Relative risk is one group's chance of an event divided by another group's chance. Absolute risk is the share of people who had that event over a stated time. That's the difference in relative risk vs absolute risk: how the groups compare, and how often the event happened.
A headline saying “20% lower risk” doesn't mean 20 fewer people out of every 100 had a problem. The starting risk matters. So do the outcome being counted and the length of follow-up.
Relative risk vs absolute risk: the official usage
The FDA's Communicating Risks and Benefits: An Evidence-Based User's Guide was published in August 2011. Its chapter on quantitative information recommends giving absolute risks alongside relative figures. It also recommends keeping the number of people and the time frame the same when making comparisons.
There isn't a special FDA definition that changes the arithmetic. These are standard statistical terms, explained in Ranganathan and colleagues' methods paper, Perspectives in Clinical Research, 2016.
| Term | What it means |
|---|---|
| Absolute risk | The proportion of people who experience an event over a stated period |
| Relative risk, also called a risk ratio | One group's absolute risk divided by the comparison group's |
| Absolute risk reduction | The comparison group's risk minus the treatment group's risk |
| Number needed to treat, or NNT | How many people would need treatment instead of the comparison for one fewer person to have the event, on average |
These terms exist because one result can look very different depending on how it's presented. “Risk fell by half” sounds substantial. Whether that means 100 fewer events or one fewer event depends on where the risk started.
A benefit example: SELECT's heart-event numbers
SELECT, reported by Lincoff and colleagues in NEJM in 2023, studied 17,604 adults with overweight or obesity and existing cardiovascular disease, meaning heart or blood-vessel disease. They did not have diabetes.
The main outcome combined cardiovascular death, nonfatal heart attack and nonfatal stroke. Average follow-up was 39.8 months, roughly three years and four months.
| Group | People with a first major event | About how many per 1,000 |
|---|---|---|
| Branded semaglutide | 569 of 8,803: 6.5% | 65 |
| Placebo, a look-alike treatment without the drug | 701 of 8,801: 8.0% | 80 |
Subtracting the published percentages gives 1.5 percentage points, meaning the gap between 8.0% and 6.5%. That's about 15 fewer people with an event per 1,000 during the trial.
SELECT tested branded Wegovy. These figures don't establish the outcomes of Promise's compounded semaglutide. The SELECT results guide explains the trial's participants, safety findings and limits.
Number needed to treat and hazard ratio meaning
NNT is calculated as 1 divided by the absolute risk reduction written as a decimal. Here, 1 ÷ 0.015 = 66.7, or approximately 67.
That is our arithmetic from SELECT's rounded published event rates. It describes roughly one fewer person having the counted event for every 67 treated instead of placebo over the observed follow-up.
People had different follow-up lengths, so this shortcut isn't a formal estimate at one fixed time. It doesn't mean the sixty-seventh person benefits or that the other 66 receive nothing.
The headline “20% lower risk” comes from a different calculation. SELECT reported a hazard ratio of 0.80. A hazard ratio compares how quickly first events occur over time among people who haven't yet had one.
It isn't the same as a risk ratio. Dividing the rounded event proportions, 6.5% by 8.0%, gives 0.8125, a relative reduction of about 18.8%. The trial's time-based analysis gives 20%. The difference comes from the calculation, not two conflicting trial results.
A harm example: a large ratio for an uncommon event
As of October 1, 2026, the day this article was written, the current Zepbound prescribing information carries “Revised: 08/2026.” Section 5.4 reports cholecystitis, meaning gallbladder inflammation, in 0.7% with Zepbound versus 0.2% with placebo across two weight-reduction trials.
Those are about seven people versus two people per 1,000. The pooled safety studies involved treatment for up to 72 weeks, plus four weeks of follow-up after treatment.
Our arithmetic from the label's rounded rates gives two descriptions:
- Relative: 0.7 ÷ 0.2 = 3.5 times the risk, or a 250% relative increase.
- Absolute: 0.7% − 0.2% = 0.5 percentage points, or about five additional people per 1,000.
Both descriptions belong beside the claim. The ratio alone can sound alarming; the absolute difference alone can make a serious condition sound trivial.
Zepbound contains tirzepatide. These labeled rates describe the studied branded product, rather than a measured rate for Promise's compounded tirzepatide.
How the numbers show up in a prescription discussion
For semaglutide or tirzepatide, a provider needs more than a percentage. They consider whether the person resembles the studied group, which preparation is being considered, and the person's medical history.
SELECT's starting risk came from adults who already had cardiovascular disease. It isn't a starting risk for every person considering weight management.
Semaglutide and tirzepatide prescribed through Promise are dispensed by a licensed U.S. compounding pharmacy as compounded medications, which are different from FDA-approved products: the formulations offered here are not FDA-approved. Compounding means a pharmacy prepares medication for a prescription. The FDA's compounding explanation says the agency doesn't review these preparations for safety, effectiveness or quality before marketing.
A licensed provider may still prescribe — that decision is between you and your doctor.
Mistakes that change what a headline means
Reading a relative percentage as a headcount. A 20% relative reduction starts with the comparison group's risk. In a hypothetical example, reducing a 10% risk by 20% brings it to 8%, not zero.
Ranking medicines across unrelated trials. Different starting risks, participants and counted events can produce different absolute benefits. A larger relative reduction alone doesn't establish which medicine suits someone better.
Leaving off the time frame. A risk measured over several years isn't an annual risk. An NNT needs its outcome and follow-up attached every time it's quoted.
Treating the average as a personal forecast. Group results describe the people studied. For a semaglutide discussion, the absolute numbers help make that evidence understandable; they don't replace the person's clinical review.
Related terms that keep a result in context
An endpoint is the outcome a study measures. An estimand defines the treatment effect the study is trying to estimate, including how treatment stops are handled.
Clinical trial phases describe the stage of testing. A boxed warning is the most prominent safety warning on a prescription label. None of these terms supplies the missing starting risk in a headline.
The useful question remains concrete: how many people had which event, in each group, over how long?
At Promise, a licensed provider reviews every request, and not everyone qualifies.
