There is a clinic in NEOM where I once sat across from a Saudi physician who had been practicing for twenty-three years. We were going over the lab results of a patient in her early thirties. Her fasting glucose was technically normal. Her HbA1c was a few points under the diabetic threshold. She had been told, repeatedly and by several doctors, that she was fine.
The physician slid the page toward me and said something I have not stopped thinking about since.
"She is not fine. I know it. I cannot prove it. The numbers were not built for her."
That sentence is the entire problem. This essay is about why he could not prove it, what would have to exist for him to prove it, and why a small team could begin building it tomorrow.
The problem, stated plainly
Type 2 diabetes is a generational catastrophe in the Gulf. Saudi Arabia's prevalence rose from 8.5% in 1992 to a modeled 39.5% in 2022. Among young Emirati women aged 17 to 25, an HbA1c-based study found 24% with pre-diabetes and 8.6% with diabetes. The same study found that fasting plasma glucose, the test most clinicians still default to, missed most of these cases. Concordance between FPG and HbA1c was kappa 0.287, which in plain language means the two tests disagree more often than they agree.
This is not a failure of awareness. It is a failure of measurement.
The diagnostic instruments we use in the Arab world were built somewhere else, on someone else's body. The reference ranges for fasting glucose, the cut-offs for HOMA-IR, the canonical genetic risk loci, the predictive equations for cardiovascular and metabolic risk. All of them were developed in cohorts that were overwhelmingly European, overwhelmingly male, and entirely outbred. As of the most recent audit, 88% of all genome-wide association study participants were of European ancestry. Arabs, who represent 5% of the world's population and carry one of the highest diabetes burdens on earth, account for fewer than 1% of GWAS participants.
When researchers tested how well European-derived metabolic risk variants generalized to Arab populations, the answer was brutal. Of 4,746 GWAS-identified metabolic risk variants, only 304, roughly 6%, replicated in Arab cohorts. Ninety-four percent of the genetic risk model we use to predict who will develop metabolic disease in this region was built on populations whose biology does not match the people we are trying to help.
The clinical consequence is that physicians like the one I sat with in NEOM are working with instruments calibrated for someone else's patient. They know it. The patient feels it. And the science continues to act as if the gap is a matter of more recruitment, not a matter of methodology.
Why this is technically hard, not just underfunded
It would be easy to write that the solution is to recruit more Arab women into existing studies. That is not the solution. That framing is what has produced two decades of marginal progress.
The technical difficulty has three layers, and each one is a real R&D problem.
The first is consanguinity. Roughly half of marriages in Saudi Arabia, Kuwait, and Qatar are between blood relatives, often first or second cousins. This is not a footnote. It changes the math of human genetics. The standard tools of population biology, from Hardy-Weinberg equilibrium to allele frequency estimation to polygenic risk score construction, all assume random mating. They do not assume anything resembling the pedigree structures of the Arabian Peninsula. In consanguineous populations you see elevated autozygosity, accumulation of homozygous loss-of-function variants, and the surfacing of recessive disease alleles that would remain hidden in outbred populations. This is a scientific opportunity disguised as a methodological problem. Homozygous human knockouts, as some geneticists have noted, are nature's clinical trial. But to read them, the model has to know they exist.
The second is the female phenotype. Most metabolic research in the GCC has been done on men, or on mixed cohorts where sex was treated as an adjustment variable rather than a biological axis. But women metabolize differently. Insulin resistance in young women often presents through the hyperandrogen pathway, with symptoms that get dismissed as cosmetic, like hair, acne, irregular cycles, rather than read as the metabolic warning they are. The HOMA-IR cut-off most labs in the region use is 2.0, derived from European studies. Korean data shows the premenopausal cut-off is closer to 2.39 and the postmenopausal cut-off 2.48. There is no published Arab female-specific HOMA-IR cut-off. None. We are diagnosing insulin resistance in millions of women in this region with a number that was not derived from any of them.
The third is the environmental and behavioral substrate that no European cohort experiences. Eighty-one percent of pregnant Saudi women are vitamin D deficient. The published predictors of that deficiency include indoor activity, full-body clothing, and parity. Variables that simply do not appear in the statistical models built around women in Stockholm or Boston. Vitamin D deficiency is not just a bone story. It interacts with insulin sensitivity, with inflammatory markers, with the entire endocrine cascade that governs whether a young woman moves toward metabolic health or metabolic disease. Add Ramadan, which restructures the circadian metabolic rhythm of an entire population for a month every year, and you have a biology that the field has never properly characterized because the field has never been in the room.
Three layers, three open scientific questions, no current dataset that addresses any of them in a population of Arab women with the depth required.
What I would build, and how a small team would start.
I would build the Arab Female Biomarker Atlas. Not a biobank. Those exist, in the form of the Qatar Biobank and the Saudi Genome Project, and they have done valuable work. The Atlas would be something narrower and sharper. A longitudinal phenotyping cohort designed from the first day around the three layers above.
Here is what it would actually contain.
A cohort of 5,000 Arab women, ages 18 to 55, recruited across at least three GCC countries to capture genetic and environmental variation. The cohort would be deliberately enriched for consanguinity. Half the participants would come from families with documented first-cousin or second-cousin marriages going back two generations, and pedigree information would be collected as primary data, not as metadata.
A biomarker panel that is wider than what is currently standard and built around the female metabolic phenotype. That means fasting glucose, fasting insulin, HbA1c, HOMA-IR, full lipid panel, hs-CRP, 25(OH)D, the full thyroid axis, the full sex hormone panel including SHBG and free androgen index, AMH, and a curated set of adipokines such as adiponectin and leptin alongside emerging markers of subclinical insulin resistance. Each biomarker measured at four points across a menstrual cycle for a subset of the cohort, because the assumption that hormonal variability is noise to be averaged out is itself a methodological error.
Whole-exome sequencing on the full cohort, with explicit modeling of runs of homozygosity rather than treating them as confounders. This is where the consanguinity-aware methodology becomes a feature rather than a problem. Standard polygenic risk scores collapse under autozygosity. New scores can be built that use it.
Lifestyle and environmental phenotyping that takes the regional context seriously. Sun exposure quantified through wearable UV sensors rather than self-report. Ramadan fasting captured as a structured exposure variable. Veiling status, indoor versus outdoor activity hours, dietary patterns specific to the region.
A multimodal model architecture that integrates these layers. The model would be trained to predict early metabolic dysfunction, including insulin resistance crossing a population-derived threshold, glucose dysregulation, and the early endocrine shifts that precede clinical diagnosis by years, using all of the above as inputs. Crucially, it would be designed to be recalibrated rather than retrained when applied to new sub-populations within the Arab world, because Lebanese, Egyptian, and Khaleeji women are not interchangeable.
A small team could start tomorrow. The minimum viable team is six people. A population geneticist with experience in consanguineous cohorts. An endocrinologist with strong female metabolic expertise. A biostatistician comfortable with autozygosity modeling. A clinical research coordinator embedded in a GCC tertiary hospital. A machine learning engineer who has worked on healthcare data in stratified populations. And a project lead who can navigate both the science and the cultural realities of recruiting Arab women into a study that asks them to talk about their bodies.
The first six months would be protocol design and ethical approvals. The next twelve months would be a pilot of 500 women in one country to validate the protocol and produce the first Arab female-specific HOMA-IR cut-off, the first Arab female-specific vitamin D-insulin axis characterization, and the first preliminary catalog of homozygous variants associated with early metabolic dysfunction in this population. After that, expansion.
Why this matters beyond the cohort
If this is built, the immediate clinical consequence is that physicians like the one I sat with in NEOM stop having to apologize to their patients for instruments they know are wrong. They get a panel calibrated to the woman in front of them. They get a risk score that does not fall apart when she tells them her parents are first cousins. They get years of warning before HbA1c crosses the diabetic threshold, because the model is reading the upstream signals the current system is blind to.
The broader consequence is scientific. Arab women, because of consanguinity, carry a higher density of homozygous loss-of-function variants than almost any other population on earth. Each of those is a natural experiment in human biology that the rest of the world cannot run. The Atlas would not just serve the women in it. It would feed back into global metabolic science as one of the most informative cohorts on the planet for understanding gene function, for the same reason Iceland and Finland did at smaller scales.
A region that has been treated as a downstream consumer of biomedical knowledge would become an upstream producer of it. That reversal is not a slogan. It is what the data, properly collected, would force.
A closing note:
The physician in NEOM did not know my name when we first met, and I did not know his. We had been put in a room together because a senior official thought it might be useful. The conversation lasted forty minutes. I have thought about it for two years.
What he was telling me, between the lines, was that he had spent his entire career watching the science arrive late for his patients. Not absent. Late. The papers were always coming, the guidelines were always updating, the cut-offs were always being revised somewhere else for someone else, and by the time any of it reached his clinic the women in front of him had already lived through the years where it would have mattered.
What I am proposing is the smallest serious version of catching up. It is not the only thing that needs to happen for science and technology in the Arab world. It is one specific, technically tractable, ethically necessary thing. A small team could start it tomorrow. The cohort exists. The methods exist in pieces, scattered across consanguinity genetics and female endocrinology and sub-population machine learning, waiting for someone to put them in one room.
The women in the data are already here. They have been here the whole time. We have just not been measuring them.