Disease-Risk Prediction From Genetics

Start here

A genetic risk report estimates how a person's DNA relates to the chance of developing a condition. It is not a diagnosis and cannot say whether you will get a disease. If a report worries you, talk with a clinician or genetic counselor before any decision about screening or treatment. See talking with a health professional.

Evidence labels used here (established, promising but limited, insufficient evidence) are explained in how we label evidence.

Two very different kinds of genetic risk

Single-gene variants. Some tests look for specific variants with large effects, such as the three BRCA1/BRCA2 variants in one FDA-authorized consumer report. The FDA states that report does not describe overall cancer risk, and that absence of a tested variant does not rule out other cancer-related variants. [1]

Many small effects combined. Most common conditions involve many variants plus lifestyle and environment. Genome-wide studies have shown that common disorders have a polygenic architecture, and the variants can be combined into a polygenic risk score. [2] See polygenic risk scores for the full explainer and genomics explained for background.

What the evidence shows about polygenic scores

Evidence label: promising but limited for clinical use; established as a research tool.

• A 2020 review said polygenic risk scores have been widely applied in research and confirm associations with disease status, but their clinical utility has yet to be established, and discrimination in the general population is currently low. [2]
• In a UK Biobank evaluation of 23 colorectal cancer models (443,888 people, 2,679 cases in 6 years of follow-up), the best model using only gene variants had an AUC (a measure where 0.5 is chance and 1.0 is perfect) of 0.62 in women and 0.64 in men. Adding age and lifestyle factors raised it to 0.64 to 0.67 in women and 0.67 to 0.71 in men. [3]
• Scores are several times more accurate in people of European ancestry than other ancestries, because most large genetic studies have drawn on European-ancestry participants. [4] See ancestry and population applicability.

These findings come from observational prediction studies, not from trials showing that acting on a score improves health. A score that predicts a disease is not evidence that using it changes outcomes; see clinical validity and clinical utility and risk prediction vs treatment selection.

Why a risk number can mislead

• A percentile is not a lifetime risk. Translating a score's position on a distribution into a lifetime risk is a stated challenge. [2]
• Different tests, different variants. Companies test different sets of variants and may give different results for the same condition. [1]
• Negative does not mean safe. Not all variants contributing to risk are known, so a negative result still leaves some risk. [1]
• Positive does not mean disease. A positive result does not mean you will develop the condition. [1]
• Wrong calls happen. For very rare variants on genotyping arrays, only 16% of heterozygous calls were confirmed by sequencing in one large evaluation. [5]

What the FDA says (United States, read 2026-10-22)

The FDA describes genetic health risk tests as providing information on genetic risk for certain conditions that may help lifestyle decisions and discussions with a provider, while not determining overall risk because environmental, other genetic and lifestyle factors also matter. [1] The page also says a positive result should not be used to decide on treatment. [1] See direct-to-consumer genetic tests for the exact tests listed.

Sensible questions to ask

• Which variants were tested, and how many contribute to this disease?
• Has the score been validated in people of my ancestry?
• Would the result change what my clinician recommends? If not, what is it for?
• Who sees my data afterwards? See genetic privacy and incidental findings.