New meta-study reinforces evidence that the herbicide glyphosate is a possible human carcinogen

Credit: Boasiedu/Wikimedia Commons

A new epidemiological study, published online earlier this week in Mutation Research, provides new evidence that the herbicide glyphosate--the most widely used pesticide in the world--may be associated with higher cancer risk. While as of this writing the journal has so far posted uncorrected proofs, and they are behind a paywall, the study finds that the highest exposed groups showed a 41% greater risk of non-Hodgkin lymphoma (NHL.)

I have read the study carefully, and I am discussing it with epidemiologists, as part of a feature story on the glyphosate debate I am preparing for a major science publication. In the meantime, however, science writer Carey Gillam--author of the award-winning book "Whitewash" about glyphosate and the decades-long efforts by Monsanto (now owned by Bayer, and manufacturer of the Roundup formulation of the herbicide) to suppress evidence of its dangers--has published an excellent article on the study and its implications in The Guardian. I strongly recommend you read that for the basic news.

The new research, led by toxicologist Luoping Zhang at the University of California, Berkeley, reports on a meta-analysis of six previous epidemiological studies of glyphosate-based herbicides and their possible association with NHL risk. Zhang, along with two other authors of the paper, were members of the Environmental Protection Agency's external Scientific Advisory Panel on glyphosate carcinogenicity, which met in 2016 to advise the agency on its updated evaluation of the herbicide's potential cancer risk. The six studies include a major cohort study, called the Agricultural Health Study (AHS), which did not, in and of itself, find a higher risk in glyphosate-exposed workers. However, each of the five other studies included in the analysis, known as case control studies, did find a positive association. (I will say more about the difference between cohort and case control studies in a moment.) By combining the studies together, and testing a novel hypothesis for glyphosate risk, the team found what it calls a "compelling link" between exposure to glyphosate-based herbicides and NHL risk.

The new study is potential bad news for Bayer, which is currently facing well over 9000 lawsuits from people exposed who are suffering from NHL and blame exposure to glyphosate/Roundup for their cancers. Last August, in a well reported verdict, a California jury awarded a school employee, DeWayne Johnson, nearly $300 million in damages for the NHL that will ultimately take his life. While the judge in the case reduced the award, it was a wakeup call for the German pharmaceutical giant and its investors that science was not necessarily on their side--despite decades of claims by Monsanto that glyphosate was safe.

The current round of fierce debate over glyphosate began in 2015, when the World Health Organization's International Agency for Research on Cancer (IARC), after a year of study, concluded that glyphosate was a "probable" human carcinogen. In response, Monsanto and its allies launched a ferocious attack on IARC and its credibility, which has been well documented in discovery documents from the lawsuits against the company currently under way. (The so-called "Monsanto Papers" also document decades of attempts to influence, and in many cases distort, the scientific process of glyphosate evaluations.) While IARC has done its best to defend itself of numerous false charges (many of which I have personally investigated in my own reporting), the campaign against the agency reached the U.S. Congress, where Republican representatives demanded that the National Institutes of Health cut its major financial contribution to the agency. (In contrast to all this, Monsanto's own representative at the IARC working group on glyphosate told me that its proceedings were entirely above board.)

In their attacks on IARC and its conclusions about glyphosate, Monsanto and its proxies (which includes scientists and industry front groups partly funded by the company, such as the American Council on Science and Health) branded the agency as an "outlier" among other regulatory bodies, such as the EPA and the European Food Safety Authority, which had concluded that glyphosate was safe--at least at normal exposures the public was likely to be exposed to. (Numerous studies have shown that the heavily used herbicide is ubiquitous in the environment, and traces are routinely found in food products.)

At the EPA, however, IARC's report on glyphosate attracted immediate attention among its scientists, who realized that the agency's earlier approvals of its use might need review. After much resistance from industry (which I will discuss in more detail in the article I am preparing), EPA finally did launch that process, which included convening the Scientific Advisory Panel. The SAP met in December 2016 (the 1261 page transcript of this meeting is available at this link.) The panel reviewed EPA's proposed evaluation, which was already in draft form and proposed to classify glyphosate as an "unlikely" human carcinogen; that evaluation had already essentially ignored input from its research wing that urged serious consideration of IARC's arguments.

In a wide-ranging discussion over several days, some members of the panel expressed reservations with EPA's proposed bill of health for glyphosate. Three of those members are now coauthors of the new meta-analysis: Berkeley's Zhang, Emanuela Taioli of the Icahn School of Medicine in New York, and Lianne Sheppard of the University of Washington in Seattle. As the transcript shows (and in interviews I conducted with all three of them) these researchers objected in various ways to the EPA's dismissing of case control studies which concluded, within the confidence limits of such studies, that glyphosate was a potential human carcinogen. In its evaluation of human epidemiology, the agency relied on an earlier version of the AHS cohort study, published in 2005, which like the most recent 2018 update (not available to EPA or the panel in 2016) did not show an association between pesticide exposure and cancer.

At this point let me pause to briefly discuss the difference between a cohort study and a case control study. In a typical cohort study, a population of individuals is identified that features a typical exposure or behavior--in the case of the AHS, pesticide applicators exposed routinely to glyphosate--along with a control population. Both populations are followed over time, to see if the exposed population shows a greater incidence of the phenomenon being studied, in this case cancers such as NHL. The bigger the cohort, and the longer the study continues, the more sensitive it is to any higher risk.

A case control study goes at the problem in pretty much the opposite direction. A population that already has the feature being studied--in this case, patients who have been diagnosed with NHL--is compared to a carefully matched control population to see how they differ. In this way, factors such as increased use of glyphosate can often be identified.

While cohort studies are often considered the "gold standard" of epidemiology, both have their strengths and weaknesses, as Zhang and her colleagues explain in the Mutation Research paper. Cohort studies are generally better at determining the exposure subjects have to a particular chemical, such as glyphosate, which can be determined when the study population is first identified. However, they require the recruiting of very large numbers of subjects, and researchers must wait years--sometimes decades--before enough cases (eg NHL) are diagnosed to provide a statistically significant finding (the current AHS study involves some 45,000 pesticide applicators who used glyphosate.)

The biggest weakness of case control studies is that they are subject to what epidemiologists call "recall bias." When researchers interview NHL sufferers, for example, they will ask them about their use of pesticides such as glyphosate. With years of news stories about the possible risk of cancer from Monsanto's Roundup and other glyphosate formulations, the possibility that a cancer victim will exaggerate their exposure in their minds must be taken into account. Although there are ways to correct for this statistically, the problem is not entirely insurmountable. On the other hand, as Zhang and her coworkers point out, case control studies are strong at identifying very rare cancers, because researchers do not have to wait years for them to pile up--the cases are already there.

In its evaluation of glyphosate, IARC took the case control studies already done very seriously, while the EPA tended to discount them. Zhang, Taioli, and Sheppard all raised concerns with this, and the meta-analysis represents their independent attempt to look into the issue further. Although a few meta-analyses had been conducted previously, none included the most recent version of the AHS study, which was an integral part of the new analysis. In addition, Zhang et al. decided to test a novel hypothesis. As they put it in the paper:

"Our a priori hypothesis is that the highest biologically relevant exposure to [glyphosate-based herbicides], i.e, higher levels, longer durations and/or with sufficient lag and latency, will lead to increased risk of NHL in humans. This hypothesis is based on the understanding that higher and longer cumulative exposures during a biologically relevant time window are likely to yield higher risk estimates. given the nature of cancer development. Hence, when cumulative exposure is higher, either due to higher level or longer duration exposures, an elevated association with the cancer of interest is more likely to be revealed if a true association exists. This a priori approach has been employed to estimate meta-risks for benzene and formaldehyde, but not in any of the previous meta-analyses exploring the [glyphosate-based herbicide]-NHL association."

To put it in simpler terms: The statistical and methodological difficulties in determining whether glyphosate causes cancer can be better addressed by focusing primarily on the most highly exposed subjects. If such an association is found, it may be possible to say more about the risk to lower-exposed populations.

The paper includes a critique of the AHS study and an examination of why it has failed so far to detect an association with NHL (although the latest version of the study did find a possible association with acute myeloid leukemia, largely ignored in media accounts, and which its authors urged be followed up.) I won't get into those details here. But the most important takeaway from the study is that even when the AHS is included, there is a clearly greater risk of NHL--41%--in the most highly exposed groups.

In her article in The Guardian, Carey Gillam quotes an EPA spokesperson saying that the agency is studying the paper, although Bayer had not responded to requests for comment when her article went to press. However, there is little doubt that Bayer is working on a response as I write, and we should see it soon. One can hope that it will be based on a serious evaluation of this new scientific study, which will be hard to dismiss out of hand given the reputations and qualifications of the researchers involved. If so, it would be a departure from Monsanto's and Bayer's previous behavior, which has largely consisted of attacking its critics--including many scientists--and trying to deflect all concerns. The DeWayne Johnson trial showed that such tactics are not working, and may be even less effective as thousands of court cases go forward.

Addendum Feb 14: One of the problems in assessing the risk that glyphosate and its formulations pose both to workers who use it as well as the general public--including vulnerable children--is that we don't have a very good idea just what those exposures are. This was underscored in another study published earlier this year, in which Zhang, Sheppard, and Taoili participated, published in the journal Environmental Health. This review of the literature (19 studies, of which five investigated occupational exposure to glyphosate, 11 documented exposure in general populations, and three reported on both) came to the following conclusion:


The current review highlights the paucity of data on glyphosate levels among individuals exposed occupationally, para-occupationally, or environmentally to the herbicide. As such, it is challenging to fully understand the extent of exposure overall and in vulnerable populations such as children. We recommend further work to evaluate exposure across populations and geographic regions, apportion the exposure sources (e.g., occupational, household use, food residues), and understand temporal trends.