The AHS classified subjects into quartiles by frequency of glyphosate exposure, ranging from fewer than 14 days to more than 108 lifetime exposure days. However, only 2 of the 5 case-control studies included information about duration of exposure. For one meta-analysis, the highest exposure is more than 10 days per year, while in the other study, high exposure was considered to be any exposure for more than 2 days per year. For the other 3 case-control studies, the exposure is simply classified as “ever” or “never.”
Another curious choice was to look only at the longest latency period, whenever available. The AHS included results for 5-year, 10-year, 15-year, and 20-year lag periods between exposure and diagnosis. The authors chose to include only the 20-year lag result in the meta-analysis. “We selected the longest lag or latency because decades may be needed for the health effects of many environmental toxicants to manifest as detectable cancers,” they wrote. However, the 20-year lag data showed the strongest association between NHL and glyphosate (RR, 1.12; CI, 0.83-1.51), spurring the suggestion by critics of cherry-picking. It is also interesting to note that most of the case-control studies were conducted less than 20 years after glyphosate was introduced onto the market in 1974, so the cancers observed across these studies necessarily had a shorter latency period.
Another reason to use the 20-year lag data, explained Dr Sheppard, was to minimize the problem of incomplete follow-up. In the 2017 update, 37% of participants — more than 20,000 people — did not complete the follow-up questionnaire to provide updated exposure information.
To fill in the missing data from the 2017 analysis, the AHS authors used a well-known technique called multiple imputation — but Dr Sheppard said their implementation of the technique may have biased the effect estimate toward the null. “It also didn’t necessarily reflect accurately the typical exposure in the study population,” she said. Because trends in glyphosate usage have dramatically increased, small biases in the modeling could throw off the final risk assessment. Using the 20-year lag data, the researchers thought, should sidestep potential problems with the imputation.
To perform a high-quality meta-analysis, it’s necessary to compare studies that are relatively similar, and some have questioned whether these older studies meet that standard. The cohort study, which provides the largest number of cancer cases, found no increased risk of NHL with glyphosate exposure, while the case-control studies vary considerably in how they were conducted. By their nature, as well, case-control studies can be troubled by recall bias, in which an individual with cancer may recall exposure patterns differently than someone who has not become sick.
The study authors did point out the limitations of their results: “41% is an important number, but you want to be careful to at least acknowledge that it’s an average of things that are similar but different,” Dr Sheppard said. But she stands behind the validity of the case-control studies. “I think that there are a lot of people, in this realm in particular, that want to discount the case-control result because the AHS isn’t showing anything,” she says. “All of us, we look at that evidence with our own lens.”