Type 2 Diabetes and a Polygenic Risk Score: What the Research Actually Shows

Type 2 Diabetes and a Polygenic Risk Score: What the Research Actually Shows

This article is for educational purposes only. It is not a substitute for advice from a licensed physician, board-certified medical geneticist, or board-certified genetic counselor. For any decisions about testing, treatment, or care, consult a qualified clinician. In emergencies, call 911.

Bottom line: A consumer “higher likelihood” result for type 2 diabetes is a population probability, not a diagnosis and not a carrier result. A carrier is a person who holds one changed copy of a single gene, and this score reports nothing of that kind. The largest study of its kind, published in Nature in 2024, pooled 2,535,601 people, 428,452 of them with type 2 diabetes. It found 1,289 signals at 611 loci, and the average signal shifts the odds by about 3 percent. The US body that speaks for medical geneticists is blunt: “there are currently no clinical guidelines available for the use of this technology”. Yet one number has already moved outcomes in a trial, because lifestyle change cut the step from prediabetes to diabetes by 58 percent over 2.8 years.

What you’ll learn

• Why 611 loci and 1,289 signals make Punnett-square math impossible here, and what one copy of the strongest variant is worth
• The honest added value of a gene score once family history and body weight are known
• The one genetic question that can change a prescription, and the red flags that raise it
• What is approved in the US and Canada as of 2026, with dates, and what is not approved at all

Polygenic is not a smaller version of Mendelian: what TCF7L2, CDKAL1 and 611 loci actually mean

Most reports in a consumer DNA account run on one gene and two copies. That is Mendelian inheritance, which means a single gene drives the trait, so copies can be counted and a clean percentage falls out. Cystic fibrosis, sickle cell and Tay-Sachs all work that way. Type 2 diabetes does not, and the gap is not a matter of degree.

Type 2 diabetes is polygenic. That word means many common gene variants each add a small push, and no single one of them is “the gene”.

Question	Single-gene (Mendelian) condition	Type 2 diabetes (polygenic)
How many genes are in play?	one	611 loci, holding 1,289 signals
How much does one variant do?	enough to cause the disease	shifts the odds by about 3 percent on average
Is there a carrier state?	yes	no
Can a number be worked out for a child?	often yes	no, and a score “cannot be used to predict the relative disease risk for other family members”
What does the family tree show?	a clear pattern	“does not have a clear pattern of inheritance”

That last quote comes from the NIH consumer genetics resource, which adds that many affected people do have a parent or sibling with the condition. The same page notes that type 2 diabetes accounts for 90 to 95 percent of all diabetes.

The scale of the map, and the size of each piece

In March 2024, a study in Nature pooled genome-wide data from 2,535,601 people, 39.7 percent of them not of European ancestry, including 428,452 cases of type 2 diabetes. The authors report “1,289 independent association signals at genome-wide significance (P < 5 x 10-8) that map to 611 loci”, of which 145 were new. A locus is simply a spot on a chromosome.

Now the part that gets left out. Those 145 new signals were “predominantly common” variants with odds ratios “lower than 1.05”. An odds ratio compares two groups: 1.00 means no change at all, and 1.05 means the odds are 5 percent higher. Across the new groupings, the mean per-allele odds ratio was about 1.03. Think of a choir of a thousand voices. Take one voice out and the song does not change.

What the strongest single variant is worth

The strongest common signal is rs7903146, inside the TCF7L2 gene. The NHGRI-EBI GWAS Catalog is a public database run by the US National Human Genome Research Institute and EMBL-EBI. It holds 45 type 2 diabetes entries for this variant that carry a per-copy odds ratio. The median across those studies is 1.38, with a range from 1.15 to 1.71. The at-risk version of that spot is common, at a frequency near 0.30 in most European-ancestry entries.

Turn that into people. About 11.6 percent of the US population had diabetes in 2021. Applying a per-copy odds ratio of 1.38 to that baseline shifts roughly 12 people in 100 to roughly 15 in 100. Applying the far more typical 1.03 leaves 12 in 100 at about 12 in 100. Neither number is a personal estimate. Neither one moves a reader from “unlikely” to “likely”.

The 2006 discovery paper matters for a different reason. It reported relative risks of 1.45 and 2.41 for people with one or two at-risk copies at a marker called DG10S478. A relative risk of 1.45 means the risk is 45 percent higher than in the comparison group, not that the risk is 45 percent. That “corresponds to a population attributable risk of 21%”, meaning about a fifth of cases in the population trace back to it. Even the strongest common signal known is not one person’s fate.

The second variant readers see named is rs7754840, near CDKAL1. Its per-copy odds ratio was 1.20 in people of European ancestry and 1.25 in a Han Chinese group in Hong Kong. The GWAS Catalog median across later studies is a little lower, at 1.15. Its mechanism is specific. Insulin response in people with two copies was “approximately 20% lower than for heterozygotes or noncarriers”, that is, than in people with one copy or none.

None of this produces a carrier state, a 25 percent figure, or a gene to name at a family dinner. Anyone who wants a number that applies to their own body should ask a primary care physician for a blood test, not a database.

Section recap: Type 2 diabetes risk is spread across 1,289 signals at 611 loci, with an average effect near 3 percent per copy. The strongest single variant sits near a 1.38 per-copy odds ratio and covers about a fifth of population risk, so no carrier status or transmission percentage exists.

One disease name, eight biological routes: the 2024 clustering result

The 2024 analysis did not stop at counting. It sorted the 1,289 signals into eight separate clusters, each one a different route into the same diagnosis.

The paper names them and gives the count for each. “+PI” and “-PI” mark whether proinsulin, the raw form the body turns into insulin, runs high or low in that cluster.

Cluster	Signals
Residual glycaemic	389
Body fat	273
Obesity	233
Metabolic syndrome	166
Beta cell +PI	91
Beta cell -PI	89
Lipodystrophy	45
Liver and lipid metabolism	3

Example loci are given for each. TCF7L2, KCNQ1 and CDKAL1 sit in Beta cell +PI. FTO and MC4R sit in Obesity, and PPARG and IRS1 sit in Lipodystrophy.

These clusters “are differentially enriched for cell-type-specific regions of open chromatin, including pancreatic islets, adipocytes, endothelial cells and enteroendocrine cells”. In plain terms, different clusters switch on in different tissues.

Why this matters at a family table

A father found to have diabetes at 52 with a thin build, and a cousin found at 60 with severe obesity, may carry quite different genetic loads. One profile may lean toward beta-cell signals, another toward body-fat signals. The single label “type 2 diabetes” hides that. Two people can reach the same address by different highways.

The trial evidence lines up with this. In the Diabetes Prevention Program gene analysis, the TCF7L2 high-risk genotype “was associated with decreased insulin secretion but not increased insulin resistance at baseline”. That is the beta-cell route, measured in living trial participants.

The honest limits

The largest cluster is the least clear. Residual glycaemic holds 389 of the 1,289 signals, and its profile is only higher fasting glucose and higher A1C. A1C is a blood test that shows average blood sugar over about three months. The two beta-cell clusters together hold just 180 signals. Most of the map does not yet have a clean story behind it.

The authors also built cluster-specific scores in a further 279,552 people of diverse ancestry, including 30,288 cases. Those scores were linked to coronary artery disease, peripheral artery disease and end-stage diabetic kidney disease across ancestry groups. That is a research result. Split-out scores are not what a consumer report returns, and no guideline uses them to pick a drug or a screening interval. A reader curious about a family pattern should bring it to a clinician, who can look at ages, body types and test results together.

Section recap: The 1,289 signals resolve into eight clusters with distinct tissue biology, which is why one family label can hide several different diseases. Cluster scores predict vascular complications in research cohorts but are not part of any consumer report or clinical pathway.

How much risk is this, in numbers you can act on

Relative language is where competing pages stop. Absolute language is what a reader needs.

The population baseline

The National Institute of Diabetes and Digestive and Kidney Diseases at NIH publishes the national figures, credited there to the CDC for data year 2021. In that year 38.4 million people of all ages had diabetes, or 11.6 percent of the population. Of those, 29.7 million were diagnosed. Another 8.7 million adults had diabetes and did not know it, which is 22.8 percent of adults with the condition. And 97.6 million adults, more than one in three, had prediabetes.

How common diabetes is differs sharply by group. These are age-adjusted rates of diagnosed diabetes for 2019 to 2021.

Group	Diagnosed diabetes
American Indian and Alaska Native adults	13.6%
Non-Hispanic Black adults	12.1%
Adults of Hispanic origin	11.7%
Non-Hispanic Asian adults	9.1%
Non-Hispanic White adults	6.9%

Those gaps reflect access, screening and living conditions rather than any built-in biological difference.

For lifetime risk, the reference point is a 2003 estimate written at the CDC for Americans born in 2000: 32.8 percent for males and 38.5 percent for females. It rests on survey data from 1984 to 2000, so read it as a dated figure rather than today’s.

What a parent’s diagnosis is worth

The Framingham Offspring Study examined parents directly, then followed 2,527 offspring from 1,303 families. Measured against offspring with no parental diabetes, the age-adjusted odds ratios were:

• Mother with diabetes: 3.4
• Father with diabetes: 3.5
• Both parents with diabetes: 6.1

Passing it down from the mother’s side and the father’s side came out about the same. In that cohort, 8.6 percent of offspring had diabetes overall. The authors describe their sample as “primarily Caucasian and at relatively low risk for diabetes”, which limits how far the numbers travel.

A father found to have diabetes at 52 already places a reader in roughly the 3.5-fold band, before any DNA is read.

What the score adds on top

This is the uncomfortable finding, and it is the reason to trust everything else here. Framingham researchers typed 18 diabetes-linked variants in 2,377 people and counted 255 new cases over 28 years. Each risk copy carried a sex-adjusted odds ratio of 1.12.

How much a score helps depends on what it is added to, so naming the model is a must. The measure below is the C statistic, sometimes called the area under the curve. It runs from 0.5, which is a coin toss, to 1.0, which is perfect sorting of who will and will not get the disease.

Model the gene score was added to	C statistic before	After
Sex alone	0.534	0.581
Sex plus self-reported family history	0.595	0.615
Full clinical model	0.900	0.901

The full clinical model held age, sex, family history, body-mass index, fasting glucose, systolic blood pressure, HDL cholesterol and triglycerides. Adding the gene score “resulted in the appropriate risk reclassification of, at most, 4% of the subjects”. In plain terms, it moved at most 4 people in 100 into a better-fitting risk group.

A British cohort found the same shape. In Whitehall II, 5,535 healthy people produced 302 cases over 10 years. A simple count of risk copies reached 0.54 on that same scale, and a weighted version 0.55. The Cambridge risk score, built from age, sex, drug treatment, family history, body-mass index and smoking, reached 0.72, and the Framingham offspring risk score reached 0.78. Genetics gave “a modest net reclassification improvement of about 5% when added to the Cambridge risk score but not when added to the Framingham offspring risk score”. Small, not zero.

The fair upper bound

A 2018 analysis is the strongest case for the other side. A genome-wide score picked out 3.5 percent of the population at more than threefold risk for type 2 diabetes. That is roughly one person in 29. Most people who see “higher likelihood” are not in that tail. They sit in a broad middle where the score changes very little. Only a clinician looking at a real glucose reading can say where a given reader stands.

Section recap: About 11.6 percent of Americans had diabetes in 2021, and nearly a quarter of adults with it were undiagnosed. A parent’s diagnosis carries roughly a 3.5-fold odds ratio, and a gene score added to a full clinical model moved the C statistic from 0.900 to 0.901.

What the consumer report is and is not

Readers assume every line in a DNA account carries the same regulatory weight. It does not, and the company says so itself.

The regulatory line, in the vendor’s own words

The clinician-facing documents draw a line inside the product list. Some health predisposition reports “meet US FDA requirements for genetic health risks”. The type 2 diabetes report, by contrast, “is based on 23andMe research and has not been reviewed by the FDA”. It is named on its own in that list, as the one report that has not been reviewed.

The same directory lists that report under “Powered by 23andMe Research”, with “1000+” variants behind it. Ethnicities listed as relevant are European, Hispanic/Latino, African, East Asian and South Asian. For scale, other reports on the same page rest on far bigger models, such as 19,000-plus for anxiety and 28,000-plus for insomnia. The company’s own disclaimer adds that the test “is not intended to diagnose any disease”. It is also “not a substitute for visits to a healthcare professional for recommended screenings or appropriate follow-up”.

Three very different things get called “a test”. They are not swappable.

Kind of test	What it looks at	Who orders it	What it can settle
Consumer type 2 diabetes report	1000-plus common variants, scored together	the customer	nothing on its own; the vendor states it “has not been reviewed by the FDA”
A1C or fasting glucose	blood sugar in the body now	a physician	whether a person has diabetes or prediabetes today
MODY gene panel	single genes such as GCK, HNF1A, HNF4A and HNF1B	a clinician, on a clinical-grade panel	a single-gene diagnosis that can change the prescription

The ancestry problem, and what causes it

Gene scores are built mostly from European-ancestry data, and they get less accurate outside that group. A 2019 analysis measured this directly. It covered 17 body and blood-panel traits in UK Biobank, using European-derived summary statistics. Accuracy was “1.6-fold lower in Hispanic/Latino Americans, 1.7-fold lower in South Asians, 2.5-fold lower in East Asians, and 4.9-fold lower in Africans on average”. Those figures come from those 17 traits, not from a type 2 diabetes score.

The cause is who got studied, not human difference. About 79 percent of all people in published genome-wide studies are of European descent, while that group is 16 percent of the world. The authors call the gap “an inescapable consequence of Eurocentric biases in genome-wide association studies”. They also note early progress, since efforts to widen the pool “show promise in leveling this vast imbalance”.

What the professional society says

In 2023, the American College of Medical Genetics and Genomics published a points-to-consider statement on polygenic risk scores, or PRS for short. Three sentences carry most of the weight.

• “there are currently no clinical guidelines available for the use of this technology”
• “PRS test results do not provide a diagnosis, instead they provide a statistical prediction of increased clinical risk”
• “there is currently limited evidence to support the use of PRS to guide medical management”

The statement also blocks the most dangerous misreading. “A low PRS does not rule out significant risk for the disease or condition in question”. A below-average result exempts nobody from screening.

That is not this site’s opinion. It is the society’s, and it is voluntary guidance rather than a rule. Not everyone in the field agrees. The 2018 team argued that “it is time to contemplate the inclusion of polygenic risk prediction in clinical care”. The disagreement is live.

One check settles the practical question. Search the full text of the 2026 American Diabetes Association Standards of Care, in the sections on diagnosis and on prevention. Across those two sections, “direct-to-consumer” and “23andMe” appear zero times. The word “polygenic” appears once, and only inside a reference title. So no US standard of care asks a physician to act on this report, which is worth saying out loud at the next appointment.

Section recap: The vendor states its type 2 diabetes report has not been reviewed by the FDA and is built from 1000-plus variants. European-derived scores lose accuracy elsewhere for reasons of data, not biology, and the ACMG says no clinical guidelines exist for their use.

The one genetic question worth escalating: ruling out MODY

There is a case where a genetic test changes a prescription. It is not the polygenic report.

Monogenic diabetes affects “a small fraction of people with diabetes (<5%)”, and unlike type 2 diabetes it comes from a single gene. The common forms are maturity-onset diabetes of the young, usually from variants in GCK, HNF1A, HNF4A or HNF1B. These are autosomal dominant, meaning one changed copy from either parent is enough. That inheritance language applies here and nowhere else in this article.

Why it gets missed

A UK study reviewed referrals for MODY testing from 1996 to 2009, covering 2,072 index patients and 1,280 relatives. MODY was confirmed in 35 percent of those tested, with HNF1A at 52 percent and GCK at 32 percent of confirmed cases. The lowest possible rate came to 108 cases per million. From that the authors concluded that “more than 80% of MODY is not diagnosed by molecular testing” in the UK. Confirmed-case rates ran from 5.3 per million in Northern Ireland to 48.9 per million in South West England. The authors put that gap down to referral patterns and access rather than biology.

US data tell the same story in children. A population-based study read three MODY genes in 586 youth chosen for being diabetes-autoantibody negative with fasting C-peptide of 0.8 ng/mL or higher. Mutations turned up in 47 of them, 8.0 percent of that pre-selected sample, giving “a prevalence of at least 1.2% in the pediatric diabetes population”. Only three of the 47 carried a clinical diagnosis of MODY, and most were on insulin.

The payoff, and why it needs a clinician

The treatment stakes are concrete. Standards of care describe HNF1A-MODY and HNF4A-MODY as “sensitive to sulfonylureas”, a class of older diabetes tablets. They describe GCK-MODY as a form that “typically does not require treatment in nonpregnant individuals” outside pregnancy. A person can move from daily insulin to a low-dose tablet, or to no medicine at all.

That switch is a clinical decision, never a self-service one. No one should change or stop a diabetes medicine on the strength of a test result read at home, or of anything read online. Only the treating clinician can make that change safely, and stopping insulin without one can be dangerous.

Several features raise the question in the first place.

• An A1C below 7.5 percent at diagnosis
• One parent with diabetes
• Signs of a specific single-gene cause, such as renal cysts or partial lipodystrophy
• A probability above 5 percent on the MODY prediction calculator at diabetesgenes.org

Mild, stable fasting glucose fits the same picture. Because these forms are autosomal dominant, diabetes in several generations in a row fits the pattern too. That follows from the inheritance rule above, not from the guideline’s own list.

Read that list as a prompt, not a verdict. In the US youth study, “no single characteristic identified all patients with mutations”, and parental history did not separate the groups. A diagnosis “can only be confirmed by molecular genetic testing” ordered through a clinician, on a clinical-grade panel, never on a consumer array.

A board-certified genetic counselor is the right person to sort this out. Genetic counseling is a session with a trained counselor. They map the family history, explain what a test can and cannot show, and help a person decide what to do next. The National Society of Genetic Counselors runs a directory of over 3,300 counselors in the US and Canada. It offers in-person and telehealth options, at https://www.nsgc.org/findageneticcounselor. Canadian readers can use the Canadian Association of Genetic Counsellors site at genetic-counsellors.ca.

Section recap: Monogenic diabetes is under 5 percent of cases, yet more than 80 percent of UK MODY goes undiagnosed by molecular testing. In one US study, 44 of 47 youth with a confirmed mutation carried the wrong label. HNF1A-MODY responds to sulfonylureas and GCK-MODY often needs no treatment, which is why a clinician-ordered panel is worth asking about.

The test that actually decides this: screening pathways in the US and Canada

A probability ends at a blood draw. Here is what each country’s rules say, and they are not the same.

The US pathway

The US Preventive Services Task Force issued a B recommendation in 2021, and it is still current. It “recommends screening for prediabetes and type 2 diabetes in adults aged 35 to 70 years who have overweight or obesity”. The graded group is set by age and body-mass index only, at 25 or above for overweight and 30 or above for obesity. Family history is not part of that trigger. Under Affordable Care Act preventive-services rules, a B grade generally means no cost-sharing on most plans. That result follows from the grade, not from anything the Task Force itself states.

The American Diabetes Association casts a wider net. Its 2026 standards say testing “should be considered in adults with overweight or obesity” who have one or more listed risk factors. The cut point is a BMI of 25, “or >=23 kg/m2 in individuals of Asian ancestry”. The first item on the risk-factor list is “First-degree relative with diabetes”. For everyone else, “testing should begin at age 35 years”, repeated at least every three years. People with prediabetes “should be tested yearly”.

Many readers therefore qualified for screening long before they ever mailed a saliva tube.

The Canadian pathway is different

Diabetes Canada’s 2018 clinical practice guidelines recommend screening “every 3 years in individuals >=40 years of age or at high risk using a risk calculator”. High risk there means a 33 percent chance of getting diabetes over 10 years. The Public Health Agency of Canada’s CANRISK questionnaire fills that role, though it “has not been validated in individuals <40 years of age”. Diabetes Canada issues interim chapter updates, so a reader should check the live chapter for changes since 2018.

	United States	Canada
Start age for most adults	35	40
Screen earlier if	BMI is 25 or above, or 23 or above for people of Asian ancestry, plus a listed risk factor such as a first-degree relative with diabetes	a risk calculator such as CANRISK shows high risk, meaning a 33 percent chance over 10 years
How often	at least every 3 years, and yearly with prediabetes	every 3 years
Graded task-force advice	screen at ages 35 to 70 with overweight or obesity, grade B	risk-calculator route instead

The thresholds

The cut points are not the same either, and the gap is wider than most readers expect.

Blood test	Prediabetes, US	Prediabetes, Canada	Diabetes
A1C	5.7 to 6.4%	6.0 to 6.4%	6.5% or above in both countries
Fasting glucose	100 to 125 mg/dL	impaired fasting glucose	126 mg/dL or above in the US, 7.0 mmol/L or above in Canada
2-hour glucose	140 to 199 mg/dL	impaired glucose tolerance	200 mg/dL or above in the US, 11.1 mmol/L or above in Canada

Risk runs on a smooth slope across those ranges. Without clear symptoms, “diagnosis requires two abnormal results from different tests”, and Canada also asks for two tests. So a person with an A1C of 5.9 percent is labelled prediabetic in Ohio and not in Ontario, from the same blood draw.

Diabetes Canada built CANRISK for a reason worth borrowing. The guidelines note that “risk scores developed in Caucasian populations cannot be applied to populations of other ethnic groups”. Scoring systems “must be validated for each considered population”. That is the same portability argument made about polygenic scores, applied to a questionnaire.

The A1C trap that belongs in this category

A1C can read falsely low when a hemoglobin variant is present. A 2017 study covered 4,620 African American adults, 367 of them with sickle cell trait. It found that “for a given fasting glucose, HbA1c values were statistically significantly lower in those with (5.72%) vs those without (6.01%) SCT”. That is a mean gap of 0.29 percentage points. Measured against 2-hour glucose the gap was 0.30 points, and it grew at higher glucose levels.

That is roughly one person in 13 carrying the trait. Standards of care already tell clinicians to use “plasma glucose criteria” in several settings. Those include “some hemoglobin variants”, pregnancy, glucose-6-phosphate dehydrogenase deficiency, HIV and altered red cell turnover. The X-linked G6PD G202A variant, “carried by 11% of Black individuals in the U.S.”, is linked to an A1C about 0.8 percentage points lower in affected men and about 0.7 points lower in women who carry two copies. Anyone who knows they carry a hemoglobin or G6PD variant should ask a physician for a fasting glucose or a glucose tolerance test instead of leaning on A1C alone.

Section recap: US screening starts at 35, or earlier with a BMI of 25 and a first-degree relative, while Canada screens from 40 or by CANRISK score. Prediabetes begins at an A1C of 5.7 percent in the US and 6.0 percent in Canada, and hemoglobin variants can push A1C about 0.3 points too low.

The modifiable arm: lifestyle, metformin and the 2022-2026 incretin drugs

This is the part of the equation a reader controls, and it has the strongest evidence in the article.

The trial

The Diabetes Prevention Program placed 3,234 people with high fasting and post-load glucose into three groups at random. They got a dummy pill, metformin at 850 mg twice daily, or a lifestyle program. That program aimed at 7 percent weight loss or more and 150 minutes of activity a week. Mean age was 51, mean body-mass index 34.0, and 45 percent of those taking part were members of minority groups.

Over an average of 2.8 years, new cases ran at 11.0, 7.8 and 4.8 per 100 person-years in the dummy-pill, metformin and lifestyle groups. That is a 58 percent drop with lifestyle and 31 percent with metformin, and lifestyle beat the drug. The plainest number is the most useful: “To prevent one case of diabetes during a period of three years, 6.9 persons would have to participate in the lifestyle-intervention program”.

Every one of those percentages describes people who already had prediabetes, over 2.8 years. None of them describes a person with a high score and normal glucose.

What happened over fifteen and twenty-one years

Follow-up kept 2,776 of the surviving group, 88 percent, in view. Over a mean 15 years, new cases fell by 27 percent in the lifestyle group and 18 percent in the metformin group, “with declining between-group differences over time”. The absolute picture is sobering. By year 15, 55 percent of the lifestyle group, 56 percent of the metformin group and 62 percent of the dummy-pill group had diabetes. In a very high-risk group, prevention meant delay, not immunity.

Standards of care sum up the longer arc, reporting “34% reduction at 10 years, 27% reduction at 15 years, and 24% reduction at 21 years”. A Chinese trial reported 39 percent at 30 years, and a Finnish one 43 percent at 7 years. At 21 years, median diabetes-free survival was 3.5 years longer with lifestyle and 2.5 years longer with metformin. Even so, “the overall treatment effect was driven almost entirely by benefit seen during the initial DPP phase”. Major heart events did not differ at 21 years. Within the original trial, “every kilogram of weight loss” brought “a 16% reduction in risk of progression over 3.2 years”.

Those who avoided diabetes had 28 percent fewer small-vessel complications than those who developed it.

Does high genetic risk mean the effort is wasted?

No, and this is the single most reassuring result available. In the trial’s gene analysis of 3,548 people, those with two high-risk copies at rs7903146 reached diabetes faster than those with none, at a hazard ratio of 1.55. A hazard ratio compares the pace of events in two groups, so 1.55 means cases piled up about 55 percent faster.

Split by arm, that gap was 1.81 in the dummy-pill group, 1.62 in the metformin group and 1.15 in the lifestyle group.

State the caveat honestly. The interaction between genotype and treatment was not statistically significant. The trial did not prove that lifestyle erases genetic risk. What it shows is that the genotype penalty was smallest in the group doing the work.

Where to actually get the program

The CDC-recognized National Diabetes Prevention Program runs the same 16-session curriculum used in the trial. Entry needs a body-mass index in the overweight range plus lab testing, prior gestational diabetes, or a positive risk test. For people on Medicare, the Medicare Diabetes Prevention Program is covered under Part B as a preventive service. It runs 16 core sessions plus six monthly follow-ups, with a cap of 22 sessions. A 2026 rule change added coverage for online delivery that does not have to be live, through the end of 2029, and dropped the need for in-person sites. That benefit is Medicare-only, so an adult in their forties with commercial insurance should ask their plan about a CDC-recognized program instead.

The drugs, with regulator and date

Metformin remains the only drug with a prevention recommendation behind it. The 2026 standards give a Grade A recommendation that metformin “should be considered in adults at high risk of type 2 diabetes”. The group named is people aged 25 to 59 with a body-mass index of 35 or above. It also names fasting plasma glucose at or above 110 mg/dL, A1C at or above 6.0 percent, and prior gestational diabetes.

Two things are true at once. No FDA prevention indication exists for metformin, and an active Grade A society recommendation does. Long-term metformin also calls for periodic vitamin B12 checks.

For the newer drugs, dates come from the regulators’ own databases, checked in 2026.

Drug (brand)	Regulator and record	Approved for	Date
Tirzepatide (Mounjaro)	FDA, NDA215866	type 2 diabetes	13 May 2022
Tirzepatide (Mounjaro)	Health Canada, DIN 02532891, status “Approved”	listed in the Drug Product Database	status date 24 November 2022, which is a database status date and not a Notice of Compliance date
Tirzepatide (Zepbound)	FDA, NDA217806	weight management	8 November 2023
Semaglutide (Ozempic)	FDA, NDA209637	type 2 diabetes	5 December 2017

No product in either database carries an approved use for preventing type 2 diabetes.

The trial data explain why that gap matters. One 176-week trial enrolled 2,539 people with obesity, 1,032 of whom also had prediabetes. Mean weight fell by 12.3, 18.7 and 19.7 percent at the three tirzepatide doses, against 1.3 percent on the dummy pill. Diabetes was diagnosed in 1.3 percent of treated people versus 13.3 percent on the dummy pill.

Then the trial stopped the drug. After 17 weeks off treatment, 2.4 percent of the tirzepatide group and 13.7 percent of the dummy-pill group had type 2 diabetes. New cases in the treated group nearly doubled within four months of stopping. The trial was funded by Eli Lilly.

Standards of care draw the line in one sentence. “There are currently no long-term data to support the use of pharmacologic treatments other than metformin for the sole purpose of preventing type 2 diabetes”. And nothing in either regulator’s record ties any of these approvals to a genetic test or a gene score. Drug decisions belong with a prescribing clinician who can weigh side effects, cost and duration.

Section recap: Structured lifestyle change cut progression by 58 percent over 2.8 years in people with prediabetes, and the benefit held at 24 percent after 21 years. Tirzepatide cut diagnoses to 1.3 percent versus 13.3 percent at 176 weeks. After 17 weeks off treatment the figures were 2.4 and 13.7 percent, and no drug is approved for prevention in either country.

Family, insurance and privacy

The last questions are usually the ones keeping people awake.

There is no cascade testing here

For single-gene conditions, relatives get tested. That is cascade testing. For a polygenic score, it does not apply. The ACMG statement is explicit. “The PRS results cannot be used to predict the relative disease risk for other family members, although some correlation may be observed”.

So a sibling in another province does not need a gene test. They need an A1C or a fasting glucose, on their own country’s schedule. Adult children should be told the family history, which counts as a risk factor in every guideline here, rather than a percentage from a website.

For children and teenagers, risk-based screening “should be considered after the onset of puberty or after 10 years of age, whichever occurs earlier”. It applies to those with overweight or obesity plus one or more risk factors. Family history of type 2 diabetes in a first- or second-degree relative is one of those factors, at Grade A. A pediatrician can start that conversation.

Two countries, two very different laws

Both countries ban the same conduct, but they picked different tools. The US Genetic Information Nondiscrimination Act, known as GINA, is a civil ban. Canada made the same conduct a crime.

	US: GINA	Canada: Genetic Non-Discrimination Act
Type of law	civil prohibition, a “floor of minimum protection” that state law may exceed	criminal law, upheld by the Supreme Court of Canada in 2020
Health insurance	“prohibits health insurers from discrimination based on the genetic information of enrollees”	no one may require “an individual to undergo a genetic test as a condition of” a contract, or require disclosure of results
Employment	“prevents employers from using genetic information in employment decisions”, but “GINA does not apply to employers with fewer than 15 employees”	covered by the same general ban
Life, disability, long-term care	“GINA’s health insurance protections do not cover long-term care insurance, life insurance, or disability insurance”	covered by the ban; “insurance companies couldn’t make people get tested to get life insurance coverage”
Military	“the U.S. military is permitted to use genetic and medical information to make employment decisions”	not addressed in the sources here
Penalty	civil enforcement	conviction on indictment brings “a fine not exceeding $1,000,000 or to imprisonment for a term not exceeding five years, or to both”

Canada’s Act also bars anyone from collecting, using or disclosing genetic test results “without the individual’s written consent”. Doctors and researchers are exempt, so a Canadian reader can still discuss results freely with their own physician.

There is a further point that surprises people. GINA covers genetic information, including family medical history. Once a person is actually diagnosed with diabetes, that is a manifested condition rather than genetic information. GINA then stops applying, and the US rule that takes over is the Affordable Care Act’s pre-existing-condition protection. That handoff follows from GINA’s own scope. The ACA rule itself is not one of the sources cited here.

Who holds the data

Custody of consumer genetic data is not permanent. The company behind these reports moved its Personal Genome Service and Research Services business lines to a nonprofit, TTAM Research Institute. The deal closed on 14 July 2025, and the former web address now redirects to a new one. The new owner states a commitment to “providing customers with choice and transparency with their data, including the option to change their decision on whether to participate in research”.

The practical step is dull. Find the research-consent setting and the data-download and account-deletion controls in the account, then decide on purpose. For any decision that touches testing, telling relatives or planning a pregnancy, a board-certified genetic counselor is the right professional to involve.

Section recap: A polygenic result says nothing usable about a specific relative, so siblings need a blood test rather than a gene test. GINA protects health insurance and larger employers but not life, disability or long-term-care coverage, while Canada’s Act is criminal law upheld by its Supreme Court.

Frequently asked questions

Will I get type 2 diabetes?

Nothing in a polygenic report can answer that. The score is “a statistical prediction of increased clinical risk”, not a diagnosis. The population baseline is 11.6 percent for diabetes and more than one in three adults for prediabetes. Only a blood test settles a person’s current status, and nearly a quarter of US adults with diabetes do not know they have it. A below-average score changes nothing either, because “a low PRS does not rule out significant risk”. A primary care physician can order the test that does answer it.

Will my children inherit it?

Not the way a single-gene condition is inherited. There is no carrier state and no transmission percentage, because risk is spread across 1,289 signals at 611 loci. What children inherit is a family history, which counts as a risk factor. In the Framingham Offspring cohort, one affected parent carried an age-adjusted odds ratio near 3.4 to 3.5, and two affected parents 6.1. Telling adult children the family history, and asking a physician about screening from age 35, is more useful than a percentage.

Will this affect my health or life insurance?

In the US, GINA bars health insurers from using genetic information, but it does not reach life, disability or long-term-care insurance. Once diabetes is diagnosed, it is a manifested condition, so GINA no longer applies. The US rule that takes over is the Affordable Care Act’s pre-existing-condition protection. That handoff follows from GINA’s scope, and the ACA rule is not among the sources cited here.

In Canada, the Genetic Non-Discrimination Act bars anyone from requiring a genetic test, or its results, as a condition of a contract. Penalties reach a $1,000,000 fine and five years in prison. The Supreme Court upheld that law in 2020.

Can my employer find out?

Not lawfully at most US workplaces. GINA’s employment protections apply to employers with 15 or more employees, so smaller employers fall outside it. The US military is also allowed to use genetic and medical information in employment decisions. State law may add protection, since GINA is a floor rather than a ceiling. In Canada, the ban applies to any person offering goods, services or contracts, with an exception only for treating clinicians and researchers.

Should I get a second opinion?

For the polygenic result, the right first contact is a primary care physician and a blood test, not another gene test. No US standard of care builds a polygenic score into type 2 diabetes screening or diagnosis. For a family pattern that suggests monogenic diabetes, escalate: an A1C under 7.5 percent at diagnosis, one parent affected, or a calculator probability above 5 percent. That needs a clinician-ordered diagnostic panel and, ideally, a board-certified genetic counselor.

Summary

A consumer type 2 diabetes result is a population probability produced by a score, and the company itself states it has not been reviewed by the FDA. The biology behind it is real and now mapped in detail: 1,289 independent signals at 611 loci, from 2,535,601 people. The average signal shifts odds by about 3 percent, and the strongest one sits near a per-copy odds ratio of 1.38.

Those signals sort into eight biological clusters with different tissue targets, which is why one family label can hide several diseases. Cluster scores predict vascular complications in research cohorts, and appear in no consumer report and no guideline.

The added value for one reader is small. Added to a full clinical model, a gene score moved the C statistic from 0.900 to 0.901 and reclassified at most 4 percent of subjects. A parent’s diagnosis already carries a roughly 3.5-fold odds ratio. Family history and a waist measurement hold most of what the report conveys.

There is one genetic question that pays. Monogenic diabetes is under 5 percent of cases, and more than 80 percent of UK MODY goes undiagnosed. In one US study, 44 of 47 youth with a confirmed mutation carried the wrong label. HNF1A-MODY responds to sulfonylureas, and GCK-MODY often needs no treatment at all. Any change to a diabetes medicine still belongs to the treating clinician.

The modifiable arm has the best evidence here. Lifestyle change cut progression by 58 percent over 2.8 years, and by 24 percent still at 21 years. The genotype penalty was smallest in the lifestyle arm, though the interaction was not statistically significant. Tirzepatide is approved for treating type 2 diabetes, in the US since 13 May 2022, and no drug in either country is approved to prevent it.

The next step is not another search. It is an A1C or a fasting glucose, ordered by a physician who can see the family history too.

This article is for educational purposes only. It is not a substitute for advice from a licensed physician, board-certified medical geneticist, or board-certified genetic counselor. For any decisions about testing, treatment, or care, consult a qualified clinician. In emergencies, call 911.

References

1. Suzuki K, Hatzikotoulas K, Southam L, et al. (2024). Genetic drivers of heterogeneity in type 2 diabetes pathophysiology. Nature 627(8003):347-357 (epub 19 February 2024). PMID 38374256. DOI 10.1038/s41586-024-07019-6. Full text PMC10937372. https://pubmed.ncbi.nlm.nih.gov/38374256/
2. Grant SF, Thorleifsson G, Reynisdottir I, et al. (2006). Variant of transcription factor 7-like 2 (TCF7L2) gene confers risk of type 2 diabetes. Nature Genetics 38(3):320-323 (epub 15 January 2006). PMID 16415884. DOI 10.1038/ng1732. https://pubmed.ncbi.nlm.nih.gov/16415884/
3. Steinthorsdottir V, Thorleifsson G, Reynisdottir I, et al. (2007). A variant in CDKAL1 influences insulin response and risk of type 2 diabetes. Nature Genetics 39(6):770-775 (epub 26 April 2007). PMID 17460697. DOI 10.1038/ng2043. https://pubmed.ncbi.nlm.nih.gov/17460697/
4. NHGRI-EBI GWAS Catalog (US National Human Genome Research Institute and EMBL-EBI). Association records for rs7903146 (TCF7L2) and rs7754840 (CDKAL1), trait “type 2 diabetes mellitus”. REST API queried live 2026-08-27. https://www.ebi.ac.uk/gwas/variants/rs7903146
5. Knowler WC, Barrett-Connor E, Fowler SE, et al.; Diabetes Prevention Program Research Group (2002). Reduction in the incidence of type 2 diabetes with lifestyle intervention or metformin. New England Journal of Medicine 346(6):393-403. PMID 11832527. DOI 10.1056/NEJMoa012512. https://pubmed.ncbi.nlm.nih.gov/11832527/
6. Diabetes Prevention Program Research Group (2015). Long-term effects of lifestyle intervention or metformin on diabetes development and microvascular complications over 15-year follow-up: the Diabetes Prevention Program Outcomes Study. Lancet Diabetes & Endocrinology 3(11):866-875. PMID 26377054. DOI 10.1016/S2213-8587(15)00291-0. Full text PMC4623946. https://pubmed.ncbi.nlm.nih.gov/26377054/
7. Florez JC, Jablonski KA, Bayley N, et al.; Diabetes Prevention Program Research Group (2006). TCF7L2 polymorphisms and progression to diabetes in the Diabetes Prevention Program. New England Journal of Medicine 355(3):241-250. PMID 16855264. DOI 10.1056/NEJMoa062418. ClinicalTrials.gov NCT00004992. https://pubmed.ncbi.nlm.nih.gov/16855264/
8. Meigs JB, Shrader P, Sullivan LM, et al. (2008). Genotype score in addition to common risk factors for prediction of type 2 diabetes. New England Journal of Medicine 359(21):2208-2219. PMID 19020323. DOI 10.1056/NEJMoa0804742. Full text PMC2746946. https://pubmed.ncbi.nlm.nih.gov/19020323/
9. Talmud PJ, Hingorani AD, Cooper JA, et al. (2010). Utility of genetic and non-genetic risk factors in prediction of type 2 diabetes: Whitehall II prospective cohort study. BMJ 340:b4838. PMID 20075150. DOI 10.1136/bmj.b4838. https://pubmed.ncbi.nlm.nih.gov/20075150/
10. Narayan KM, Boyle JP, Thompson TJ, Sorensen SW, Williamson DF (2003). Lifetime risk for diabetes mellitus in the United States. JAMA 290(14):1884-1890. PMID 14532317. DOI 10.1001/jama.290.14.1884. https://pubmed.ncbi.nlm.nih.gov/14532317/
11. Meigs JB, Cupples LA, Wilson PW (2000). Parental transmission of type 2 diabetes: the Framingham Offspring Study. Diabetes 49(12):2201-2207. PMID 11118026. DOI 10.2337/diabetes.49.12.2201. https://pubmed.ncbi.nlm.nih.gov/11118026/
12. National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK), National Institutes of Health. Diabetes Statistics (page last reviewed January 2024; figures attributed to the CDC, data year 2021). Retrieved live 2026-08-27. https://www.niddk.nih.gov/health-information/health-statistics/diabetes-statistics
13. US Preventive Services Task Force; Davidson KW, Barry MJ, Mangione CM, et al. (2021). Screening for Prediabetes and Type 2 Diabetes: US Preventive Services Task Force Recommendation Statement. JAMA 326(8):736-743. PMID 34427594. DOI 10.1001/jama.2021.12531. Status confirmed current on the USPSTF site 2026-08-27. https://www.uspreventiveservicestaskforce.org/uspstf/recommendation/screening-for-prediabetes-and-type-2-diabetes
14. American Diabetes Association Professional Practice Committee (2026). 2. Diagnosis and Classification of Diabetes: Standards of Care in Diabetes-2026. Diabetes Care 49(Supplement_1):S27-S49. PMID 41358893. DOI 10.2337/dc26-S002. Full text PMC12690183. https://pubmed.ncbi.nlm.nih.gov/41358893/
15. American Diabetes Association Professional Practice Committee (2026). 3. Prevention or Delay of Diabetes and Associated Comorbidities: Standards of Care in Diabetes-2026. Diabetes Care 49(Supplement_1):S50-S60. PMID 41358891. DOI 10.2337/dc26-S003. Full text PMC12690170. https://pubmed.ncbi.nlm.nih.gov/41358891/
16. Diabetes Canada Clinical Practice Guidelines Expert Committee (2018). Chapter 3, Definition, Classification and Diagnosis of Diabetes, and Chapter 4, Screening for Diabetes in Adults. Canadian Journal of Diabetes; with the Public Health Agency of Canada CANRISK questionnaire. Chapters retrieved live 2026-08-27. https://guidelines.diabetes.ca/cpg/chapter4
17. Abu-El-Haija A, Reddi HV, Wand H, Rose NC, Mori M, Qian E, Murray MF; ACMG Professional Practice and Guidelines Committee (2023). The clinical application of polygenic risk scores: A points to consider statement of the American College of Medical Genetics and Genomics. Genetics in Medicine 25(5):100803. PMID 36920474. DOI 10.1016/j.gim.2023.100803. https://www.acmg.net/PDFLibrary/Clinical-Application-Polygenic-Risk-Scores.pdf
18. Martin AR, Kanai M, Kamatani Y, Okada Y, Neale BM, Daly MJ (2019). Clinical use of current polygenic risk scores may exacerbate health disparities. Nature Genetics 51(4):584-591 (epub 29 March 2019). PMID 30926966. DOI 10.1038/s41588-019-0379-x. Full text PMC6563838. https://pubmed.ncbi.nlm.nih.gov/30926966/
19. Khera AV, Chaffin M, Aragam KG, et al. (2018). Genome-wide polygenic scores for common diseases identify individuals with risk equivalent to monogenic mutations. Nature Genetics 50(9):1219-1224 (epub 13 August 2018). PMID 30104762. DOI 10.1038/s41588-018-0183-z. Full text PMC6128408. https://pubmed.ncbi.nlm.nih.gov/30104762/
20. Shields BM, Hicks S, Shepherd MH, Colclough K, Hattersley AT, Ellard S (2010). Maturity-onset diabetes of the young (MODY): how many cases are we missing? Diabetologia 53(12):2504-2508 (epub 25 May 2010). PMID 20499044. DOI 10.1007/s00125-010-1799-4. https://pubmed.ncbi.nlm.nih.gov/20499044/
21. Pihoker C, Gilliam LK, Ellard S, et al.; SEARCH for Diabetes in Youth Study Group (2013). Prevalence, characteristics and clinical diagnosis of maturity onset diabetes of the young due to mutations in HNF1A, HNF4A, and glucokinase. Journal of Clinical Endocrinology & Metabolism 98(10):4055-4062. PMID 23771925. DOI 10.1210/jc.2013-1279. Full text PMC3790621. https://pubmed.ncbi.nlm.nih.gov/23771925/
22. Lacy ME, Wellenius GA, Sumner AE, et al. (2017). Association of Sickle Cell Trait With Hemoglobin A1c in African Americans. JAMA 317(5):507-515. PMID 28170479. DOI 10.1001/jama.2016.21035. https://pubmed.ncbi.nlm.nih.gov/28170479/
23. Jastreboff AM, le Roux CW, Stefanski A, et al.; SURMOUNT-1 Investigators (2025). Tirzepatide for Obesity Treatment and Diabetes Prevention. New England Journal of Medicine 392(10):958-971 (epub 13 November 2024). PMID 39536238. DOI 10.1056/NEJMoa2410819. ClinicalTrials.gov NCT04184622; funded by Eli Lilly. https://pubmed.ncbi.nlm.nih.gov/39536238/
24. US Food and Drug Administration, Drugs@FDA records queried via the openFDA drug/drugsfda API, and Health Canada Drug Product Database queried via the health-products.canada.ca API, both 2026-08-27. Mounjaro NDA215866 approved 2022-05-13; Zepbound NDA217806 approved 2023-11-08; Ozempic NDA209637 approved 2017-12-05; Health Canada DIN 02532891 status “Approved”, status date 2022-11-24. https://www.accessdata.fda.gov/scripts/cder/daf/
25. 23andMe / TTAM Research Institute. Health Predispositions clinician reference page, the Personal Genome Service test disclaimer, and the TTAM Research Institute acquisition press release dated 14 July 2025 (company product documentation, cited only for the company’s statements about its own product). Retrieved live 2026-08-27. https://www.23andme.org/clinicians/genetic-reports/health-predispositions/
26. US National Human Genome Research Institute (NIH), Genetic Discrimination and the Genetic Information Nondiscrimination Act of 2008; Genetic Non-Discrimination Act, S.C. 2017, c. 3 (Justice Laws Website, consolidation current to 2026-06-21); Supreme Court of Canada, Reference re Genetic Non-Discrimination Act, 2020 SCC 17 (judgment of 10 July 2020, Case in Brief). All retrieved live 2026-08-27. https://www.genome.gov/about-genomics/policy-issues/Genetic-Discrimination
27. Centers for Medicare & Medicaid Services, Innovation Center. Medicare Diabetes Prevention Program (MDPP) Expanded Model, including CY 2026 Physician Fee Schedule Final Rule updates. Retrieved live 2026-08-27. https://www.cms.gov/priorities/innovation/innovation-models/medicare-diabetes-prevention-program
28. US National Library of Medicine, National Institutes of Health. MedlinePlus Genetics: “Type 2 diabetes” (page last updated 1 November 2017). Retrieved live 2026-08-27. https://medlineplus.gov/genetics/condition/type-2-diabetes/
29. National Society of Genetic Counselors, Find a Genetic Counselor directory, and the Canadian Association of Genetic Counsellors (now at genetic-counsellors.ca). Both retrieved live 2026-08-27. https://www.nsgc.org/findageneticcounselor

Last updated: 2026-08-27

Author: Yu Mizuno (Editor-in-Chief, non-physician), GeneLumen editorial team. This article aggregates 29 sources from peer-reviewed medical literature and public health agencies (tier 1=22 / tier 2=6 / tier 4=1), including NIH resources (NIDDK, MedlinePlus Genetics, NHGRI, the NHGRI-EBI GWAS Catalog), the US Preventive Services Task Force, American Diabetes Association Standards of Care, Diabetes Canada guidelines, ACMG statements, the FDA and Health Canada drug databases, CMS, Canadian federal statute and Supreme Court sources, and PubMed-indexed publications. Editorial lead: Yu Mizuno, a non-physician research editor.

This article is for educational purposes only and is not a substitute for medical advice from a licensed physician, board-certified medical geneticist, or board-certified genetic counselor. In emergencies, call 911 (US/Canada).

Related: Metabolic and Hematologic Genetic Diseases category

🇯🇵 For readers in Japan — a separate Japanese edition written for Japan’s healthcare system (not a translation): https://genelumen.com/ja/ja-metabolic-hematologic-genetic/type2-diabetes-tcf7l2-polygenic-risk-kazokureki-japan