“This genetic finding doubles your risk” sounds precise, but it leaves out the number most people need first: what was the risk before it doubled? A change from 1% to 2% and a change from 10% to 20% both double risk. They describe very different numbers of people affected.
To read a genetics headline or research paper, put the baseline risk, comparison risk, outcome and time period together. The arithmetic is usually straightforward. Deciding whether those numbers apply to a particular person takes more care.
The three numbers to put beside a headline
Absolute risk is the probability of an outcome in a specified population over a stated period. Relative risk, or the risk ratio, compares two risks by division. The absolute risk difference compares them by subtraction. The National Cancer Institute’s explanation of risk gives the clinical context for these distinctions.
| Measure | Example A | Example B |
|---|---|---|
| Reference-group risk | 1%: 10 in 1,000 | 10%: 100 in 1,000 |
| Comparison-group risk | 2%: 20 in 1,000 | 20%: 200 in 1,000 |
| Relative risk | 2 ÷ 1 = 2 | 20 ÷ 10 = 2 |
| Absolute increase | 1 percentage point: 10 more in 1,000 | 10 percentage points: 100 more in 1,000 |
These numbers are invented to demonstrate arithmetic; they are not estimates for any gene or disease. “A 100% relative increase” describes both columns. “An increase of 100 percentage points” describes neither. When percentages become confusing, translate them into counts with the same denominator.
Example 1: turn a headline into a complete sentence
Fictional teaching example. A headline says, “People with variant X have twice the risk.” The underlying study reports five-year risks of 1% and 2% for a defined outcome.
A more informative summary is: “In this study population, the five-year risk was 20 in 1,000 in the comparison group and 10 in 1,000 in the reference group—an absolute difference of 10 in 1,000 and a risk ratio of two.”
Now ask what the headline still leaves unresolved. Who was included? How was the outcome measured? Were the groups comparable? How uncertain are the estimates? An association may be useful evidence without establishing that the genetic difference alone caused the outcome.
Example 2: a large relative benefit can be a modest absolute benefit
Fictional teaching example. In a hypothetical trial with comparable groups and complete five-year follow-up, an outcome occurs in 20 of 1,000 people receiving usual care and 10 of 1,000 receiving an intervention.
- Absolute risks: 2% and 1%.
- Absolute risk reduction: 1 percentage point, or 10 fewer events per 1,000 people over five years.
- Relative risk: 1% ÷ 2% = 0.5.
- Relative risk reduction: 50%.
If the trial validly establishes that reduction, the point estimate corresponds to treating 100 people for that period to prevent one additional event on average. It does not identify which individual benefits. This calculation also omits uncertainty and possible harms; it is an arithmetic illustration, not a treatment recommendation.
The Cochrane Handbook’s guidance on interpreting results explains why absolute effects, the chosen baseline risk and benefits and harms belong together. For a fuller reading process, see our guide to interpreting gene therapy trials.
Odds ratios and hazard ratios need their own labels
A risk compares events with everyone at risk; odds compare events with non-events. They are not interchangeable. For another invented cohort, risks of 20% and 40% give a risk ratio of two. Their odds are 20/80 and 40/60, giving an odds ratio of approximately 2.67.
The difference matters when interpreting studies that report odds ratios, including many case-control analyses. A hazard ratio concerns the relative instantaneous event rates in a time-to-event analysis; it is not automatically a ratio of five-year probabilities. Preserve the measure the authors actually estimated. The Cochrane guidance on effect measures explains these distinctions.
Example 3: a polygenic percentile is not a disease probability
Fictional teaching example. A report places a person at the 90th percentile of a polygenic score. That means the score is high relative to its reference distribution. It does not mean a 90% chance of developing the disease.
A polygenic score combines information across many genetic variants. Translating it into an absolute prediction requires context, including the outcome, time horizon and the population in which the model was evaluated. The NHGRI explanation of polygenic risk scores describes why people with similar genetic scores can have different disease risks.
A useful question for the report is: “Does this number rank my score, estimate my probability of disease, or compare me with another group?” Those are three different statements, even if all are expressed with numbers.
Check the timeframe and the fit to the population
Lifetime risk cannot be substituted for risk during the next year. Developing a disease, dying from it and having a recurrence are also different outcomes. Copy the exact outcome definition before interpreting the number.
A prediction model should be evaluated beyond the data used to build it. One important property is calibration: whether predicted probabilities agree with observed outcomes in the intended setting. NCI describes independent validation and calibration monitoring in its absolute-risk modeling overview. A model that separates higher- from lower-risk groups can still give inaccurate absolute probabilities.
For researchers, record uncertainty intervals alongside the estimate. A narrow interval does not eliminate bias or show that the model transfers to a different population. Avoid reducing the result to whether a p value crosses a threshold; the effect’s size, precision and relevance remain essential.
A five-line worksheet for your next paper
| Complete this line | What it prevents |
|---|---|
| The outcome and time period are… | Mixing different events or lifetime and short-term risks. |
| The groups being compared are… | Losing the reference population behind the multiplier. |
| The absolute numbers are… | Reporting only a relative change. |
| The effect measure and uncertainty are… | Relabeling odds, risks or hazards, or presenting a point estimate as exact. |
| This does and does not establish… | Turning a study result into an unsupported personal prediction. |
Use the paper-comparison workspace to keep these notes. If the question concerns a genetic report, our guides to what a genetic test can miss and secondary findings explain two other sources of uncertainty.