Statistics Versus Facts: A case study on aspartame

By Lisa Watson
A case study on aspartame
A recent communication "Aspartame intake is associated with greater glucose intolerance in individuals with obesity" has garnered some press attention and will no doubt be cited in future publications – at least by scientists who fail to actually read the brief and look at the data presented.
May 28, 2016

A recent "brief communication" Aspartame intake is associated with greater glucose intolerance in individuals with obesity1 has garnered some public press attention and will no doubt be cited in the Introduction section of future publications – at least by scientists who fail to actually read the brief and look at the data presented. It is an example of statistics interfering with an understanding of the broader observation.

Before examining this paper, it is important to note that there are well documented human clinical trials that have evaluated whether or not aspartame influences blood glucose levels and other indicators of diabetes or pre-diabetes2.  It does not. These clinical trials are not rodent experiments, which are subject to clear limitations in extrapolation to humans. They are not observational/epidemiological associations, such as the recent report, in which large pools of data may be mined using exhaustive statistical analyses until something "significant" is identified. No, these were carefully designed intervention research studies with real people (both with and without diabetes) in which dietary intake is known and controlled, the sweetener aspartame is consumed, and blood glucose and/or other meaningful indicators of metabolic factors are measured. These human studies demonstrate that aspartame has no effect on glucose levels, which is not surprising given the sweetener’s composition2. It is a simple ingredient made of components found in the food we eat every day, and our bodies break it down it just like other proteins quite soon after it’s consumed . . . just as food is digested.

The Kuk and Brown Study
Kuk and Brown obtained a subset of data collected as part of the much larger NHANES III survey between 1988 and 1994 (2856 people, or about 8.4% of the 33,994 total NHANES III participants). As part of NHANES, this smaller group had an overnight fast, and was administered a 2-hour glucose tolerance test. Their dietary intakes were assessed using a 24-hour recall. What the individuals reported eating the day before was assumed to represent their average diets.

The researchers divided these individuals in various ways. First the group was divided based on sucrose intake ("low/high"). Second, the same people were divided based on fructose intake ("low/high") in the previous 24 hours; next, the same individuals by aspartame intake (no consumption or a consumer); and last, the group was divided based on whether or not they consumed saccharin. They then looked at average characteristics of each subgroup, including BMI, fasting glucose, and insulin resistance.

A regression analysis was run for each group based on results of the 2-hour blood glucose levels and BMI. 

Findings
The researchers make sweeping statements in the paper itself, and in the press release from their institution promoting their study to the media, most of which mirror the study’s journal title suggesting a negative association in glucose tolerance with increasing BMI and attributed to aspartame. 

What do the data show?

  • There was no difference in fasting glucose among aspartame consumers and non-consumers: Fasting glucose levels were identical in the "aspartame" group (5.6mmol/L) as in 6 of the other 7 subgroups.
  • There was no difference in mean measurements of glucose tolerance between aspartame consumers and non-consumers.
  • There was no difference in mean measurements of insulin resistance between aspartame consumers and non-consumers.
  • Most important, the "steeper slope" the authors use to conclude that aspartame is related to "significantly greater impairments in glucose tolerance for individuals with obesity" is a false indicator because the low BMI aspartame consumers had lower (healthier) glucose tolerance than non-consumers. (See below.)

Misleading Statistics That Miss the Data
The authors' headline is that individuals with obesity who consumer aspartame may have "worse glucose management" than others3

The "evidence" cited to support this statement is a comparison of a regression line comparing 2-hr glucose results against increasing BMI in the various subgroups. At first glance, the "aspartame" group line does have a steeper slope, which could suggest an aspartame influence with increasing BMI. BUT in looking more closely, it is clear that the reason for the steeper slope is that the leaner individuals in the aspartame group had much lower 2-hr glucose levels than any of the other subgroups. (See diagram below.) Most of the low BMI groups had 2-hr levels of just at or slightly below 7, whereas the aspartame group was approximately 6. In fact, the authors state, "our results suggest a beneficial effect of aspartame in lean individuals." The "steeper slope" in the aspartame group with increasing BMI appears to be solely related to the fact that the starting point of the line was at a lower, presumably more positive, point. Had the 2-hr numbers among aspartame users with lower BMI been similar to the other low BMI groups, the entire basis for the authors’ declarations would be null. Yet this critical point is not addressed in the authors' discussion. 

Figure below is copied from Kuk and Brown. All notations in red were added by this author and are not part of the Kuk and Brown analysis. Red marks on y-axis highlight 2-hr levels among low BMI. Dotted line in graph D shows what slope of "aspartame" group would be had initial values been similar to other groups. 



According to the NIH (http://1.usa.gov/1UhVDM8), a glucose level of 7.8 mmol/L is clinically considered normal. Although data from the graph is difficult to interpret precisely, it suggests, as one would expect, a tendency toward increasing glucose intolerance with increasing BMI, although most mean subset values appear to fall within the normal range. This occurs across groups in the Kuk and Brown analysis.
 
Conclusion
There are many other concerns about the methodology of this paper that should suggest caution in extrapolation of the findings, including an apparent small sample size in some subgroups, accuracy of a 24-hour dietary recall, overlap across groups in terms of stated sweetener intake, no difference observed in fasting glucose and insulin resistance and the type/limited nature of statistical analyses carried out. Most important is the broad extrapolation of these findings in the face of human clinical trials with contrary findings. 

The question of whether low calorie sweeteners are helpful to those wishing to manage caloric intake has been the subject of exhaustive attention in recent years. Human clinical trials consistently point to modest benefits when low calorie sweeteners like aspartame are substituted for caloric sweeteners. It is important to look below the statistics in studies such as this one before making sweeping generalizations.



  1. Kuk JL and Brown RE. App. Physiol Nutr Metabol 41:1-4 (2016).
  2. See reviews: Magnuson et al. Crit Rev Tox 37:629-727 (2007); Renwick Physiol Behav 55:139-143 (2004); Maresk et al. Am J Clin Nutr. 95:283-289.
  3. " 'Our study shows that individuals with obesity who consume artificial sweeteners, particularly aspartame, may have worse glucose management than those who don't take sugar substitutes,' says Professor Jennifer Kuk, obesity researcher in the School of Kinesiology and Health Science." York University press release, May 24, 2016.

Lisa Watson is principal at Watson Green LLC, a food and nutrition consulting firm. She holds a Master of Science Degree in human nutrition from Clemson University.


A study on aspartame

Latest Blogs

Aspartame Resource Center
For scientific information about aspartame
Read more