Genetic Report says 89% Risk of Type 2 Diabetes? Why That Number Is Probably Misleading

I recently received a genetic report estimating my lifetime risk of Type 2 Diabetes at 89%.

That’s not “high risk.” That’s near certainty.

There’s just one problem: it’s almost certainly wrong.

As the CTO of a genomic technology company, I spend a lot of time evaluating reports like this. This one turned out to be a perfect example of how risk can be presented in a way that is mathematically plausible — and deeply misleading.

After digging into how it was likely calculated, it became clear that the number says more about how risk is calculated than about my actual future health. This post is about a simple yet critical distinction that often gets lost: the difference between absolute and relative risk.

The headline number

Here’s what the report said:

• A “typical” male with my BMI (~29) has a 29.7% lifetime risk of Type 2 Diabetes
• Based on my genetics, sex at birth, age, and BMI, my risk is 89.0%

At face value, that suggests: I am almost guaranteed to develop diabetes. But that’s not what the underlying science actually supports.

Relative risk: how much higher than average?

The key input in the report is a polygenic risk score (PRS) — an aggregate measure of the genetic variants that contribute to Type 2 Diabetes.

My score: 3.33

That likely means something like: My genetic risk is ~3× higher than average.

Digging into the methodology, this is a Z-score that is standardized to follow a normal distribution with a mean of 0 and a standard deviation of 1. Importantly, a Z-score of 3.3 says I am 3.3 standard deviations above the population mean. This is fine, I know my genetics place me at high risk of Type 2 Diabetes.

This is relative risk.

Relative risk answers the question: “How does my risk compare to someone else’s?”

It does not answer: “What is the probability I will get this disease?”

Absolute risk: what is my actual probability?

Absolute risk is what most people think they’re being told:

“You have an 89% chance of developing diabetes.”

But to estimate absolute risk properly, you need much more than genetics:

• Age
• Weight trajectory
• Diet and physical activity
• Blood glucose markers
• Family history
• Environment

Even then, estimates are subject to significant uncertainty.

Where things go wrong

The report appears to do something like this:

• Start with a baseline risk: 29.7%
• Multiply by genetic risk (~3×)
• Arrive at ~89%

The baseline risk seems legit, as there are published results that quantify BMI-specific lifetime diabetes risk in the U.S. for age-, sex-, and ethnicity-specific subgroups (1). However, the issue here is the apparent multiplication of baseline risk by the relative genetic risk to arrive at an absolute of 89%. This is appealingly simple — and deeply misleading.

Why?

1. Risk doesn’t scale linearly

You can’t just multiply probabilities and expect reality to follow. If you could, stacking risk factors would quickly push everyone toward 100%.

That’s not what we observe in real populations.

2. Real-world data doesn’t support near-certainty

Even among people with:

• Obesity
• Strong family history
• High genetic risk

You do not see 90% of them developing diabetes. If nearly 90% of individuals with this profile developed diabetes, we would already observe that in epidemiological cohorts. We don’t. In real-world population data, the lifetime risk of type 2 diabetes is ~30–60% (2, 3), not 89%. Even in high-risk populations, it rarely approaches certainty. Polygenic risk scores for Type 2 Diabetes typically confer a 2–4× increase in relative risk in the highest-risk individuals (4,5). Applied to real-world lifetime risks of ~30–60%, this raises risk, not to certainty, but into a higher, still far-from-deterministic range.

My risk is elevated — but far from deterministic.

Incidentally, the Z-score of 3.33 aligns with a relative risk of 2–4x. In polygenic risk score models for Type 2 Diabetes, each standard deviation increase typically corresponds to a ~20–40% increase in odds (OR ≈ 1.23–1.43 per SD) (6). For a Z-score of 3.33, this translates to a relative risk of roughly (1.23–1.43)^(3.33) ≈ 1.9–3.3, consistent with the ~2–4× relative risk observed in individuals at the high end of polygenic risk distributions (4,5). The issue here is not the Z-score of 3.33; it is how it has been applied to get an absolute risk estimate. One cannot directly multiply a probability by relative risk measures such as odds ratios or risk ratios.

What I think this should’ve been:

• Start with a baseline probability of p = 0.29
• Convert the baseline probability to odds: Odds = p/(1-p) = 0.297/0.703 = 0.42
• Apply the converted baseline odds ratio to the polygenic risk odds of 1.9–3.3, depending on the odds increase per standard deviation we want to use. Lower: 0.42 1.9 = 0.8; upper: 0.42 3.3 = 1.39
• Convert back to probability: Lower: 0.8/(1+0.8) = 0.44; upper: 1.39/(1+1.39) = 0.58.

Even if we accept the genetic effect size — roughly a 3× increase in risk — the way it’s applied matters. Odds ratios must be applied to odds, not probabilities. When done correctly, the same inputs yield an absolute risk range of 44–58%, not 89%.

The difference isn’t biology — it’s math. The model didn’t discover a near-certain outcome — it computed one by applying the right inputs in the wrong mathematical space.

Even so, I would not advocate for this corrected approach in practice. Estimating absolute risk properly requires incorporating multiple covariates and fitting well-calibrated survival models on large, representative cohort data. That’s a fundamentally different problem from scaling a baseline probability with a genetic risk multiplier — and it’s where most simplified approaches break down.

3. Lifestyle is doing invisible work

In addition, the model doesn’t meaningfully account for:

• Diet
• Exercise
• Weight changes over time

These factors can cut risk in half or more, even in high genetic risk groups. So the 89% implicitly assumes something like: “average or unfavorable lifestyle, indefinitely.”

That’s a big assumption — and one that isn’t stated clearly.

A more honest interpretation

A more scientifically grounded way to phrase the same result would be:

“You are in a high genetic risk group for type 2 diabetes. Given your current BMI, your risk is meaningfully higher than average.”

That’s useful. But it’s very different from: “You have an 89% chance of getting diabetes.”

Why this matters

Numbers like “89% risk” feel precise and authoritative. But when they’re built on stacked assumptions, they can:

• Create unnecessary anxiety.
• Undermine trust in genetics.
• Distract us from what actually matters.

Which is this:

For Type 2 Diabetes, genetics loads the gun, but lifestyle pulls the trigger

The actionable takeaway

The most important thing about high genetic risk for Type 2 Diabetes is not the number — it’s the opportunity.

Research consistently shows:

• People at high genetic risk benefit at least as much from lifestyle changes
• Weight loss, exercise, and diet can dramatically reduce risk
• Many high-risk individuals never develop the disease

So the takeaway isn’t: “This is your destiny.” It’s: “This is your early warning system.”

Final thought

If you’re given a risk number — especially a big, scary one — ask:

• Is this absolute risk or relative risk?
• What assumptions went into this number?
• What factors are missing?

Because in many cases, the number isn’t wrong — it’s just presented in a way that makes it sound far more certain than it really is.

I will also add that our internal reporting at Simplify Genomics puts me at the 81st percentile of risk for developing Type 2 Diabetes, but importantly, it does not convert this to an absolute risk. Rather, we report a risk ratio, in my case 1.4, that is the ratio of those who have developed Type 2 Diabetes in my percentile group compared to the average genome (40th-60th percentile, see 7). Naively, we could combine the prevalence of Type 2 Diabetes in the United States for males aged 50 of 15% (8) and conclude my absolute probability is 15*1.4 = 21%. This is also incorrect for the same reason: a relative risk cannot be directly multiplied by prevalence to yield an individual probability without a properly calibrated model.

References

1. Narayan KMV, Boyle JP, Thompson TJ, Gregg EW, Williamson DF. Effect of BMI on lifetime risk for diabetes in the U.S. Diabetes Care. 2007;30(6):1562–1566. Available from: https://pubmed.ncbi.nlm.nih.gov/17372155/
2. Narayan KMV, Boyle JP, Thompson TJ, Sorensen SW, Williamson DF. Lifetime risk for diabetes mellitus in the United States. JAMA. 2003;290(14):1884–1890. Available from: https://pubmed.ncbi.nlm.nih.gov/14532317/
3. Gregg EW, Buckley J, Ali MK, Davies J, Flood D, Griffiths B, et al. Lifetime risk of type 2 diabetes in 23 high-income jurisdictions: a multinational population-based study. Lancet Diabetes Endocrinol. 2022. Available from: https://www.sciencedirect.com/science/article/abs/pii/S2213858722002522
4. Khera AV, Chaffin M, Aragam KG, Haas ME, Roselli C, Choi SH, et al. Genome-wide polygenic scores for common diseases identify individuals with risk equivalent to monogenic mutations. Nat Genet. 2018;50:1219–1224. Available from: https://www.nature.com/articles/s41588-018-0183-z
5. Mars N, Koskela JT, Ripatti P, Kiiskinen TTJ, Havulinna AS, Lindbohm JV, et al. Polygenic and clinical risk scores and their impact on age at onset and prediction of cardiometabolic diseases. Nat Med. 2020;26:549–557. Available from: https://pubmed.ncbi.nlm.nih.gov/32273609/
6. Ashenhurst et al. A Polygenic Score for Type 2 Diabetes Improves Risk Stratification Beyond Current Clinical Screening Factors in an Ancestrally Diverse Sample. Front Genet 2022 26;13:871260. Available from: https://pmc.ncbi.nlm.nih.gov/articles/PMC9086969/
7. Simplify Genomics. Genomic Clinical White Paper: Polygenic Risk Score v1.0 [Internet]. 2023. Available from: https://docs.simplifygenomics.com/gcr/Genomic-Clinical-White-Paper-Polygenic-1-0.pdf
8. Koyama AK, et al. Lifetime risk of diabetes in the United States. [Internet]. 2022. Available from: https://pmc.ncbi.nlm.nih.gov/articles/PMC7310804/