Thursday, June 20, 2013

GM Oh No! Part 1.5 of 3: Pig health considerations

A blog of Bridge Environment, updated most Thursdays

This entry is a follow up to the first in a three-part series about genetically modified organisms (GMOs)

Which technological creations do pigs fear more,
GMO feed or mutant super birds?
Last week, I defined, put into historical context, and considered the human health effects of genetically modified (GM) food items, also known as GMOs. While promising to explore environmental and political/economic considerations in future blogs, I concluded that human health concerns were more modest than critics would have you believe, but that there is potential value in conducting long-term testing of food varieties, GM or otherwise, that differ in substantial ways from their predecessors. As I will describe in the political/economic post, there may also be reasons to worry about financial and political influences on the approval process.

About the same time as that blog posting, a scientific study was published on the health of pigs which ate a GM diet (Carman et al. 2013). The authors used what one might call a shotgun approach. Instead of a targeted study with a specific concern in mind, they examined many anatomical and biochemical characteristics of 84 slaughtered pigs fed a GMO diet and 84 fed an equivalent non-GMO diet. Out of the many characteristics they tested, they found differences they described as statistically significant for two characteristics: uterine size and the rate of severe stomach inflammation. Though there have been a few balanced blog posts and media reports about this study, the majority used headlines such as “GMO feed turns pig stomachs to mush!,” perhaps not surprising for a source called Natural News. But MSN Now ran “GMO feed wreaked havoc on pigs’ stomachs.” If general media reports are to be believed, this study confirmed our fears: GMO foods do horrible things to our health.

Is it true? Was my advice last week off-base? Let me start by reassuring you that the study does not change my conclusions at all. Here is why.

First, let’s consider uteruses. According to the authors, they were 25% larger in pigs fed GMO corn and soy (median 0.105% of total body weight versus 0.086% body weight) and this difference was stastically different at the 2.5% level (known as a p-value, where p stands for probability). However, their claim is complicated and, in some cases misleading, due to several factors:
  • the math: as reported in their tables, uteruses of GMO-fed pigs were 22% larger than those of non-GMO-fed pigs, but this may be a typo since the results from the table do not match what is reported in the text;
  • attrition: several pigs died in the experiment (11 non-GM-fed pigs and 12 GM-fed pigs) and one non-GM-fed female pig failed to develop a uterus at all, so there may be a bias based on which pigs developed and survived until the end of the experiment; and
  • the health significance: we have no understanding of whether larger uteruses for pigs at this stage in development is a good or bad thing; in fact GMO-fed pigs were slightly larger at slaughter so the difference may simply indicate faster sexual maturity.
The authors’ claim of statistical significance raises even more concerns. Scientists typically only make strong claims about results if observed differences have a 5% or smaller chance of occurring due to random variation. Scientists picked the 5% p-value threshold because of a desire to maintain high standards prior to claiming that an observed difference is real. Even then, one in twenty times a scientist will report a meaningful finding that was simply due to random differences among similar individuals.

This grey area of scientific proof becomes far murkier when multiple comparisons are made. Because of the shotgun appraoch of this study, where one treatment was conducted and many comparisons were made, we would expect a far greater chance of an observed difference being due to chance than if only one observation had been made. My first real statistics professor referred to such shotgun approaches as p-ing all over the page, and this paper is guilty. For example, they measured eight separate organs from the same set of pigs. In each of those eight comparisons, there would be a 5% chance of mistakenly thinking there was an effect of GMO feed when in fact the difference was random chance. Collectively over the eight comparisons, there would be a 1 in 3 chance1 of thinking at least one organ size difference was attributable to diet when in fact the pigs were essentially the same. To correct for this multiple comparison bias, scientists are supposed to adjust the threshold for considering a result significant. In the case of eight comparisons, the new standard of significance would be 0.64% for each organ, and the authors’ p-value of 2.5% would not be adequate to claim a true difference between the uteruses of GMO- and non-GMO-fed pigs.

Inflamed stomachs were even more problematic than large uteruses. The authors claim that severe inflammation occurred over 2.5 times more often in GM-fed pigs and that the difference at a p-value of 0.4%. However, the authors made 16 separate comparisons of pathological conditions. To correct for the multiple comparisons, they should have adjusted their significance level per condition down to 0.32%. Once again, valid use of statistics would keep them from claiming a true difference. It is even more interesting when we examine the other observed differences between GM-fed and non-GM-fed pigs, many of which were related to the stomach. Whereas GM-fed pigs more often had severe stomach inflammation, they also more commonly had no inflammation, and less often had mild or moderate inflammation. There are statistical tests to compare multiple category data like these, but it is not surprising that the authors failed to use them considering their failure to address multiple comparisons. Had they performed it, such an analysis would have provided ambivalent results because of the fact that GM-fed pigs had higher incidence of stomach health but also of severe inflammation. GM-fed pigs also had lower incidence of stomach erosion, pin-point ulcers, and bleeding ulcers, but higher incidence of frank ulcers (not sure what they are…aren’t all ulcers honest?). GM-fed pigs also had lower incidence of heart, liver, and spleen abnormalitlies. Mind you, none of these differences were statistically significant, either, so all of this analysis should be taken with a very large grain of salt.

What really stands out for me in this study is not the effect of a GM diet, but the condition of all pigs raised commercially for meat production. Over the course of these pigs’ short lifetime (less than six months), more than one in eight died prior to slaughter, even with veterinary treatment. The article reassures us that these death rates are “within expected rates for US commercial piggeries.” Of the survivors, more than 1 in 10 had heart abnormalities, 1 in 5 had abnormal lymph nodes, over half had moderate to severe stomach inflammation, nearly 3 in 5 had pneumonia, and 4 in 5 had stomach erosions. The condition of these animals definitely makes me ponder eating more seafood.

Back to GMO health effects…applying scientific standards for statistical interpretation, this study becomes inconclusive. We could choose to be like the authors and interpret trends in the data that may simply be a result of random chance. This exercise yields a complex picture without any obvious indication that GM-fed pigs were healthier or less healthy than their non-GMO-fed counterparts. Should we dismiss the findings entirely? I don’t think so. Some of those trends may be a result of real effects. However, follow up study would be necessary and should be focused on particular concerns and analyzed correctly. At this point, though, there still is no credible evidence of health effects associated with common GMO food supplies. I maintain my conclusions from last week, and promise to flesh out larger concerns surrounding environmental impacts and political/economic influence over the coming weeks.

News outlets that presented this research otherwise have shown you their lack of respect for understanding science and, purposely or inadvertently, played on our human tendency to panic over uncertainties. You might want to consider better news sources in the future.


For more information, read our other blog posts and visit us at Bridge Environment.

1 If each comparison is treated as significant when the statistics report a 5% chance of mistaking random variation for a true result, then each has a 95% chance of correctly identifying random differences as being just that. To get it correct for eight different comparisons, we have to multiply 0.95 by itself eight times, 0.958 = 0.66. In other words, the likelihood that we correctly eight observed differences as due to random chance is only 66%, leaving a 34% chance…one in three, of seeing at least one false positive.