A claim such as “20% weight loss” is easy to remember. A fair interpretation needs more information: who took part, for how long, with what support and using which analysis? A few practical questions can make study news much easier to assess without a statistics qualification.
Start with the participants
A study in adults without diabetes does not automatically answer the same question as a study in type 2 diabetes. Age, starting weight and accompanying conditions also affect how well results apply elsewhere. Read the description of the population before focusing on the largest percentage.
For example, SURMOUNT-5 studied adults with obesity without type 2 diabetes. SELECT examined people with existing cardiovascular disease and no diabetes. Both are important trials, but they address different questions and support different conclusions.
An average is not a typical individual
An average compresses many outcomes into one number. Some participants lose more weight, others less. Some cannot continue treatment. The mean therefore does not tell you what you should have achieved by a particular month.
Consider a made-up example: three people lose two, eight and twenty kilograms. The average is ten kilograms, although none lost exactly ten. This illustrates arithmetic only. It is not evidence about a particular medicine or a claim about a usual treatment course.
Separate percentages from percentage points
If average weight loss is 15% in one group and 10% in another, the difference is five percentage points. Relative to the smaller figure, 15% is 50% greater. Both calculations can be correct, while creating very different impressions.
A percentage of weight also needs a reference. Ten percent of a starting weight of 100 kilograms equals ten kilograms. This is a calculation example, not a target. Always check whether a report describes weight change, the proportion of participants or the risk of an event.
Include time and the comparison group
An outcome at 64 weeks cannot simply be treated as equivalent to one at 72 weeks. Comparing against placebo, another medicine or a substantial support programme also means different things. Trial conditions are part of the result, not background decoration.
OASIS 4 compared oral semaglutide with placebo. SURMOUNT-5 directly compared two injected active ingredients. Placing their figures next to one another does not create a direct tablet-versus-injection comparison. A useful review makes that limitation visible.
Look at analysis and discontinuation
Check how researchers handled people who stopped treatment or missed measurements. Different scientific questions can justify different analyses. A number estimating continued treatment can answer a different question from an analysis that incorporates interruptions.
If a report does not explain this, look for the original paper or ask for clarification. A company presentation may emphasise one analysis. That does not automatically make its number false, but the associated question and uncertainty need to travel with the claim. Avoid removing those conditions when sharing results with other people.
Common questions
Can I calculate my expected result from the average?
You can understand the arithmetic, but cannot create a reliable personal forecast. Medical circumstances, tolerability and actual treatment experience vary between individuals.
Is an older study automatically useless?
No. Its question, quality and relevance matter. New evidence can add to existing knowledge. An older appropriate comparison can be more informative than a recent headline without published results.
Questions and experiences
Share a question or a useful experience about this topic. Every contribution is checked before publication. Comments may be written in different languages.
Loading reader contributions…
Opening the form sets a necessary session cookie. No email address is required. Publication is subject to moderation and is not guaranteed. For deletion requests: Contact