Epidemiology Expert: Pooled Results Confirm No Association with Cancer

Harvard epidemiologist, Dr. Lorelei Mucci, testified during the Pilliod v. Monsanto trial that two recent “pooled” population or epidemiological studies conclusively demonstrated there is no association between glyphosate based formulations (GBFs) such as Roundup and Non- Hodgkin’s Lymphoma (NHL). The two pooled studies represent a body of evidence more than five times larger than the evidence for the epidemiological conclusion limited by chance, bias and confounding of the International Agency for Cancer Research (IARC) in 2015 and concurrently resolved the IARC conclusion-limiting issues. She also noted during cross-examination the plaintiff’s quest for biological plausibility of cancer via animal and mechanistic tests for a non-existent human cancer association is not meaningful.

One of the “pooled” population or epidemiological studies refers to the soon to be published North American Pooled Project (NAPP) which pooled the results of McDuffie and De Roos (2003) studies. The De Roos study was a linchpin of the IARC epidemiological conclusion and is also heavily relied upon by plaintiffs’ Roundup cases. The NAPP study paper was drafted in 2015 and an abstract and presentation materials were peer-reviewed and presented at professional conferences. Dr. Mucci presented testimony and exhibits showing the NAPP results.

• During direct and cross examinations, Dr. Mucci testified regarding the hypothesis driven process of the NAPP study: NAPP includes includes the Canadian data from McDuffie as well as the Hohenadel study (Hohenadel et al 2011) which used the same data set for identification of the confounder relationship between GBFs and malathion insecticides. And then also all of the U.S. case-control studies that were included the publication from Dr. De Roos. And the NAPP study kind of a little bit different from these earlier studies because it was specifically addressing the hypothesis of whether glyphosate was associated with non-Hodgkin’s lymphoma,and the reason that’s important is that the way they sought to analyze the data was specific for glyphosate. Q. All right. And it’s larger, the number of cases there are more than the prior ones. A. Yeah, so much, much larger. You can see, you know, five to ten times larger than these individual studies with 113 exposed cases. So it’s, as I mentioned, the earlier publications, including Dr. De Roos’ study, weren’t specifically looking at Roundup. They were looking — in that particular study they were looking at 47 different pesticides. The approach they took was to put all of the 47 pesticides into these mathematical models. And the challenge with that earlier study was there were only 36 cases exposed to Roundup. And if you have 47 different pesticides, you’re going to have some pesticides for which there’s no exposed cases and that can cause a problem in your analysis. Q. And what’s that called in epidemiology? A. We call that that a sparse data bias (Sparse Data Bias, Greenland et al 2014) And what happens is because you have very few to no cases in specific cells, it can lead to your estimates what we say as being unstable and so it can lead to spurious associations or getting the wrong answer. So — and the reason that I — so that looking at the approach that was taken in the NAPP study was the correct approach that we do in terms of adjusting for confounding. That’s kind of the standard epidemiology approach where you look at a specific exposure and disease and try to identify what are the specific confounders in this set of data for that exposure and disease.
• NAPP multivariable analysis adjusting for use of three pesticides and excluding proxy respondents reflected an odds ratio for ever use of GBFs of 0.95 with a 95% confidence interval (CI) of 0.69–1.32 for the pooled analysis. The lifetime days of GBF exposure risk metric (>7 days) , when restricted to self-respondents, resulted in odds ratio of 1.06 with 95% CI (0.62–1.81).
• The August 2015 NAPP presentation reported 45 cases of NHL sub-type of diffuse large B-cell lymphoma (DLBCL) and adjusted for use of the three pesticides provided an ever use odds ratio of 1.23 with 95% CI (0.81–1.88). The sub-type analysis did not present the results for only self respondents vs. proxy respondents.

The second pooled study is from AGRICOH, a program of the IARC which supports data pooling of agricultural cohort studies for improved analysis of disease-exposure associations. As part of this program, Leon et al, published a pooled cohort analysis on March 18, 2019. The cohort groups included those from AGRICAN, a program of the Mutualité Sociale Agricole, the French national health insurance system of agricultural workers; CNAP, an aggregate group of farm holders and families compiled by Statistics Norway; and the Agricultural Health Study (AHS), which included farmers and farm workers from Iowa and North Carolina. The pooled data set included more than 300,000 subjects with potential exposure to 33 pesticides tracked for 3.5 million years of follow-up for an association with 2,400 NHL cases. Notably, the follow-up time was longer in the AHS publication (up to 2012 and 2013), and thus more cases were included than in the AGRICOH analysis (130 vs 113 DLBCL cases). The primary analysis provided risk metrics for ever use GBFs of .96 with CI(.77–1.18) for NHL and 1.36 with CI(1.00–1.86) for DLBCL. Testimony of Dr. Mucci and alternative analysis within Leon study provided more up-to-date insights:

-Trial cross-examination of Dr. Mucci: Q.Which (Leon study) shows a statistically significant increased risk for diffuse large B-cell lymphoma; true? A. I actually don’t — it’s — it’s — it’s probably borderline significant. I’m not going to argue with that. But it also didn’t include the most up-to-date AHS data. And as I showed, when you include that data it goes from 1.36 to 1.21 and is not significant.

-The Leon paper also includes an alternative calculation with a minimal set of variables for NHL and diffuse large B-cell lymphoma (DLBCL). Supplemental Table 3 shows ever use and meta-analysis estimates, minimally adjusted, of NHL and sub-type DLBCL for GBFs:

• 1131 NHL cases with Hazard Ratio risk measure of .98 95%CI(.76–1.25).
• 221 DLBCL cases and Hazard Ratio of 1.12 with 95%CI(.86–1.45).

The importance of adjusting risk measures for other pesticides was mentioned several times during the examinations of Dr. Mucci, but the following excerpt from the direct examination relates to a “hypothetical” differential analysis (etiology) of plaintiff’s expert in ruling in GBFs based on dose response risk estimates from Ericsson and McDuffie studies which were not adjusted for other pesticides:

Q. If the jury heard from Dr. Nabhan that in looking at dose-response issues you don’t have to even consider confounding, would you agree or disagree with that, if that’s what they heard? A. I would disagree. Q. Why? A. So in epidemiology when you’re looking, for example, at dose-response and you see an association, the first thing you need to ask is could bias or confounding have led to that dose — apparent dose-response. Again, a statistical association does not mean a causal association. So you first want to rule out that there’s bias and confounding. And so that’s why it’s incredibly important to always adjust for confounding. And in fact, actually there’s many examples where you get a dose-response because of confounding. Because those who are in the highest group of the exposure are much more likely, for example, to be exposed to other pesticides even more so than those in the lower level of exposure. Q. And the whole issue about whether you have to adjust for other pesticides when you’re looking at this issue, do you think that’s important or is that something that, you know, only a rookie would do or someone who’s sort of making core baseline epidemiology mistakes? A. No. In fact, it’s very critical. You know,confounding in epidemiology is one of the core issues we worry about. Again, we can’t do the time machine. And the reality is people who — I know you don’t like the physical activity example — but people who are, you know, physically active, they’re less likely to smoke and they’re more likely to eat a healthy diet and they are more likely to go regularly to the physicians. And so confounding is something as an epidemiologist we’re concerned about. And the good thing about it is there’s something we can actually do with it in our mathematical model. So it’s always something that we should be concerned about. And we should look within a study to see if confounding is present.

Dr. Mucci provided a day of highly pertinent testimony regarding the recent epidemiological studies which confirm the lack of association between GBFs and NHL and major sub-types. Whether juries appreciate the difference between isolated animal and laboratory tests and recent large scale conclusive studies on human populations remains to be seen. Ultimately, if juries cannot grasp the distinction, appellate courts have consistently overruled jury verdicts when epidemiological evidence derived from human populations was not appropriately valued.