Re: {MEDSTATS} Ratios as DVs and IVs: BMI (was Re: Interesting results, how to deal with these interaction effects?)

[Peter Flom]

I neglected to mention that Kronmal does use BMI as one of his examples of what not to do

Peter

-----Original Message-----
>From: Peter Flom <peterflom...@mindspring.com>
>Sent: Oct 26, 2009 10:58 AM
>To: meds...@googlegroups.com, MedStats <meds...@googlegroups.com>
>Subject: {MEDSTATS} Ratios as DVs and IVs: BMI (was Re: Interesting results, how to deal with these interaction effects?)
>
>
>Thanks to Frank Harrell and Bruce Weaver for their insights and
>recommendations.
>
>In the article by Kronmal to which Bruce refers below, he argues against any use
>of a ratio DV or IV
>
>This got me thinking about studies that use BMI, either as DV or IV. BMI is, of course,
>a ratio.
>
>Are these studies therefore bogus?
>
>Peter
>
>-----Original Message-----
>>From: Bruce Weaver <bwe...@lakeheadu.ca>
>>Sent: Oct 23, 2009 10:16 AM
>>To: MedStats <meds...@googlegroups.com>
>>Subject: {MEDSTATS} Re: Interesting results, how to deal with these interaction effects?
>>
>>
>>On Oct 23, 9:26 am, Frank Harrell <f.harr...@vanderbilt.edu> wrote:
>>> The laws of arithmetic are violated any time you use % change as a
>>> dependent variable.  A simple way to see this is that an increase of
>>> 100% is balanced by a decrease of 50%.  Percent changes may only be
>>> computed on group summary measures, not per-patient measures.
>>
>>Peter, here are a couple articles that might help you argue Frank's
>>point about % change with the investigators.
>>
>> http://www.jstor.org/stable/2983064
>> http://www.biomedcentral.com/1471-2288/1/6
>>
>>If you don't have JSTOR access, I can send you the Kronmal article--
>>just drop me a line.
>>
>>--
>>Bruce Weaver
>>bwe...@lakeheadu.ca
>>http://sites.google.com/a/lakeheadu.ca/bweaver/Home
>>"When all else fails, RTFM."
>>
>>>
>
>
>Peter L. Flom, PhD
>Statistical Consultant
>Website: www DOT peterflomconsulting DOT com
>Writing; http://www.associatedcontent.com/user/582880/peter_flom.html
>Twitter: @peterflom
>
>>


Peter L. Flom, PhD
Statistical Consultant
Website: www DOT peterflomconsulting DOT com
Writing; http://www.associatedcontent.com/user/582880/peter_flom.html
Twitter: @peterflom

[Juliet Hannah]

Hi Peter,

Any more insights into the BMI question you brought up. It seems like
a good question. BMI
is a well-studied variable, and seems to approximate some measure of
fat (along with waist-hip ratio).
It is expensive to obtain other measures of fat. Maybe in this case,
there is evidence that BMI
approximates the underlying measure (central adiposity) so this makes
it more justifiable?

Juliet

[Peter Flom]

Hi Juliet

I have not had any replies about BMI. I am not very impressed with BMI as a measure, even ignoring the fact that it's a ratio. It's just not a good measure of overweight; it ignores too many factors that vary both across individuals and across groups.

Peter

[Martin Holt]

I agree with Peter. When BMI is discussed as a measure, I think of a forward
in a rugby union team, who will have a high BMI despite not carrying much
fat. But then I wonder, "What alternative measures are there ?"

Martin Holt

[Peter Flom]

Martin Holt <m861...@btinternet.com> wrote

>
>I agree with Peter.

:-)

>When BMI is discussed as a measure, I think of a forward
>in a rugby union team, who will have a high BMI despite not carrying much
>fat. But then I wonder, "What alternative measures are there ?"
>

Being American, I think of running backs in American football. There was
one who was 5'9" and over 200 pounds (1.75 m and 95kg) and not any excess fat.

Better measures? The best simple one is probably waist-hip ratio. Although it, too,
is a ratio.

Peter

[John Sorkin]

It is unfortunate when people besmirch a useful measure on theoretical basis without considering the actual track record of the measure. Yes, there are theoretical reasons for not liking BMI. Two that come quickly to mind are the fact that BMI is a ratio (and in some rare instances one can obtain incorrect inferences when a ratio is used in a model), and second BMI which as a measure of relative obesity will not work well for everyone (whiteness the example of the rugby player of American professional football player). Despite the fact that is does not work for some "odd" people, it works well for a vast segment of the population and is very helpful in studies where it is not feasible to get more direct measures of obesity such as DXA or CT scans. Despite these drawbacks, BMI has proven a very useful measure. When used properly it is a very strong predictor of many outcomes of interest including cardiovascular disease, diabetes, hypertension, stroke, and mortality. BMI is easily and reliably measured (unlike waist-to-hip ratio which can vary widely with the observer especially in obese subjects). Is BMI perfect, absolutely not, but it is useful both clinically and statistically.

I await the poison arrows that will certainly come my way . . .

John

John David Sorkin M.D., Ph.D.
Chief, Biostatistics and Informatics
University of Maryland School of Medicine Division of Gerontology
Baltimore VA Medical Center
10 North Greene Street
GRECC (BT/18/GR)
Baltimore, MD 21201-1524
(Phone) 410-605-7119
(Fax) 410-605-7913 (Please call phone number above prior to faxing)

>>> Peter Flom <peterflom...@mindspring.com> 11/2/2009 9:00 AM >>>

Confidentiality Statement:
This email message, including any attachments, is for the sole use of the intended recipient(s) and may contain confidential and privileged information. Any unauthorized use, disclosure or distribution is prohibited. If you are not the intended recipient, please contact the sender by reply email and destroy all copies of the original message.

[William Stanbury]

The C19 Belgian inventor of the now Anglicised "Body Mass Index" or BMI was - guess what folks - a research statistician. He spent many years if not decades ploughing through data and came up with his Quetelet Index, ie. mass in kg divided by height in metres squared. There were other ratios at the time, e.g. the Broca Index, yet the Quetelet Index prevailed to become the BMI.  My understanding (to be confirmed or infirmed) of the recent waist-hip ratio is that it is most applicable to individuals in a given population who are very obese, in danger of - or needing - surgery, as well as being even more sensitivethan the BMI to persons of different origins. Such individuals are comparatively rare in western populations (though perhaps no longer in anglophone populations...). Clinical trials people are often more focused on "individuals" and sometimes do not think or work as epidemiologists on "populations".

Perhaps for now, as with democracy, we should make do with the BMI as the least bad method?

 

 

2009/11/2 Peter Flom <peterflom...@mindspring.com>

[John Whittington]

Returning to the general, rather than the specific (I realise that there is
a lot of the debate about pros and cons of BMI) ....

Whilst I understand the mathematical concerns about analysing a ratio,
rather than the constituent components of that ratio, it seems to me that
there are often cases in which the ratio (or some other mathematical
derivation of two or more observed variables) is the quantity which has
practical and meaningful value, in which case it seems (to me) to make
sense to analyse that ratio (and little, if any, sense to analyse the
components of that ratio). To take a rather extreme illustrative example:

Average speed is the ratio of distance to time. If, say, we had data which
consisted of time taken to cover a particular distance (that distance
differing between observations), if our interest was in average speed, then
it would surely make sense to calculate and analyse that 'ratio'? To my
mind, to conduct an analysis in which time and distance were treated
separately would be silly and meddlesome, to say the least.

Kind Regards,
John

At 11:58 01/11/2009 -0500, Juliet Hannah wrote:
>Hi Peter,
>
>Any more insights into the BMI question you brought up. It seems like
>a good question. BMI
>is a well-studied variable, and seems to approximate some measure of
>fat (along with waist-hip ratio).
>It is expensive to obtain other measures of fat. Maybe in this case,
>there is evidence that BMI
>approximates the underlying measure (central adiposity) so this makes
>it more justifiable?
>
>Juliet

John

----------------------------------------------------------------
Dr John Whittington, Voice: +44 (0) 1296 730225
Mediscience Services Fax: +44 (0) 1296 738893
Twyford Manor, Twyford, E-mail: Joh...@mediscience.co.uk
Buckingham MK18 4EL, UK
----------------------------------------------------------------

[Martin Holt]

I think John Whittington's comments (Nov 02, 2009 below) about not examining
the components of a ratio make a lot of sense. Rugby players are not that
extreme IMO~they are fit and healthy and one would think perfect specimens
for a measure aimed at discriminating between such fit people and
others...yet the BMI measure does not seem to do this. To say that this is
because they are extreme misunderstands the role of the BMI, IMO; that is to
discriminate between fit and healthy people in terms of size, and others. If
BMI is expected not to work with these people, for some reason, it should
state the conditions under which it is expected to work.

I remembered hearing once that the average Norwegian male was 4" taller than
a British one ...here I am doing just what John warns against....so I dug up
a table that gives average heights across nationalities
http://en.wikipedia.org/wiki/Human_height#Average_height_around_the_world

I was surprised at the range of heights. I wondered if the statistician who
invented the Quetelet index (later BMI) had included figures like these to
validate the index (if he had data for the corresponding weights, and used
the ratios as John says). Is the BMI an international index ?

Best Wishes,

Martin Holt


----- Original Message -----

From: "John Whittington" <Joh...@mediscience.co.uk>
To: <meds...@googlegroups.com>
Sent: Monday, November 02, 2009 3:31 PM
Subject: {MEDSTATS} Re: Ratios as DVs and IVs: BMI (was Re: Interesting
results, how to deal with these interaction effects?)


>

[Peter Flom]

More even than difference in heights (with which the BMI at least attempts to deal) there could be differences in the relative length of torsos and legs across cultures. I tried to find data on this, and found very little. This would cause cross-cultural problems.

But there are certainly wide variations in this across people, but even that is hard to find data on. On a purely anecdotal level, I am 5'9" (a little shorter than average) but when I buy pants, I usually am at the the low end on inseam length (30" or 31" in jeans).


Peter

[William Stanbury]

Martin,

 

From memory Henri Quetelet worked on the Belgian population in C19, and possibly also in early C20. He may have worked with data from other cultures or lands to validate his index however this would need to be checked. This checking would answer the question whether the BMI was derived from international data or not and hence whether it is a bona fide international index. Even if the index is demonstrated not to be a bona fide international one, Belgium is in a unique population position in Europe at a certain cultural and economic crossroads. It has been invaded and occupied historically by many different European cultures, many of whom bred in Belgium. The land also has a long history of welcoming refugees, whether political or economic. Furthermore I understand that its capital, Brussels, overtook New York in 2007 to become the most international city on the planet. Hence the Belgian population in C19 was already quite heterogeneous (at the very least by European standards) and is particularly heterogeneous now by planetary standards. Thus, in addition to the hard, lengthy, and systematic research work pre-IT performed by Quetelet to conclude upon the BMI to be the least bad method deduced from his data, perhaps this Belgian population heterogeneity explains why the Quetelet Index competed well with other methods at the time to become a planetary "gold standard" - because it worked best through having been developed from a heterogeneous population?

 

Nevertheless, if there remain a number of physicians, statisticians, or epidemiologists on the planet who are not happy with the BMI and are researching, or about to research, in an endeavour to find a better method, in addition to cities such as Brussels, Geneva, New York, and Sydney, I understand that a number of Caribbean islands may also be quite heterogeneous.   

 

 

2009/11/3 Peter Flom <peterflom...@mindspring.com>

[Bruce Weaver]

On Nov 2, 9:35 am, "John Sorkin" <jsor...@grecc.umaryland.edu> wrote:
> It is unfortunate when people besmirch a useful measure on theoretical basis without considering the actual track record of the measure. Yes, there are theoretical reasons for not liking BMI. Two that come quickly to mind are the fact that BMI is a ratio (and in some rare instances one can obtain incorrect inferences when a ratio is used in a model), and second BMI which as a measure of relative obesity will not work well for everyone (whiteness the example of the rugby player of American professional football player). Despite the fact that is does not work for some "odd" people, it works well for a vast segment of the population and is very helpful in studies where it is not feasible to get more direct measures of obesity such as DXA or CT scans. Despite these drawbacks, BMI has proven a very useful measure. When used properly it is a very strong predictor of many outcomes of interest including cardiovascular disease, diabetes, hypertension, stroke, and mortality. BMI is easily and reliably measured (unlike waist-to-hip ratio which can vary widely with the observer especially in obese subjects). Is BMI perfect, absolutely not, but it is useful both clinically and statistically.
>
> I await the poison arrows that will certainly come my way . . .

Hi John. I have no poison arrows. But, in the examples of BMI's
usefulness you provide above, it is used as an explanatory variable.
Kronmal addresses this situation in his article, and argues pretty
persuasively (I think) that one is better off including height and
weight (and possibly their product) as terms in the model. One
argument he offers is that using BMI makes it impossible to see the
relative contributions of height and weight. He gives another example
where there is no association with BMI, but when height & weight are
entered as separate variables, height is strongly associated while
weight is not.

http://www.jstor.org/stable/2983064 - Kronmal's article

Cheers,
Bruce