How to read clinical trial results without being misled by risk

A headline says a drug "cut the risk of stroke by 69%". That sounds like a near cure. Then you open the paper and find that 13 people in one group and 4 in the other had the event, out of more than 1,800. Both descriptions are true. They answer different questions, and knowing which question you care about is most of the skill of reading a trial result.

This guide walks through the numbers you will meet in almost every trial report, using one real study as the worked example: SINGLE-AF, a trial of blood thinners in people with atrial fibrillation (AF) and an intermediate stroke risk, presented at the European Society of Cardiology congress on 28 August 2026 and published in the New England Journal of Medicine. We covered the trial itself in our SINGLE-AF report. Here we use it to explain the maths.

This guide is for general information and is not medical advice. Talk to your own doctor before changing any treatment.

Start with the question the trial asked

Every trial is built around a primary outcome, the one result it was designed and sized to answer. Everything else is secondary, and secondary results carry less weight because the trial was not planned around them.

SINGLE-AF enrolled 1,803 adults in South Korea who had AF and a CHA2DS2-VASc score that put them in a grey zone, where guidelines do not give a firm answer on whether to start an anticoagulant. Half were assigned to a direct oral anticoagulant (DOAC) and half to no anticoagulant. The registration record, NCT04437654, describes the setup. The average age was 60.4 years and 23.7% of participants were women.

The primary outcome combined stroke, systemic embolism and death from cardiovascular causes. Before looking at any number, ask three things:

• Who was in the trial, and do they look like me?
• What was compared with what?
• What counted as the main result, and over how long?

In SINGLE-AF, a reader who is 80, or who already has a high stroke score, is outside the group studied. The result tells you little about that person.

Relative risk versus absolute risk

The primary outcome happened in 4 of 902 people on a DOAC (0.5%) and 13 of 901 people with no anticoagulant (1.5%).

There are two ways to describe that gap.

The relative way compares the two rates as a ratio. The trial reported a hazard ratio of 0.31. Put simply, the event rate in the DOAC group was about 31% of the rate in the comparison group, which is where a "69% lower risk" headline comes from.

The absolute way subtracts one rate from the other. 1.5% minus 0.5% is a difference of 1.0 percentage point. Out of every 100 people like those in the trial, about one fewer had the main event over the study period.

Both numbers are correct. The relative figure tells you how strong the effect of the drug was. The absolute figure tells you how much difference it is likely to make to one person, and it depends heavily on how common the problem was to begin with. A 69% cut in a rare event is a small change in your own chances. The same 69% cut in a common event would be large.

When you read any report, look for the absolute numbers. If an article only gives you the percentage reduction, go to the paper or the registry and find the raw event counts.

Number needed to treat

The number needed to treat (NNT) turns the absolute difference into something easier to picture. You get it by dividing 100 by the absolute difference in percentage points.

For SINGLE-AF, 100 divided by 1.0 gives an NNT of about 100. Around 100 people in this group would need to take a DOAC over roughly two years for one of them to avoid the primary outcome. The other 99 would take the drug without that particular benefit, because they were never going to have the event, or they had it anyway.

NNT is a guide and has wide uncertainty when events are rare. It is still one of the clearest ways to discuss a treatment with your doctor, because it sets the benefit beside the cost, the side effects and the effort of taking a medicine every day.

Confidence intervals show the uncertainty

The hazard ratio of 0.31 came with a 95% confidence interval of 0.10 to 0.94.

A confidence interval is the range of values that fits reasonably well with the data. The single best estimate is 0.31, but the real effect could plausibly be anywhere from a 90% reduction (0.10) to a 6% reduction (0.94). An interval that wide tells you the trial had few events to work with.

Two practical rules help:

• If the interval for a ratio crosses 1.0, the trial cannot rule out no effect at all.
• The wider the interval, the less precise the estimate, even if the result is called significant.

In SINGLE-AF, the upper end of 0.94 sits close to 1.0. The trial points towards benefit, but the size of that benefit is uncertain.

What a p-value does and does not tell you

The primary result had a p-value of 0.03. A p-value answers a narrow question: if the drug truly had no effect, how surprising would a result at least this large be? A value below 0.05 is the conventional threshold for calling a result statistically significant.

A p-value does not tell you how big or important the effect is, and it does not give the probability that the drug works. A tiny, unimportant effect in a huge trial can have a very small p-value. A large, meaningful effect in a small trial can miss the 0.05 line. Read the p-value next to the absolute difference and the confidence interval, never on its own.

Count the events

SINGLE-AF had 17 primary outcome events in total. That small number explains the wide confidence interval and is a reason to treat the result with care. When a handful of events would change the picture, one or two extra events in either group could shift the hazard ratio a long way.

A quick way to sense how solid a result is: look for the number of people who actually had the outcome in each arm, as well as the percentages. Trials with hundreds of events produce tighter, more reliable estimates than trials with a dozen or two.

Look at harms in the same way

Benefits mean little without the risks next to them. In SINGLE-AF, major bleeding occurred in 0.3% of the DOAC group and 0.5% of the no-anticoagulant group. Any stroke occurred in 0.3% and 1.1% respectively.

Apply the same tools. Turn harms into absolute differences and, where you can, a number needed to harm. Then check whether the trial was big enough and long enough to pick up uncommon side effects. Bleeding risk with anticoagulants rises with age and over time, and a two-year trial in people with an average age of 60 may not show what happens to older people over a decade.

Primary results, secondary results and subgroups

Trials report many numbers. Secondary outcomes and subgroups, such as results split by sex or age, can be useful hints. They are also where chance findings hide, because the more comparisons you run, the more likely one will look striking by luck.

Treat a secondary or subgroup result as a question for the next trial unless it was planned in advance and is backed by other evidence. Our guide to what it means when a trial fails goes further on subgroups, using the CARDIO-TTRansform trial of eplontersen.

How the trial was run

The design shapes how much weight a result can carry. SINGLE-AF was open-label, which means participants and their doctors knew who was taking the anticoagulant. Open-label designs can still give reliable answers, especially when outcomes are hard events such as stroke and are judged by people who do not know the treatment assignment. They are more open to bias when outcomes depend on judgement or on how people feel.

Our guides on placebo and blinding and open-label versus blinded trials explain the difference with examples.

Other design points worth checking:

• Randomisation. Were people assigned to groups by chance? That is the main protection against unfair comparisons.
• Comparator. Was the new treatment compared with placebo, with nothing, or with the current standard? A win over nothing is easier than a win over a good existing drug.
• Follow-up. How long were people watched, and how many dropped out?
• Funding. Who paid for the trial, and did independent researchers analyse the data?

Where the trial was done

SINGLE-AF recruited in South Korea. People there may differ from people elsewhere in body size, genetics, diet and how health care is organised. Trial results often travel well, but it is fair to ask whether a population looks like yours.

A short checklist for any trial report

When you see a new result, work through these in order:

1. What was the primary outcome, and did the trial meet it?
2. What were the event rates in each group, as plain percentages?
3. What is the absolute difference, and what does that mean as an NNT?
4. How wide is the confidence interval, and does it come close to 1.0?
5. How many events were there in total?
6. What were the harms, in absolute terms?
7. Was it randomised, and was it blinded?
8. Do the people studied look like you?

You can find the registration record of most trials on ClinicalTrials.gov or the WHO International Clinical Trials Registry Platform. The record lists the planned primary outcome, which lets you check whether the published paper stuck to it.

Applying the checklist to SINGLE-AF

Using the checklist, SINGLE-AF looks like this. The trial met its primary outcome. Event rates were 0.5% and 1.5%, an absolute difference of 1.0 percentage point, or an NNT of about 100 over two years. The confidence interval ran from 0.10 to 0.94, with only 17 events. Major bleeding was uncommon in both groups. The design was randomised but open-label, in a fairly young Korean population.

That picture supports a conversation between a patient and a doctor about starting an anticoagulant in this grey zone, and it does not settle the question for everyone. Larger trials and data from other countries will add to it. Guideline committees will weigh SINGLE-AF next to earlier studies before they change advice.

Talking to your doctor about a new result

If a trial in the news seems relevant to you, it can help to bring the specific study to your appointment. Useful questions include:

• Was I the kind of person who was in this trial?
• What is my own baseline risk, and what would the treatment change in absolute terms?
• What are the side effects, and how likely are they for me?
• Is this result strong enough to change what we do now, or should we wait for more evidence?

You do not need to become a statistician. Knowing the difference between relative and absolute risk, and asking how many events a result rests on, puts you in a far better position to judge what a headline is really saying.

Related reading

For more on this, see Single-AF: anticoagulation in intermediate stroke risk, How new medicines get approved and reach patients and How to find a clinical trial that fits you.