborators. Global burden of 87 risk factors in 204 countries and territories, 1990-2019. Lancet 2020;396:1223-1249. doi:10.1016/S0140-6736(20)30752-2
9. Knowler WC, Barrett-Connor E, Fowler SE, et al. Reduction in the incidence of type 2 diabetes with lifestyle intervention or metformin. New England Journal of Medicine 2002;346:393-403. doi:10.1056/NEJMoa012512
10. Rothschild D, Weissbrod O, Barkan E, et al. Environment dominates over host genetics in shaping human gut microbiota. Nature 2018;555:210-215. doi:10.1038/nature25973
11. Reicher L, Shilo S, Godneva A, et al. Deep phenotyping of health-disease continuum in the Human Phenotype Project. Nature Medicine 2025;31:3191-3203. doi:10.1038/s41591-025-03790-9
12. Lutsker G, Sapir G, Shilo S, et al. A foundation model for continuous glucose monitoring data. Nature 2026;650:978-986. doi:10.1038/s41586-025-09925-9
13. Hengist A, Ong JA, McNeel K, Guo J, Hall KD. Imprecision nutrition? Intraindividual variability of glucose responses to duplicate presented meals in adults without diabetes. American Journal of Clinical Nutrition 2025;121:74-82. doi:10.1016/j.ajcnut.2024.10.007
14. Berry SE, Valdes AM, Drew DA, et al. Human postprandial responses to food and potential for precision nutrition. Nature Medicine 2020;26:964-973. doi:10.1038/s41591-020-0934-0
15. Bermingham KM, Linenberg I, Polidori L, et al. Effects of a personalized nutrition program on cardiometabolic health: a randomized controlled trial. Nature Medicine 2024;30:1888-1897. doi:10.1038/s41591-024-02951-6
16. Lennon NJ, Kottyan LC, Kachulis C, et al. Selection, optimization and validation of ten chronic disease polygenic risk scores for clinical implementation in diverse US populations. Nature Medicine 2024;30:480-487. doi:10.1038/s41591-024-02796-z
17. Martin AR, Kanai M, Kamatani Y, et al. Clinical use of current polygenic risk scores may exacerbate health disparities. Nature Genetics 2019;51:584-591. doi:10.1038/s41588-019-0379-x
18. Moreno-Grau S, Vernekar M, Lopez-Pineda A, et al. Polygenic risk score portability for common diseases across genetically diverse populations. Human Genomics 2024;18:93. doi:10.1186/s40246-024-00664-y
19. Kachuri L, Chatterjee N, Hirbo J, et al. Principles and methods for transferring polygenic risk scores across global populations. Nature Reviews Genetics 2024;25:8-25. doi:10.1038/s41576-023-00637-2
20. Fatumo S, Chikowore T, Choudhury A, et al. A roadmap to increase diversity in genomic studies. Nature Medicine 2022;28:243-250. doi:10.1038/s41591-021-01672-4
21. Ding Y, Hou K, Xu Z, et al. Polygenic scoring accuracy varies across the genetic ancestry continuum. Nature 2023;618:774-781. doi:10.1038/s41586-023-06079-4
22. Vazani Y, Daneshfard B, Nimrouzi M. Toward a holistic model of personalized and integrative medicine: the distinctive role of Persian medicine. Journal of Ayurveda and Integrative Medicine 2026;17:101287. doi:10.1016/j.jaim.2025.101287
23. Akhtari M, Moeini R, Mojahedi M, Gorji N. Assessment the studies on the concept of Mizaj (temperament) in Persian Medicine. Journal of Complementary and Integrative Medicine 2020;17(3). doi:10.1515/jcim-2018-0122
24. Zendehboodi Z, Saberikia Z. Association of temperament with genetic polymorphisms in SOD1, GSTM1 and GSTT1 genes. Molecular Biology Research Communications 2021;10:33-38. doi:10.22099/mbrc.2020.38820.1565
25. Mukerji M. Ayurgenomics-based frameworks in precision and integrative medicine: translational opportunities. Cambridge Prisms: Precision Medicine 2023;1:e29. doi:10.1017/pcm.2023.15
26. Govindaraj P, Nizamuddin S, Sharath A, et al. Genome-wide analysis correlates Ayurveda Prakriti. Scientific Reports 2015;5:15786. doi:10.1038/srep15786
27. Dua P, Punera DS, Dholakia D, et al. Association of coronary artery disease related single nucleotide polymorphisms with extreme Prakriti types: insights from a case control study. Journal of Ayurveda and Integrative Medicine 2026;17:101371. doi:10.1016/j.jaim.2026.101371
28. Jiang M, Zhang C, Zheng G, et al. Traditional Chinese medicine Zheng in the era of evidence-based medicine: a literature analysis. Evidence-Based Complementary and Alternative Medicine 2012;2012:409568. doi:10.1155/2012/409568
29. Dai Z, Liao X, Wieland LS, et al. Cochrane systematic reviews on traditional Chinese medicine: what matters, the quantity or quality of evidence? Phytomedicine 2022;98:153921. doi:10.1016/j.phymed.2021.153921