In a study, I measure BMI of a large sample of adult people.

Among them, there is some who have very low and very high height.

As I can read on the internet (like here or even in the wiki, for instance), BMI is not relevant for very short or very tall people.

Some math professor even wrote an article with a "better" calculation of BMI, but with a very low impact in science world (didn't find it on pubmed) so it's not really peer-reviewed.

Let's take some examples, assuming "normal BMI" is 21:

  • for a 110cm tall adult, ideal weight is 25kg
  • for a 130cm tall adult, ideal weight is 36kg

These weights appear really low to me. Note that dwarfism cutoff is 145cm in France.

My question is then the following: How could I choose a cutoff on height to exclude BMI values ?

NB: I didn't find anything on pubmed nor on google scholar, but maybe I missed an important article

  • 1
    $\begingroup$ I don't really know much about this subject, but I doubt there is a precise cutoff. Isn't it simply a continuous thing? Gradually, the more you move away from mean height (in either direction), the less value the BMI measurement has. One thing you could do is a weighted analysis, with more average height people having heigher weight. $\endgroup$
    – Eff
    Commented Jun 29, 2018 at 8:07
  • $\begingroup$ I don't think this has been studied this precisely but I agree, it should be continuous. Still, I guess it is not linear (x3 maybe?), so there should be a value or a range of value where it become reasonable to think BMI is not relevant anymore. $\endgroup$ Commented Jul 3, 2018 at 9:53

1 Answer 1


The best procedure is going to depend on what exactly you're going to do with your data. Regardless, you should NOT exclude the BMI values. Just include the height, weight, and BMI. If you're simply reporting BMI as a characteristic of your study population, you can annotate that figure and let your reviewers and readers decide what to make of it. This follows the principles in Chapter 4 of Hulley's Designing Clinical Research. You want to get and maintain all the data.

If you're going run some further analysis on BMI, or if BMI is an outcome in your study, then you can evaluate the impact of a cutoff on your results, but you have to decide ahead of time how you'll do your primary analysis. I recommend using the entire data set (no height cutoff) for your primary analysis (because *there is no established practice for a BMI height cutoff in the general medical literature), and then running a secondary analysis that excludes those research subjects above some threshold (you can pick it, but depending on the size of your sample, 2 standard deviations could be reasonable). That secondary analysis would be a hypothesis generating analysis. If excluding very short and very tall people changed the results of your analysis, that's something to mention in your discussion and something to consider for the next study.

*If you're in a specialized field, then follow the practice of that field for your primary analysis.

  • $\begingroup$ Very interesting answer. Keeping height along with BMI is a very good idea. Still, I don't understand why excluding people is such a bad idea. My outcome is overweight and I don't have any other measurement than BMI to assess it, but I know BMI is a bad overweight assesser for very short people. Thus, by keeping short people in my analysis, I think I'd induce bias (I would not really measure my outcome for this very population), wouldn't I ? $\endgroup$ Commented Jul 11, 2018 at 9:50
  • $\begingroup$ If you're not going to apply your results to people below a certain height, then you have an excellent argument for excluding those people from your study. I was under the impression you already had the data and it included people who were very short (who, presumably, had other measurements taken that were used in other parts of the analysis). $\endgroup$
    – De Novo
    Commented Jul 11, 2018 at 16:03

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .