“20% lower cardiovascular risk” sounds precise, but the percentage alone does not tell you how common the events were, which events counted, or how long people were followed. A useful reading note puts the relative effect beside the actual group results and the population studied.
SELECT provides a real example. Its findings can be described accurately in several ways, but those descriptions are not interchangeable. Working through the arithmetic helps reveal what a headline includes, what it leaves out, and where a trial result stops being a personal forecast.
First identify who and what the trial studied
SELECT randomized adults aged 45 or older with established cardiovascular disease, BMI of at least 27, and no history of diabetes to semaglutide or placebo. The primary outcome combined cardiovascular death, nonfatal heart attack, or nonfatal stroke, analyzed by time to the first qualifying event. Mean follow-up was 39.8 months. Read the original SELECT trial report.
Those details define the question. “Cardiovascular outcome” here does not mean every possible heart symptom, and an event count for this composite is not a count of deaths alone. The study population also differs from people without established cardiovascular disease. Keeping the eligibility criteria visible prevents a result for secondary prevention from silently becoming a claim about everyone.
Put both groups on the page
| Group | Participants with an event | Reported proportion |
|---|---|---|
| Semaglutide | 569 of 8,803 | 6.5% |
| Placebo | 701 of 8,801 | 8.0% |
The absolute difference using the rounded proportions is 8.0% − 6.5% = 1.5 percentage points. Expressed with a common denominator, that is about 80 versus 65 participants with an event per 1,000 enrolled, or about 15 fewer per 1,000 during the observed follow-up. This is a rescaling of trial percentages, not a newly observed group of exactly 1,000 people.
“Percentage points” matters. A fall from 8.0% to 6.5% is not a 1.5% relative reduction. Dividing the 1.5-point difference by 8.0 gives an 18.75% relative reduction in these rounded event proportions. That calculation answers a different question from the reported hazard ratio.
Why the headline can say 20% while that division gives 18.75%
SELECT reported a hazard ratio of 0.80, with a 95% confidence interval of 0.72 to 0.90. That interval describes uncertainty around the relative-hazard estimate under the analysis; it is not a range of personal risks or absolute benefits. Subtracting 0.80 from 1 gives the headline’s 20% lower relative hazard. A hazard ratio comes from a time-to-event analysis, which uses information about when events occur and handles follow-up through the analysis method. It is not simply the division of two rounded percentages.
The raw proportions and the hazard ratio are related summaries of the same study. They need not match exactly. Labeling each correctly avoids turning a small arithmetic difference into an apparent contradiction. For practical reading, keep the reported measure’s name: hazard ratio, risk ratio, and odds ratio should not all be relabeled “risk reduction” without explanation.
The event proportions also do not mean every participant was observed for exactly 39.8 months. That is a mean follow-up, not a uniform deadline. Dividing the 1.5-point difference by the number of years would not create a valid annual treatment effect. For a comparison at a specified time, look for the study’s appropriate time-specific analysis.
An invented example shows why starting risk matters
Imagine two fictional populations in which an intervention produces the same 25% relative reduction in risk over the same defined period. These numbers teach arithmetic; they are not SELECT results or estimates for any medicine.
| Starting risk | Risk after reduction | Absolute difference |
|---|---|---|
| 4% | 3% | 1 percentage point, or 10 per 1,000 |
| 20% | 15% | 5 percentage points, or 50 per 1,000 |
In both rows, divide the absolute difference by the starting risk to get 25%. The larger absolute difference in the second row comes from the higher starting risk. This is why the relative figure alone cannot describe the size of a benefit in everyday terms.
You cannot make this a personal SELECT calculator by inserting an estimated risk and multiplying by 0.80. A hazard ratio is not automatically a risk ratio, and the treatment effect may not transfer to a different population or follow-up period. Personal application requires assessment of both baseline risk and whether the evidence fits the person.
Read the outcome definition before interpreting the benefit
A composite outcome groups different events together. A reduction in the combined outcome does not establish that every component fell by the same amount or that every component carries the same consequence for a patient. When a headline says “fewer cardiovascular events,” check whether it is reporting the primary composite, one component, or a separate analysis.
Also ask whether the number counts people with at least one event or all events, including repeat events. SELECT’s primary analysis concerns the first qualifying event. Repeating its percentage as a reduction in the total number of all future events would change the question.
Put harms beside benefits without subtracting unlike outcomes
Ask how adverse events were collected, which participants were included, and how long observation continued. A harms percentage based only on treated participants may use a different denominator from a benefit analysis based on randomized participants. Clear reporting makes these differences inspectable. CONSORT Harms 2022 sets out reporting expectations for benefits and harms, including analysis populations.
A treatment discontinuation, a nonfatal event, and a death are different outcomes. Their percentages cannot be subtracted to produce one universal “net benefit” score. The same person can also contribute to more than one outcome. A useful comparison preserves outcome definitions, severity, time frame, and uncertainty, then considers what matters in the clinical decision.
A five-line note for checking the next headline
- Population: who was eligible, and who does the headline imply?
- Outcome: what counted, and was it a composite or an individual event?
- Group results: what were the event counts, denominators, and absolute proportions?
- Relative measure and time: what measure was reported, and over what follow-up?
- Limits: what uncertainty, harms, missing data, or population differences affect interpretation?
If a summary supplies only the relative percentage, leave the other lines blank until you can inspect the original report. Missing context is a reason to investigate, not proof that the result is unimportant. Our guide to reading weight-loss trials applies the same habit to weight outcomes and analysis choices.



