Communicating Polygenic Absolute Risk

Why the Wording of Risk Matters

A polygenic risk estimate can be statistically sound and still be misunderstood. “Your risk is 4%” leaves several questions unanswered: risk of which event, over what period, and relative to which population and assumptions? It can also sound like a forecast that says what will happen to one particular person. A useful risk communication makes the quantity interpretable without implying certainty.

The central object is an absolute risk over a stated time horizon. It is the predicted probability that a defined event occurs for a person with specified characteristics by a specified time, under a stated prediction model and target setting. A polygenic score may contribute to that prediction alongside clinical predictors and baseline incidence, but the communication should report the resulting absolute risk rather than relying on the score or a relative effect alone.

State the Event, Horizon, and Setting

The time horizon is part of the risk itself, not a detail to add later. “A 5% risk” is incomplete if the estimate is 5% over 5 years, 10 years, or a lifetime. The event also needs to be explicit: for example, a first diagnosis of a specified disease, rather than a broad or ambiguous label. If the model estimates the probability in the presence of competing events, the reported event probability should follow that model’s definition.

The setting matters because an estimate is conditional on a prediction model, its input values, and the population and care context for which it is intended. Earlier material on calibration and local baseline incidence establishes why a risk estimate should be appropriate to its target setting. In communication, do not imply that a number is universal if its application depends on a defined population or assumptions.

A concise report can state the estimate first, then make its scope visible: “Based on this model and the information available, the estimated chance of developing [event] over the next 10 years is 3.2%.” The phrase “estimated chance” signals that this is a probability produced by a model, not a guarantee. Where appropriate, name the model’s target setting and note that the estimate depends on the accuracy and completeness of the recorded predictors.

Use Natural Frequencies Without Changing the Meaning

A natural frequency preserves the probability when the denominator is stated and used consistently. If a risk is \(p\), then among \(N\) people with the relevant profile the expected number of events is \(Np\), and the expected number without the event is \(N(1-p)\). These are expectations, not exact observed counts. Choose a denominator that makes the numbers easy to understand, such as 100 or 1,000, and make clear that the people are being described by the same risk profile and horizon.

The complementary number can help prevent a high-sounding relative comparison from eclipsing the absolute chance of the event. For a 3.2% risk, the complement is 96.8%; in a group of 1,000, that corresponds to 968 people who would not have the event within the stated period, in expectation. Do not use this complement to imply that those people have no health concerns: it refers only to the specified event and time horizon.

Frequency statements are most useful when the denominator is not allowed to drift. Comparing “32 per 1,000 over 10 years” with “32 per 1,000 over a lifetime” would compare different quantities, even though the counts match. Likewise, a frequency from a group with a different profile or target setting is not automatically a fair comparison.

Report Uncertainty as Uncertainty

A point estimate alone can conceal how imprecisely the risk has been estimated. An uncertainty interval puts a range around the estimate, but the interval must be identified and interpreted according to how it was constructed. For example, a 95% frequentist confidence interval for a model-based risk has a repeated-sampling interpretation: under the model and sampling assumptions, the procedure used to construct such intervals would cover the relevant fixed risk parameter in 95% of repeated samples. It does not mean there is a 95% probability that the fixed parameter lies in this particular interval.

An interval also does not ordinarily describe the range of outcomes for one person. A person either experiences the specified event within the horizon or does not; the interval reflects uncertainty about the estimated probability, not a prediction that the person’s outcome will fall between its endpoints. The interval may reflect sampling uncertainty or other components included by the calculation. Model misspecification, incomplete inputs, and changes in the target setting are not automatically captured. State the interval’s type and principal source when known.

A Predicted Probability Is Not a Deterministic Forecast

A risk model assigns a probability conditional on the information it uses. Even if the estimate is well calibrated for people with similar predicted risks, it does not determine the outcome for a particular person. A 10-year risk of 8% means that, under the model and target conditions, the event is estimated to occur for about 8 in 100 comparable people over 10 years. It does not mean the model knows that one named person will or will not have the event.

This distinction also matters for the complementary outcome. At an estimated 8% risk, the model implies an estimated 92% probability of no event within the same horizon, for the same event definition and setting. Neither probability should be phrased as certainty. “The model estimates an 8% chance over 10 years” is more accurate than “You will develop the disease” or “You are safe from the disease.”

A probability is not a causal effect. A higher estimated risk associated with a polygenic score does not mean that the score itself caused the disease, and a lower score does not guarantee protection. Risk communication should describe what the model predicts, rather than turning a statistical association into a causal claim or a promise about an individual’s future.

Common Mistakes in Risk Communication

• Leaving out the horizon. “The risk is 4%” is incomplete. State, for example, “the estimated risk of the defined event over the next 10 years is 4%.”
• Using only a relative comparison. “Three times the risk” can sound dramatic without showing the absolute quantities. Give both risks and, where useful, the difference in percentage points or events per 1,000.
• Treating natural frequencies as guaranteed counts. The wrong reasoning is that a 3% risk means exactly 30 events in every group of 1,000. The correction is that 30 is the expected count under the model; actual counts vary.
• Calling an uncertainty interval an individual outcome range. The interval describes uncertainty in the estimated risk as defined by its method, not a range of outcomes that the person might “land in.”
• Presenting a model estimate as certain or causal. A predicted probability is not a deterministic forecast, and an association between score and risk does not by itself show that the score causes disease.
• Mixing settings or denominators. Frequencies are comparable only when event, horizon, risk profile definition, and target setting are aligned.