Pesticide poisoning study resubmitted

Four members of the Pesticide Action Network (PAN) have resubmitted a new study on the global estimation of pesticide poisonings. The study was published on September 2 in the journal Frontiers in Public Health and claims the number of cases of acute accidental pesticide poisonings worldwide is between 402 and 433 million per year.

Study claims large portion of agricultural population affected

The study estimates that approximately 46 percent of the global agricultural population, which is around 934 million people, are victims of poisoning. The PAN, which is also the organization behind the research, claims there is no conflict of interest despite the fact that the authors are permanent members of the network.

New study addresses retraction of earlier research

The new study was intended to respond to the retraction of a similar publication from 2020, which was retracted in October 2024 due to methodological issues. The 2020 study had been previously examined in February 2023, which found that the French estimate was overestimated by a factor of 30, and that the Indian averages had been calculated without weighting based on samples of 80 to 1,577 people.

New study updates and expands previous research

The new study updates the scientific literature review up to 2023, covering 214 articles and 62 countries, and adds several sensitivity analyses. It arrives at an order of magnitude higher than the previous one. The authors mention the retraction of the previous study on the first page but do not contest it.

Confusing "cases" and "people" in the study

The first trick in the study consists of confusing "cases" and "people." The figure of 433 million refers to episodes of poisoning, not distinct individuals. The authors explicitly acknowledge this, indicating that the incidence of cases usually exceeds the incidence of poisoned people. However, they transform this total into "46 percent of poisoned farmers," which amounts to confusing cases and people.

Methodology raises concerns about case estimation

The publications collected list 369,632 actually observed cases. The study gets from this figure to 433 million cases through successive projections, by first applying the rates from local surveys to entire national agricultural populations, then extrapolating again from the 49 countries documented to the neighboring regions lacking data.

French case highlights methodological issues

The case of France shows the process. The authors attribute 45,839 annual cases to France, a figure raised to 105,862 once extrapolated to all of Western Europe. However, this is not a count but a rate of 6.1 percent applied to the French agricultural population. This rate comes from a single article devoted to the risk of asthma in the Agrican cohort.

Lack of clarity on what constitutes a poisoning

What is called a "poisoning" adds a more fundamental problem: opinions differ on what constitutes a poisoning. "There is no universally accepted definition," admit the authors. Therefore, each survey uses its own criteria, and the results vary accordingly: among the 140 aggregated surveys, the proportion of farmers declared poisoned ranges from 2 percent to 100 percent.

Randomized trial shows no biological evidence of poisoning

The authors also cite a randomized trial comparing users of chlorpyrifos to a group treated with neem oil under the same working conditions. However, both groups reported an identical frequency of nonspecific symptoms, while blood analyses showed no decrease in plasma cholinesterase, the enzyme blocked by organophosphates, whose collapse indicates poisoning by these products.

Study fails to address margin of error

Finally, the study never addresses the margin of error of its result. The range of 402 to 433 million does not respond to it: it measures the gap between five ways of sorting the same data. A true margin of error should add the uncertainty inherent to the size of each survey, the transition from a local territory studied to an entire country, and the regions devoid of any data. However, this calculation is done nowhere.