Relative vs. Absolute Risk:
How to Read Any Trial Headline Number

Two Numbers, One Trial: Relative Risk vs. Absolute Risk

Every clinical trial that measures whether a treatment prevents something, a heart attack, a hospitalization, a death, produces two different ways of describing the same result: a relative risk reduction and an absolute risk reduction. They are not competing statistics; they are the same underlying data, viewed from two different distances, and almost every trial headline shows you only one of them.

Here is a worked example, using round numbers instead of a real trial, so the mechanics are easy to see. Imagine a randomized trial with 100 people in a placebo group and 100 in a treatment group, followed for a year, tracking one outcome. In the placebo group, 5 people have that event. In the treatment group, 4 do.

• Relative risk reduction: the treatment group's rate (4%) is 20% lower than the placebo group's rate (5%). This is the number that becomes a headline: "cuts risk by a fifth."
• Absolute risk reduction: 5% minus 4% is 1 percentage point. Out of 100 people treated for a year, one additional person avoided the event, compared with placebo.

Both numbers are true. Both are calculated from the identical 5-versus-4 result. But "20 percent" and "1 point out of 100" land very differently, even though nothing about the underlying data changed between them, only the arithmetic used to describe it.

Number Needed to Treat: Putting the Number in Perspective

Once you have an absolute risk reduction, there is one more calculation that turns it into something genuinely useful: number needed to treat, usually abbreviated NNT. It answers a very practical question: how many people, on average, would need to receive this treatment for the length of the study to prevent one additional occurrence of the outcome, compared with not treating them at all?

The formula is simple: NNT equals 1 divided by the absolute risk reduction, expressed as a decimal. In the worked example above, the absolute risk reduction was 1 percentage point, or 0.01, so NNT = 1 ÷ 0.01 = 100. On average, 100 people would need to take the treatment, for the trial's duration, for one of them to avoid the event who otherwise wouldn't have.

NNT does not, by itself, say whether a treatment is worth taking. A NNT of 100 for something serious and irreversible, a stroke, can be entirely reasonable, especially for a safe, well-tolerated treatment. A NNT of 100 for a minor, reversible symptom is a different conversation. What NNT does is move the discussion out of vague language like "reduces risk" and into a concrete number to weigh, ideally alongside its counterpart, number needed to harm (NNH), when a paper reports one.

Applying This to the GLP-1 Cardiovascular Outcome Trials

This distinction matters directly for GLP-1 medications, because some of the largest, most consequential trials in this drug class are not weight-loss trials at all. They are dedicated cardiovascular outcome trials, where the primary endpoint is a composite of major adverse cardiovascular events (MACE): typically cardiovascular death, non-fatal heart attack, and non-fatal stroke, tracked over years in tens of thousands of people.

The best known is SELECT, published by Lincoff and colleagues in the New England Journal of Medicine in 2023 and registered on ClinicalTrials.gov before it began. It enrolled more than 17,000 adults with a BMI of 27 or higher and an existing cardiovascular condition, but without diabetes, randomized to semaglutide 2.4 mg or placebo and followed for a median of roughly three years. The topline result reported publicly was a relative risk reduction, commonly described as approximately a 20 percent reduction in major adverse cardiovascular events, with a hazard ratio of 0.80.

Notice what that sentence does not contain: the absolute risk reduction. Calculating it takes the actual MACE event counts in each of the two arms, figures published in the paper's results tables and supplementary appendix, rarely repeated in a press release or a summary article. That is the number that tells you what changed for those specific 17,000-plus people, rather than how the two event rates compared to each other in percentage terms. It is available, but it takes going to the source.

The same reporting pattern shows up across earlier trials in this drug class. LEADER (liraglutide) and SUSTAIN-6 (semaglutide, studied in a population with type 2 diabetes at high cardiovascular risk) also reported cardiovascular benefit primarily as relative reductions; SUSTAIN-6, for instance, is commonly described as showing roughly a 26 percent reduction in major adverse cardiovascular events, driven largely by fewer strokes (Marso et al., New England Journal of Medicine, 2016). None of this makes these trials misleading; large, peer-reviewed cardiovascular outcome trials are exactly the evidence this checklist is built to reward. It means converting a relative number into an absolute one is the reader's job, and doing it properly requires the paper itself, not just the headline drawn from it.

Surrogate Endpoints vs. Hard Outcomes

A separate distinction matters just as much as relative versus absolute: what was actually measured. A surrogate endpoint is a stand-in, something like body weight, HbA1c, or LDL cholesterol, believed to track an outcome a patient actually cares about, but not that outcome itself. A hard outcome is the real thing: a heart attack, a stroke, a death.

Most of the earliest, largest GLP-1 trials, including STEP and SURMOUNT, were built around surrogate and functional endpoints: how much weight was lost, how much HbA1c dropped. Those are legitimate, clinically meaningful measurements, and the reason these drugs were approved for weight management and type 2 diabetes in the first place. But moving a surrogate marker in the right direction is not automatically the same as proving fewer heart attacks, strokes, or deaths. That is precisely why dedicated cardiovascular outcome trials like SELECT exist as a separate category of evidence, run to test whether the benefit extends to hard outcomes rather than being assumed from the surrogate marker moving.

The same applies to muscle and body composition. A trial reporting total weight lost is reporting a surrogate for what most people actually want: fat loss without a meaningful loss of lean mass. Understanding how much of a reported weight-loss number was lean tissue takes different data, the kind our guide to preserving lean mass on a GLP-1 medication walks through.

Trial Population vs. the Person Reading About It

A result is only as relevant to you as the population it was tested in. Every major trial publishes detailed inclusion and exclusion criteria: age range, BMI threshold, presence of diabetes, prior cardiovascular history, kidney function, and dozens of other filters for who was allowed to enroll. SELECT specifically excluded people with diabetes and required an existing cardiovascular condition. A different, equally real result in people with type 2 diabetes and no cardiovascular history, such as SUSTAIN-6, is not automatically the same finding transplanted onto a different population, even when it points the same direction.

Before treating a trial's topline number as personally relevant, it is worth checking, even briefly, whether you resemble the people who were actually studied: similar age, starting weight or BMI, health history, sex. A genuinely large, well-run trial in a population unlike your own is still good evidence, it is just evidence that may not transfer cleanly to your situation, and that gap is exactly what a prescribing clinician is positioned to evaluate.

Duration, and What Happens After the Trial Ends

Trial results describe what happened for the specific length of time people were followed, not indefinitely, and not after treatment stops. SELECT followed people for a median of roughly three years; STEP and SURMOUNT trials typically ran 68 to 72 weeks. A benefit measured over three years of continuous treatment is evidence about those three years, and extrapolating beyond that window is a judgment call, not a data point from the trial itself.

The discontinuation question deserves the same scrutiny as the headline number. Weight-loss trials that included a placebo-controlled withdrawal phase have generally shown that much of the benefit erodes once the medication stops, a separate, published finding worth reading directly rather than assuming. Whether a similar pattern holds for cardiovascular benefit after stopping is a genuinely open question the existing trials were not designed to answer cleanly, since most did not run a controlled off-drug follow-up phase long enough to isolate it.

Industry Funding and How to Read a Disclosure Statement

Nearly all large GLP-1 trials, including SELECT, LEADER, and SUSTAIN-6, were funded by the drug's manufacturer. That is standard, not unusual: trials of this size, tens of thousands of participants followed for years, cost far more than most independent academic funders can support, and regulators require this level of evidence before approval. Funding source alone does not tell you whether a result is trustworthy.

Worth checking instead: whether the trial was registered on ClinicalTrials.gov with its primary endpoint specified before results came in, so the endpoint could not be quietly redefined after seeing the data; whether it was independently peer-reviewed in a reputable journal; and whether the paper's funding and conflict-of-interest section is stated plainly rather than buried. Reading a disclosure statement is a skill worth having generally, not only for trials. It works the same way our own site's affiliate disclosure does: the point is not to make you distrust the content by default, it is to give you what you need to weigh it, which only works if you actually read it.

A 60-Second Checklist for Any Health Headline

You don't need a statistics background to run this. Six questions, in order:

Frequently Asked Questions

What is the difference between relative risk reduction and absolute risk reduction?

Relative risk reduction compares the event rate in a treatment group to the event rate in a comparison group, as a percentage difference between the two. Absolute risk reduction is the plain difference between those two rates, in percentage points. If 5 out of 100 people on placebo have an event and 4 out of 100 on the drug have one, the relative risk reduction is 20 percent, but the absolute risk reduction is 1 percentage point. Both numbers are correct; they answer different questions, and headlines almost always report the relative number because it looks larger.

What does number needed to treat (NNT) mean?

Number needed to treat is the number of people who would need to receive a treatment, for the length of the study, to prevent one additional occurrence of the outcome being measured, compared with not treating them. It is calculated as 1 divided by the absolute risk reduction expressed as a decimal. An absolute risk reduction of 1 percentage point (0.01) gives an NNT of 100. NNT does not say whether a treatment is worth taking; it gives you a concrete number to weigh against the cost, inconvenience, and side effects of treating that many people.

Why do the GLP-1 cardiovascular outcome trials report relative risk instead of absolute risk?

Trials like SELECT (Lincoff et al., New England Journal of Medicine, 2023) report a hazard ratio and a relative risk reduction because that is the standard statistical output of a randomized, time-to-event trial design, and it is also the number most likely to be repeated in a headline or press release. The absolute risk reduction can be calculated from the actual event counts in each study arm, which are reported in the paper's results tables and supplementary materials, but they are far less commonly quoted outside the published paper itself.

Is a surrogate endpoint like weight loss the same as a hard outcome like fewer heart attacks?

No. A surrogate endpoint, such as body weight or HbA1c, is a measurable stand-in that is believed to track something a patient actually cares about. A hard outcome, such as a heart attack, stroke, cardiovascular death, or all-cause mortality, is the thing itself. A treatment can move a surrogate marker in a favorable direction without being proven to change hard outcomes, which is exactly why dedicated cardiovascular and renal outcome trials, separate from the weight-loss or glucose-lowering trials, exist for this drug class.

Does industry funding automatically make a clinical trial untrustworthy?

No, and most large drug trials, including the major GLP-1 cardiovascular outcome trials, are funded by the manufacturer, since running a trial of that scale is expensive and regulators require this evidence before approval. Funding does not by itself invalidate a result. What matters is whether the trial was registered on ClinicalTrials.gov with its primary endpoint defined before the results came in, whether it was peer-reviewed and published in a reputable journal, and whether the paper's funding and conflict-of-interest disclosures are clearly stated, which is standard practice and worth reading, not a red flag on its own.