The distribution of fat in the visceral and gastroesophageal junction is associated with BE

Reference: Nelsen, E.M., et al., Distribution of Body Fat and Its Influence on Esophageal Inflammation and Dysplasia in Patients With Barrett's Esophagus. Clin Gastroenterol Hepatol, 2012.

A study out of the Mayo Clinic in Minnesota on the distribution of body fat and its influence on  Barret’s Esophagus. 

Brief Summary: It is known that body fat can increase the risk of Barret’s Esophagus (BE). However, it is not known how the distribution of the body fat (visceral, abdominal, etc.) in particular influences Barret’s Esophagus.   The main objective of this retrospective study was to study the distriubution of the fat in patients with Barret’s Esophagus – particuarly with respect to the distriubtion of the fat.  A total of 100 pateints were sampled for this study from the Mayo clinic: 50 of those with BE (case) and 50 without BE (controls).  Fat measurements were then taken from all of the patients according to the following speicific fat regions: subcutaneous fat (SF), visceral fat (VF), and gastroesophageal junction (GEJ) fat.  

Results: The distribution of fat in the visceral and gastroesophageal junction region is signficantly associated with Barret’s Esophagus (BE).     

Implications  for Practice: Patients with a high fat content in the gastroesophageal and visceral regions are at a higher liklihood of acquiring Barret’s Esophagus (BE) if shown in unison with other symptoms. 

Discussion:  A good observational study which shows how the distribution of fat influences BE.   I liked how the authors talked about the potential mechanisms between the body fat distributions and their influece on BE – many authors do not do this.   

In order to further elucidate the relationship between fat and BE, it would be nice to see a comprehensive multi-variate study which includes the full array of interaction variables.  For instance, the authors mentioned that BMI (body-mass index) alone does not explain the male and Caucasian pre-dominance of BE in males, and the authors further go on to explain that this may be due to males and Caucasians having a greater proportion of fat in the abdominal area (visceral and/or subcutaneous fact).  All this being well known, it would be nice to see if there was an interaction between gender and nationality in conjunction with fat distibution area. For instance, does the effect on abdominal fat tendencies with Barreth's E ?  These types of questions can be answered by throwing interaction terms in the model  and by utilizing a multi-variate regresssion technique which I will get into with the next section.

Commentary on Statistics and Study Design:  I really liked some of the statistical consideratiosn in this papers.  Most notably, it was good to see that the authors used the concept of matching.  Matching is done to balance out the effects of any confouding variable in an observational study and should always be done if the investigator is highly suspicious of any confounding variable in the study.   This is especially pertinent to this study since it’s thought that gender may be highly associated with BE.  You don’t see many investigators do this even though it should be done for many investigations.   Also, I liked how the author’s gave explicit detail on the fat measurements in the Methods section.     

                However, I would have taken a different statistical approach to analyzing the relationship between the fat distribution variables and the BE response variable.  The authors investigated the uni-variate and bi-variate relationship (while controlling for BMI) between the fat distribution variables and the BE response variable.  In doing this, the author’s have obviously flagged BMI as an essential variable due to past literature knowledge for BE, and this would  seem to be correct.  However, for the fat distribution variables, I would have investigated the nature of the relationship between the fat distribution variables and the BE response while controlling for all variables – not just BMI.  This can be done with a typical multi-variate regression technique which includes all variables, and then the investigators could have simpy eliminated the un-necessary variables from the model in a backwards-stepwise approach.  At the end, one would then report both the uni- and multi- variate results.  Taking this approach would allow the investigator to come to a small set of variables which are most associated with BE. For instance, if may be that GEJ fat is not associated with BE while also controlling for Visceral fat, and GEJ fat can thus be eliminated as a significant predictor variable.  Furthermore, using this approach, one could also look at the presence of different interaction variables - which as I alluded to in the discussion section - may be present here.  For whatever reason, many investigators do not include the presence of interaction variables even though they should.  

                From a presentation standpoint, for the second aim in Figure 3, the author’s decided to present a box and whisker plot instead of a table.  Since one is only looking an uni-variate associates here, this seems to be OK.  Personally, for uni-variate assocations, I like the use of a box-and whisker plot more than a table since the actual numbers are visually more explicit, but either one works well.  As long as the knowledge comes across, it’s all good.

                A big thanks to our Mayo Clinic buddies for doing this!