Difficulties in establishing a causal link between chemical exposures and cancer cannot be overcome by court assessments

Abstract

Scientific data are often used in lawsuits to prove, or dismiss, causation by a claimed factor of a claimed disease. Recent media reports of million-dollar compensations awarded to some cancer patients who had been exposed to certain chemical substances motivated me to examine how solid the causal links really were. Here, I discuss the limitations of epidemiological research on cancer causation and highlight how new knowledge of cancer genetics makes it unrealistic to expect that cancer causation can be clearly demonstrated. I then present two exposure–cancer cases, namely talcum powder–ovarian cancer and glyphosate–non-Hodgkin lymphoma, that led to civil lawsuits decided, in the United States, in favor of the claimants. Both these cancers have several risk factors, among which the claimed exposure presents only a minor, if any, increased risk. Through these cases, I explain why the use of epidemiological data is inappropriate to define causal associations in complex diseases like cancer. I close by suggesting a fairer approach, called proportional liability, to resolving future cancer litigation cases.

Keywords

Asbestos associations causal inference estimates glyphosate tumor heterogeneity non-Hodgkin lymphoma mutation load observational research ovarian cancer risk factors talc

Introduction

In 1965, Hill presented nine “features to be specially considered” when determining causation from observational data.¹ These features (or “viewpoints from…which we should study association”) regard the adequacy of the epidemiological evidence and the coherence of the evidence with additional data; they include strength, consistency, specificity, temporality, biological gradient, plausibility, coherence, experiment, and analogy. Since then, several dozen papers have discussed the adequacy or not of epidemiological results for assessing causality between an exposure and a disease, in particular cancer.^2

–7 Most of these papers proposed variants of Hill’s list, but none established a set of criteria or a pipeline to follow that, if satisfied, would ascertain causality between an exposure and a disease in an individual.

When deciding about causation, we should remember that epidemiological studies on the risk of cancer associated with any kind of environmental exposure are often limited by uncertainties regarding the extent of exposure, especially when direct measurements are not available or multiple environmental exposures occurred.^8,9 Uncertainties also regard the diagnosis, which is usually taken from death certificates. The accuracy of the declared cause of death may differ for different cancer types^10,11 and among different registers and countries.

Epidemiological studies can analyze the risk of disease in a specific subpopulation (cohort) compared to the general population or in a series of patients (cases) compared to healthy controls. Both study designs have their pros and cons. Cohort studies are suitable for assessing associations between exposures and different endpoints, for example the incidence of or mortality from different types of cancer. In these studies, the study population is defined in terms of exposure to a risk factor and is followed over time. Therefore, this design lends itself to prospective studies. Prospective cohort studies are more accurate in assessing risk factor exposure than are case–control studies. In case–control studies, the population is defined in terms of disease diagnosis. In other words, patients who have developed a particular disease are identified and then matched with healthy controls extracted from the same general population and presenting approximately the same characteristics of age, sex, ethnicity, and so on. Case–control studies are always retrospective, and exposure to the risk factor of interest is ascertained in different ways, often through indirect methods such as the use of questionnaires.

Both study designs assume that the factors that make each person different, for example age, sex, lifestyle, germline DNA variants, access to medical care, and exposure to pathogens and toxins, can be controlled for by analyzing sufficiently large populations. In other words, epidemiology assumes that unknown positive and negative risk factors in the individual annul each other in the population by their algebraic summation. However, the validity of this assumption has never been scientifically demonstrated.

The relative likelihood of a given disease occurring between two groups that differ in exposure or condition is typically expressed using the odds ratio (OR) or the relative risk (RR). If a disease is rare in the general population, then OR is similar to RR.^12,13 The larger the OR or RR, the more likely the disease is associated with the factor. However, if the confidence interval (CI) for the OR or RR includes 1, then statistical significance is not reached irrespective of the value of the OR or RR. Many statistically significant associations between cancer and an environmental or occupational risk factor have ORs or RRs in the range of 1.1–1.3, meaning that exposure to such factors increases disease risk by just a small amount. This epidemiologic evidence does not reach the threshold of the legal burden of proof termed “more likely than not” or “more probable than not” that is required in most civil lawsuits in the United States. In epidemiological terms, this burden of proof corresponds to an RR > 2.0 (i.e. the incidence rate in the exposed population is twice that in the nonexposed controls) and means that the risk attributable to an exposure, beyond the underlying risk due to other causes, exceeds 50%.¹⁴

Different epidemiological studies on a particular exposure–disease association often find widely different risks because, in each study, the analyzed groups are small compared with the general population. Moreover, the characteristics of exposed persons or other aspects of study design may vary among studies (called “study heterogeneity”). Thus, one study may find that a given exposure is associated with an excess relative likelihood of disease occurrence (i.e. OR or RR > 1), while other studies may show, instead, that the same exposure has no effect on or even reduces the relative likelihood of a given disease.

Because of the limitations of one or a few epidemiological studies, meta-analyses have become popular for the assessment of risks associated with a given exposure, lifestyle, condition, genetic variation, and so on. Meta-analyses combine the results of individual studies to reach cumulative conclusions, which can be considered more solid. The results of meta-analyses are weighted means of the results of individual studies and are expressed as meta-ORs and meta-RRs. An increasing number of meta-analyses is published each year, with redundant meta-analyses of the same studies or of overlapping sets of studies being published, sometimes with conflicting results.¹⁵ To overcome problems in the selection of original studies and in reporting methods and results of meta-analyses and systematic reviews, the PRISMA reporting guideline has been developed.¹⁶ However, even when a meta-analysis shows a significant association, if there is study heterogeneity it is difficult to conclude that a particular exposure causes the disease of interest. This approach to calculating risk, therefore, has limitations.

To overcome limitations in current methods of assessing risk, Keyes and Galea are proposing the use of a “causal architecture approach,” which focuses on understanding the complex causal networks underlying a disease rather than the effects of individual exposures.¹⁷ Despite the novelty of this approach, a protocol on how to assign a causal link between exposure and disease in an individual is lacking. As I discuss below, there is no agreed method for the allocation of causality between an exposure and a disease in an individual because, on the basis of current scientific knowledge, this is simply not possible.

Mechanics of cancer causation and development

Today, it is not possible to determine the exact cause of a given cancer in an individual. However, progress in genetics and immunology has shown that cancer causation, growth, and development are associated with particular somatic mutations, several of which are attributed to specific environmental exposures¹⁸ and others to chance in the unchecked propagation of DNA replication errors.^19,20 Cancer risk and prognosis also depend on an individual’s germline constitution^21

–24 and immunological status.^25,26

Environmental exposure to toxins can result in a specific mutational profile, or “signature,” in tumors. For example, in patients with lung cancer, the total number of somatic mutations (mutation load) and the frequency of GC>TA mutations are both strongly correlated with smoking history.^27
–29 In skin melanoma, unsuccessful repair of ultraviolet-induced DNA lesions leads to a predominance of C>T transitions.^30,31 The specific mutational profile depends, however, on the individual’s genetic background. For example, in melanoma, the mutation load was found to be higher in patients whose melanocortin 1 receptor gene had one or two R alleles (with genetic variants associated with red hair color and fair skin) than in patients with only r alleles.³²

Different individual exposures to cancer risk factors, and the different responses based on genetic and immunological status, manifest as huge intertumoral heterogeneity. Cancer genome sequencing has revealed that tumors arising in the same organ and having the same histological characteristics display considerable differences among patients in terms of genomic changes, with each tumor having its own set of somatic mutations in its own set of mutated genes, depending on the specific combination of risk factors.^33

–36 This scenario is illustrated in Figure 1, where different hypothetical tumors, of the same tissue origin and histological type, have different but overlapping sets of somatic mutations in different genes. This heterogeneous pattern of somatic mutations has been observed in many tumor types, including ovarian cancer³⁷ and non-Hodgkin lymphoma.³⁸

Figure 1.

Distribution of somatic mutations across 8 genes, in a hypothetical series of 200 tumors of the same tissue origin and histological type. In the Oncoprint (a type of matrix table), individual tumors are presented in columns, and individual genes are presented in rows. The bar plot at the top indicates the number of mutated genes per tumor. Values on the left indicate, for each gene, the percentage of tumors in which the gene is in a mutated state, while the bar charts on the right show the number of mutations in each gene. The columns are arranged so as to group mutations by gene, and the rows are arranged from the most mutated to the least mutated gene. Exposures to different risk factors and chance can cause mutations in the genes. Each risk factor can potentially act on different genes and each gene can be mutated by more than one risk factor.

Case studies

Talcum powder and ovarian cancer

Talc, a mineral composed of magnesium, silicon, oxygen, and hydrogen, is widely used in beauty care products. Talcum powder is sometimes used by women on their genitals or sanitary napkins. However, because talc mineral can be contaminated with asbestos, groups of women in the United States filed claims that their genital use of talcum powder caused ovarian cancer. From 2016 to 2018, courts ordered the Johnson & Johnson company, the main producer of talcum powder, to pay several hundred million dollars to these women (https://www.bloomberg.com/news/articles/2018-12-21/j-j-s-tainted-talc-risk-expands-as-cancer-trials-triple-in-2019).

The chance of exposure to asbestos through talcum powder may be very low. Industrial producers claimed that asbestos fiber content in talcum powder was under the detectable level, according to the analytical methodology used in the 1970s, but undetectable does not mean asbestos-free, as recently argued.³⁹ Still, an analysis using up-to-date analytical methods did not find asbestos fibers in six talcum powder-containing products sold between 1940 and 1977.⁴⁰ In that period, there was widespread exposure of the population to asbestos, due to its extensive use in many consumer products.⁴¹ Furthermore, asbestos fibers were found in indoor air of certain buildings⁴² and even in outdoor air samples from the 1960s through the early 1980s.⁴³

Ovarian cancer is a common cancer. Each year in the United States, more than 22,000 women are diagnosed with ovarian cancer and about 14,000 women die from it.⁴⁴ Ovarian cancer has several known risk factors (Table 1). Among these, germline mutations in BRCA1 gene pose the highest risk. Other important risk factors are BRCA2 gene mutations (RR = 8.4) and a family history of breast or ovarian cancer (RR = 2.69–3.49), followed by asbestos exposure and genital talcum powder usage with lower risk estimates.

Table 1.

Main factors known to modulate the risk of ovarian cancer.

Risk factor	Type of estimate	Risk estimate	Reference
Age at menopause (≥53 vs. <45 years)	OR	1.54 (1.18–1.99)	La Vecchia⁴⁵
Asbestos exposure	meta-SMR	1.77 (1.37–2.28)	Camargo et al.⁴⁶
	meta-SMR	1.75 (1.45–2.10)	Reid et al.⁴⁷
Family history of ovarian cancer^a	SIR	2.69 (mother-daughter),3.49 (sister-sister)	Hemminki et al.⁴⁸
	FRR	2.96 (2.35–3.72)	Jervis et al.⁴⁹
Genital talcum powder use	OR	1.24 (1.15–1.33)	Terry et al⁵⁰
	meta-RR	1.22 (1.13–1.30)	Berge et al.⁵¹
Genital powder use	HR	1.09 (0.97–1.23)	O’Brien et al.⁵²
Germline mutations in BRCA1 gene^b	RR	50 (22–114)	Antoniou et al.⁵³
Germline mutations in BRCA2 gene^b	RR	8.4 (2.2–32)	Antoniou et al.⁵³
Obesity	meta-RR	1.28 (1.16–1.41)	Liu et al.⁵⁴
Olive oil consumption^c	OR	0.68 (0.50–0.93)	La Vecchia⁴⁵
Oral contraceptive use (ever vs. never)	meta-OR	0.73 (0.66–0.81)	Havrilesky et al.⁵⁵

FRR: familial relative risk; HR: hazard ratio; meta-OR: meta (or pooled) odds ratio; meta-RR: meta (summary, overall or pooled) relative risk; meta-SMR: meta (summary, overall or pooled) standardized mortality ratio; OR: odds ratio; RR: relative risk; SIR: standardized incidence ratio.

^a One affected relative.

^b Values are for the age group 60–69 years, and RR refers to the incidence of ovarian cancer in England and Wales in 1973–1977.

^c Fifth quintile of consumption versus first quintile.

The association between ovarian cancer and asbestos exposure (by any route) was deemed “causal” by the International Agency for Research on Cancer (IARC),⁵⁶ resulting in asbestos being listed in group 1 of its four-stage classification of carcinogenicity. This assessment was validated in a meta-analysis by Camargo et al., who calculated an overall pooled standardized mortality ratio (SMR) of 1.77 (95% CI: 1.37–2.28).⁴⁶ Effect estimates were stronger for heavy asbestos exposure, that is, in cohorts that received compensation for asbestosis and in those with an SMR > 2.0 for asbestos-related lung cancer. Another meta-analysis by Reid et al.,⁴⁷ published the same year, found a meta-SMR of 1.75 (95% CI: 1.45–2.10). However, in a sub analysis of four studies that examined cancer incidence, there was no significant association between asbestos exposure and ovarian cancer risk (effect size, 1.29; 95% CI: 0.97–1.73). These authors suggested that the reported associations may be due to misclassification of peritoneal mesothelioma as ovarian cancer. The slightly different findings and conclusions of these two meta-analyses may be due to differences in the selection of studies: Camargo et al. included 18 cohort studies, while Reid et al. included 14 cohort and two case–control studies; 11 cohort studies were included in both analyses.

Regarding the genital use of talcum powder, IARC found little evidence for its carcinogenicity (in 2006) and concluded that “[p]erineal use of talc-based body powder is possibly carcinogenic to humans ([g]roup 2B).”^57,58 A pooled analysis of eight case–control studies (total, 8525 cases and 9859 controls), published in 2013, found an increased risk of epithelial ovarian cancer (OR = 1.24; 95% CI: 1.15–1.33) relative to women who never used powder.⁵⁰ The study did not find a significant trend in risk with increasing number of lifetime applications nor an increase in risk among women who reported only nongenital powder use. These results were confirmed in 2018 by two meta-analyses. Berge et al.⁵¹ reported a meta-RR of 1.22 (95% CI: 1.13–1.30) for ovarian cancer associated with ever use of genital talcum powder. Penninkilampi and Eslick⁵⁹ reported a meta-OR of 1.31 (95% CI: 1.24–1.39) for ovarian cancer associated with any perineal talc use; they also observed a slight dose–response relationship. However, when cohort studies were analyzed, no significant associations were observed.^51,59 It should also be noted that numerous confounders, such as douching, are often not included in epidemiologic studies and thus remain as unknown factors. Indeed, in the Sister Study on 50,884 women in the United States and Puerto Rico, no association between talc use and ovarian cancer risk was observed (HR = 0.73; 95% CI: 0.44–1.2), but douching was associated with an increased risk of ovarian cancer (HR = 1.8, 95% CI: 1.2–2.8).⁶⁰ A large new multicohort study of over 250,000 women found no significant association between the use of powder (typically talc) in the genital area and the incidence of ovarian cancer (frequent use vs. never use: HR = 1.09; 95% CI: 0.97–1.23).⁵²

Little is known about how talc or asbestos could cause ovarian cancer. It is not known, for example, if ovary is subject to the same mechanism of asbestos carcinogenesis that leads to pleural mesothelioma and lung cancer, namely the persistence of inhaled asbestos fibers causing chronic inflammation.^61
–63 Asbestos fibers can reach the ovary.⁶⁴ However, to my knowledge, there is no evidence that ovarian asbestos fibers cause ovarian inflammation. Ovarian inflammation is associated with obesity,⁶⁵ which itself is a risk factor for ovarian cancer.⁵⁴ Finally, the epidemiological evidence for the association of ovarian cancer with asbestos is not supported by experimental evidence. Indeed, studies conducted in the late 1970s found that, in rodents, a lifetime diet containing 1% amosite or crocidolite (types of asbestos) had no carcinogenic effects.^66
–68 This lack of mechanistic or experimental data means that two of Hill’s features (plausibility and experiment) are not met. As shown in Table 2, the other features are similarly not fulfilled.

Table 2.

Lack of fulfillment of Hill’s features by existing scientific data on the association between talc exposure and ovarian cancer risk.

Feature	Comment	Fulfillment
Strength	Estimates of increased risk are not consistent across studies. Two meta-analyses reported estimates of 1.22⁵¹ and 1.31⁵⁹ for genital exposure to talc, indicating a modest increase in risk, while two others reported estimates of 1.77⁴⁶ and 1.75⁴⁷ for any exposure to asbestos, indicating a larger increase in risk.	No
Consistency	Individual clinical studies found either no association or only a weak association between talc use and ovarian cancer.	No
Specificity	Misclassification of peritoneal mesothelioma as ovarian cancer on death certificates may have occurred.⁴⁷ The studies are not specific for exposure, as asbestos fibers were not detected in six talcum powders sold in 1940–1977.^39,40 Association was limited to the serous carcinoma histotype ⁵¹ or to serous and endometrioid histotypes.⁵⁹ Widespread exposure of the general population to asbestos-containing products occurred until the 1980s.	No
Temporality	Ovarian cancer was diagnosed after exposure to talc. Not all studies reported the timing of talc exposure or the latency of the disease.	Not certain
Biologic gradient	Overall, an exposure–response relationship was inconsistent or not demonstrated in most studies. A “suggestion of dose-response” was reported in Berge et al.⁵¹	No
Plausibility	Mechanistic data on talc- or asbestos-induced ovarian cancer are missing.	No
Coherence	Differences in risk estimates have been reported in individual studies, with variation also associated with study design, that is, case–control versus cohort studies.	No
Experiment	Lifetime treatment of rodents with asbestos in their diet was not carcinogenic for the ovary.	No
Analogy	Not applicable.	No

The strongest risk factors for this cancer are genetic (Table 1). In patients without these germline variations, weak risk factors such as environmental exposure and individual habits or just chance may be the only plausible causal factors, as pointed out for most types of cancer.¹⁹ Therefore, in an individual with a diagnosis of ovarian cancer, it is almost impossible to ascertain a causal relationship between the disease and exposure to asbestos or asbestos-contaminated talcum powder.

Glyphosate and non-Hodgkin lymphomas

Lymphomas are a large, heterogeneous group of tumors that originate from the clonal proliferation of lymphocytes. Lymphomas are often distinguished into Hodgkin and non-Hodgkin types, but this classification is now considered obsolete, particularly on the basis of advances in genetic pathology. This division has been abandoned by the World Health Organization, whose classification (from 2008 and revised in 2016⁶⁹) now lists dozens of different forms of non-Hodgkin lymphomas. Notwithstanding the new classification, most epidemiological studies still refer to Hodgkin and non-Hodgkin lymphomas for practicality and statistical power.

Non-Hodgkin lymphomas are common. In the United States, 125,850 new cases of non-Hodgkin lymphoid neoplasms were expected in 2016.⁷⁰ Established risk factors for non-Hodgkin lymphoma (Table 3) include autoimmune conditions, celiac disease, type 2 diabetes, some viral infections,⁸⁵ some eating habits, polymorphisms on chromosome 6,⁸⁶ and the occupations of carpentry, cleaning, farming, hairdressing, and spray-painting. One of these risk factors, namely farming, involves an exposure to the herbicide glyphosate.

Table 3.

Main known or suspected factors to modulate the risk of non-Hodgkin lymphomas.

Risk factor	Type of estimate	Risk estimate	Reference
Alcohol drinking	meta-RR	0.85 (0.79–0.91)	Tramacere et al.⁷¹
Any autoimmune condition	OR	1.26 (1.09–1.44)	Wang et al.⁷²
Celiac disease	meta-OR	2.61 (2.04–3.33)	Tio et al.⁷³
Dairy product consumption^a	meta-RR	1.20 (1.02–1.42)	Wang et al.⁷⁴
Diabetes mellitus, type 2	meta-OR	1.22 (1.07–1.39)	Castillo et al.⁷⁵
Fat consumption^a	meta-RR	1.26 (1.12–1.42)	Han et al.⁷⁶
Genetic loci at 6p21.32			Conde et al.⁷⁷
rs10484561 – G allele	meta-OR	1.95 (1.72–2.22)
rs7755224 – G allele	meta-OR	2.07 (1.76–2.42)
Glyphosate exposure, high levels	meta-RR	1.41 (1.13–1.75)	Zhang et al.⁷⁸
Glyphosate exposure, ever vs. never	OR	1.13 (0.84–1.51)	Pahwa et al.⁷⁹
Glyphosate exposure, >2 days/year	OR	1.73 (1.02–2.94)	Pahwa et al.⁷⁹
Gluten-sensitive enteropathy	meta-RR	4.42 (3.72–5.26)	Kane et al.⁸⁰
Hepatitis B virus infection	meta-OR	2.50 (2.20–2.83)	Li et al.⁸¹
Hepatitis C virus infection	meta-OR	1.69 (1.40–2.03)	Zhu et al.⁸²
Occupational carpentry	OR	1.42 (1.04–1.93)	‘t Mannetje et al.⁸³
Occupational cleaning	OR	1.17 (1.01–1.36)	‘t Mannetje et al.⁸³
Occupational farming	OR	1.26 (1.05–1.51)	‘t Mannetje et al.⁸³
Occupational hairdressing	OR	1.34 (1.02–1.74)	‘t Mannetje et al.⁸³
Occupational spray-painting	OR	2.07 (1.30–3.29)	‘t Mannetje et al.⁸³
Red meat consumption^a	meta-RR	1.22 (1.03–1.44)	Caini et al.⁸⁴

meta-OR: meta (summary, overall or pooled) odds ratio; meta-RR: meta (summary, overall or pooled) relative risk; OR: odds ratio.

^a Highest versus lowest category of consumption.

The IARC classified glyphosate as a probable human carcinogen (group 2A), stating that “[t]here is limited evidence in humans for the carcinogenicity of glyphosate. A positive association has been observed for non-Hodgkin lymphoma” (emphasis in the original).⁸⁷ In contrast, the European Food Safety Authority concluded that “glyphosate is unlikely to pose a carcinogenic hazard to humans and the evidence does not support classification with regard to its carcinogenic potential.”⁸⁸ The United States Environmental Protection Agency wrote (in April 2019) that it “continues to find that there are no risks to public health when glyphosate is used in accordance with its current label and that glyphosate is not a carcinogen.”⁸⁹ Similarly, the Canadian Pest Management Regulatory Agency concluded (in April 2017) that “[g]lyphosate is not genotoxic and is unlikely to pose a human cancer risk.”⁹⁰ The Joint Meeting on Pesticide Residues (JMPR) concluded that “…long-term dietary exposure to residues of glyphosate from uses considered by the JMPR is unlikely to present a public health concern.”⁹¹ A report from the European Chemicals Agency concluded that “…based on the epidemiological data as well as the data from long-term studies in rats and mice, taking a weight of evidence approach, no classification for carcinogenicity is warranted.”⁹²

Despite these official decisions that glyphosate is not carcinogenic, in August 2018, a court in San Francisco, United States, ordered Monsanto (producer of glyphosate-containing Roundup) to pay US$289 million to a patient with non-Hodgkin lymphoma that, according to the court, was caused by the chemical; the compensation was later reduced to US$78 million. In March 2019, another court in San Francisco granted compensation of US$80 million for a similar dispute (https://www.motherjones.com/environment/2019/03/glyphosate-roundup-cancer-non-hodgkin-lymphoma-epa-panel-hardeman-lawsuit-jury-verdict/).

The association between glyphosate exposure and lymphohematopoietic cancers, including non-Hodgkin lymphoma, was assessed in a meta-analysis published in 2016.⁹³ The study reported a meta-RR of 1.3 (95% CI: 1.0–1.6) for the association between glyphosate (ever vs. never use) and non-Hodgkin lymphoma risk. The authors concluded that a causal relationship cannot be established between glyphosate exposure and risk of any type of lymphohematopoietic cancer, because of the few studies available and the crude exposure metric, that is, ever or never use of glyphosate.

Two cohort studies also failed to demonstrate a link between glyphosate and non-Hodgkin lymphoma. The Agricultural Health Study enrolled, in the mid-1990s, over 50,000 licensed pesticide applicators from North Carolina and Iowa (United States), of whom 44,932 (82.8%) used glyphosate.⁹⁴ Data as of 2012 indicated no association between glyphosate and any solid tumor or lymphoid malignancy, including non-Hodgkin lymphoma. For this disease, the incidence rate ratio was 0.87 (95% CI: 0.64–1.20) comparing the highest exposure quartile to persons with no exposure. The AGRICOH Consortium did a pooled analysis of three large agricultural cohorts from France, Norway, and the United States, to investigate possible associations between non-Hodgkin lymphoma and 14 pesticide classes and 33 active ingredients, including glyphosate.⁹⁵ They registered 2430 cases of non-Hodgkin lymphoma among 316,270 farmers. Glyphosate use did not associate with non-Hodgkin lymphoma overall (meta-HR = 0.95; 95% CI: 0.77–1.18, ever vs. never users), but there was a borderline association for one type of non-Hodgkin lymphoma, namely diffuse large B-cell lymphoma (HR = 1.36; 95% CI: 1.00–1.85). A limitation of the study was the crude exposure measurement (ever vs. never users).

A recent meta-analysis, instead, found a significantly increased risk of non-Hodgkin lymphoma due to high cumulative exposure to glyphosate (meta-RR = 1.41; 95% CI: 1.13–1.75).⁷⁸ The meta-analysis, which considered the Agricultural Health Study and five case–control studies, cannot, of course, overcome the limits of the individual studies. Although the report described the method for identifying cases with high cumulative exposure, the defined “high level” was not consistent among the six studies. Also, the report did not provide a risk estimate comparing subjects using the ever–never exposure categories. In a recent pooled analysis of case–control studies, which examined 1690 cases of non-Hodgkin lymphoma and 5131 controls, the ever use of glyphosate (adjusted for other pesticides) showed a nonstatistically significant association with non-Hodgkin lymphoma (OR = 1.13; 95% CI: 0.84–1.51). A weak statistically significant association was observed for glyphosate use >2 days per year (OR = 1.73, 95% CI: 1.02–2.94).⁷⁹ Thus, overall epidemiological studies show that the association between glyphosate exposure and non-Hodgkin lymphoma is weak or null. When the available evidence is analyzed according to Hill’s nine features (Table 4), it is clear that none is fulfilled.

Table 4.

Lack of fulfillment of Hill’s features by existing scientific data on the association between glyphosate exposure and non-Hodgkin lymphoma risk.

Feature	Comment	Fulfillment
Strength	One meta-analysis reported a meta-RR of 1.3 ⁹³ while a more recent meta-analysis, focused on “high cumulative exposure,” reported a meta-RR of 1.41.⁷⁸	No
Consistency	Different studies found either low risk estimates or no association.	No
Specificity	Non-Hodgkin lymphomas are a heterogeneous group of tumors. Crude estimates of exposure to glyphosate were used in the different studies. Subjects were also exposed to other herbicides.	No
Temporality	Most studies did not report time since exposure for the diagnosis of non-Hodgkin lymphoma. It is presumed that the diagnosis occurred after exposure. A reasonable latency period for this disease is unknown.	Not certain
Biologic gradient	Data on exposure–response relationships are not available. One study found an association with a “high cumulative exposure” and “longest exposure duration.”⁷⁸	No or not certain
Plausibility	Mechanistic studies are missing.	No
Coherence	Epidemiological studies have variable results.	No
Experiment	A toxicology study of rodents administered glyphosate for their life span did not find an increased incidence of lymphomas.⁹⁶	No
Analogy	Not applicable.	No

meta-RR: meta (summary, overall or pooled) relative risk.

Most risk factors for the disease have weak effects, and the only factors with an OR, meta-OR, or meta-RR > 2 are gluten-sensitive enteropathy, celiac disease, hepatitis B virus infection, occupational spray-painting, and allele G at rs7755224 (Table 3). Moreover, toxicology studies of rodents administered high doses of glyphosate for almost their entire lives did not find an increased incidence of lymphomas.⁹⁶ Thus, there are no mechanistic or experimental data to support the epidemiological results. Like for most types of cancer,¹⁹ even for non-Hodgkin lymphomas chance may be the only plausible causal factor in most cases. Therefore, in a given individual with a diagnosis of non-Hodgkin lymphoma, it is almost impossible to ascertain a causal relationship between the disease and glyphosate exposure.

Conclusions

Most types of cancer have genetic causes (inherited mutations/variations or somatic mutations) or complex genetic and environmental causes. In most cases, however, the real cause of a particular cancer in an individual cannot be determined. Epidemiology provides evidence for defining risk factors for any given disease. A few factors have an enormous impact on risk, for example smoking (which increases the risk of lung cancer by about 2000%^97,98) and germline mutations (which increase the risk of hereditary cancer hundreds or thousands of times^53,99,100). But for most cancers, environmental, genetic, and individual (e.g. diabetes, obesity) factors are associated with small increases in risk, typically in the 10% to 30% range. Therefore, on the basis of the “more likely than not” burden of proof, the use of talcum powder cannot be considered a cause of ovarian cancer in patients who have developed this disease. For the same reason, glyphosate exposure cannot be considered the cause of non-Hodgkin lymphoma. These conclusions are based on simple probability calculations, as epidemiological risk estimates for associations between the claimed exposures and these two cancers are low (with meta-RR values of 1.3–1.5).

By this analysis, the attribution by a court of a causal relationship between a claimed exposure and a cancer, including the cases of talcum powder–ovarian cancer and glyphosate–non-Hodgkin lymphoma, is arbitrary and in contrast with the scientific evidence. Consideration should also be given to the possibility that there is no risk attributable to the claimed factors and that what is presented to the courts is actually a misleading presentation of science. A possible solution to helping the courts make decisions based on the evidence has been proposed by O’Connell¹⁰¹ for radiation-induced cancer and is termed “proportional liability.” To explain this term, I must premise that common law follows a rule of “all or nothing,” which requires a plaintiff to receive full compensation from the accused or no compensation. Full compensation was awarded in the two cases discussed here. But attribution of these diseases to exposures alone is scientifically wrong, as the exposures are only weak factors among the multiple known risk factors.

Proportional liability allows a defendant (a company) to be held liable for damages only in proportion to its fault in creating the event. The scientific evidence would be used in court to determine the excess risk attributable to the chemical exposure, also considering the presence and strength of association of other risk factors with the claimed disease. The risk attributed to the chemical exposure would then define the defendant’s share of the total damage sustained by the claimant, which could be calculated from insurance tables similar to those used to compensate victims of road accidents, whose cause of illness or death is certain.

Under proportional liability, the defendant would be liable only for the portion of damage attributed by a court to the chemical exposure. With this approach, the interests of both the plaintiff and industry are defended and secured. However, it would be extremely difficult for the court to ascertain proportional liability for each of the multiple factors that may be involved in the risk of a given complex disease such as cancer. The potential benefits of proportional liability or other alternative approaches for future cancer litigation cases merit further study.

Footnotes

Acknowledgment

I am grateful to Valerie Matarese, PhD, for her professional scientific editing.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

TA Dragani

References

Hill

. The environment and disease: Association or causation? Proc R Soc Med 1965; 58: 295–300.

Vandenbroucke

Broadbent

Pearce

. Causality and causal inference in epidemiology: the need for a pluralistic approach. Int J Epidemiol 2016; 45: 1776–1786.

Rothman

Greenland

. Causation and causal inference in epidemiology. Am J Public Health 2005; 95(Suppl 1): S144–S150.

Vineis

. Causality assessment in epidemiology. Theor Med 1991; 12: 171–181.

Popper

Miller

. A proof of the impossibility of inductive probability. Nature 1983; 302: 687–688.

Morabia

Hume, Mill, Hill, and the sui generis epidemiologic approach to causal inference. Am J Epidemiol 2013; 178: 1526–1532.

Phillips

Goodman

. Causal criteria and counterfactuals; nothing more (or less) than scientific common sense. Emerg Themes Epidemiol 2006; 3: 5.

Dopart

Friesen

. New opportunities in exposure assessment of occupational epidemiology: use of measurements to aid exposure reconstruction in population-based studies. Curr Environ Health Rep 2017; 4: 355–363.

Steckling

Gotti

Bose-O’Reilly

, et al. Biomarkers of exposure in environment-wide association studies—opportunities to decode the exposome using human biomonitoring data. Environ Res 2018; 164: 597–624.

10.

German

Fink

Heron

, et al. The accuracy of cancer mortality statistics based on death certificates in the United States. Cancer Epidemiol 2011; 35: 126–131.

11.

James

Bull

. Information on death certificates: Cause for concern? J Clin Pathol 1996; 49: 213–216.

12.

Norton

Dowd

Maciejewski

. Odds ratios-current best practice and use. JAMA 2018; 320: 84–85.

13.

Viera

. Odds ratios and risk ratios: What’s the difference and why does it matter? South Med J 2008; 101: 730–734.

14.

Green

Freedman

Gordis

Reference guide on epidemiology. In: Reference Manual on Scientific Evidence. 3rd ed. Washington, DC: National Research Council. The National Academies Press, 2011, p. 549.

15.

Riaz

Khan

Riaz

, et al.

Disorganized systematic reviews and meta-analyses: Time to systematize the conduct and publication of these study overviews?

Am J Med 2016; 129: 339.e11–339.e18.

16.

Moher

Liberati

Tetzlaff

, et al. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. PLoS Med 2009; 6: e1000097.

17.

Keyes

Galea

. Commentary: the limits of risk factors revisited: Is it time for a causal architecture approach? Epidemiology 2017; 28: 1–5.

18.

Phillips

. Mutational spectra and mutational signatures: insights into cancer aetiology and mechanisms of DNA damage and repair. DNA Repair (Amst) 2018; 71: 6–11.

19.

Tomasetti

Vogelstein

Cancer etiology. Variation in cancer risk among tissues can be explained by the number of stem cell divisions. Science 2015; 347: 78–81.

20.

Moolgavkar

Knudson

Jr . Mutation and cancer: a model for human carcinogenesis. J Natl Cancer Inst 1981; 66: 1037–1052.

21.

Noci

Dugo

Bertola

, et al. A subset of genetic susceptibility variants for colorectal cancer also has prognostic value. Pharmacogenomics J 2016; 16: 173–179.

22.

Galvan

Colombo

Frullanti

, et al. Germline polymorphisms and survival of lung adenocarcinoma patients: a genome-wide study in two European patient series. Int J Cancer 2015; 136: E262–E271.

23.

Pintarelli

Cotroneo

Noci

, et al. Genetic susceptibility variants for lung cancer: replication study and assessment as expression quantitative trait loci. Sci Rep 2017; 7: 42185.

24.

Sud

Kinnersley

Houlston

. Genome-wide association studies of cancer: current insights and future perspectives. Nat Rev Cancer 2017; 17: 692–704.

25.

Fridman

Zitvogel

Sautes-Fridman

, et al. The immune contexture in cancer prognosis and treatment. Nat Rev Clin Oncol 2017; 14: 717–734.

26.

Hanahan

Weinberg

. Hallmarks of cancer: the next generation. Cell 2011; 144: 646–674.

27.

Seo

Lee

, et al. The transcriptional landscape and mutational profile of lung adenocarcinoma. Genome Res 2012; 22: 2109–2119.

28.

Krishnan

Ebert

Ting

, et al. Whole-genome sequencing of Asian lung cancers: second-hand smoke unlikely to be responsible for higher incidence of lung cancer among Asian never-smokers. Cancer Res 2014; 74: 6071–6081.

29.

Nagahashi

Sato

Yuza

, et al. Common driver mutations and smoking history affect tumor mutation burden in lung adenocarcinoma. J Surg Res 2018; 230: 181–185.

30.

Elliott

Bostrom

Filges

, et al. Elevated pyrimidine dimer formation at distinct genomic bases underlies promoter mutation hotspots in UV-exposed cancers. PLoS Genet 2018; 14: e1007849.

31.

Mao

Brown

Esaki

, et al. ETS transcription factors induce a unique UV damage signature that drives recurrent mutagenesis in melanoma. Nat Commun 2018; 9: 2626.

32.

Robles-Espinoza

Roberts

Chen

, et al. Germline MC1R status influences somatic mutation burden in melanoma. Nat Commun 2016; 7: 12064.

33.

Lawrence

Stojanov

Polak

, et al. Mutational heterogeneity in cancer and the search for new cancer-associated genes. Nature 2013; 499: 214–218.

34.

Jia

Pao

Zhao

. Patterns and processes of somatic mutations in nine major cancers. BMC Med Genomics 2014; 7: 11.

35.

Poulos

Wong

JWH

. Finding cancer driver mutations in the era of big data research. Biophys Rev 2019; 11: 21–29.

36.

Wang

Fang

Cheng

, et al. The molecular landscape of synchronous colorectal cancer reveals genetic heterogeneity. Carcinogenesis 2018; 39: 708–718.

37.

Murakami

Matsumura

Brown

, et al. Exome sequencing landscape analysis in ovarian clear cell carcinoma shed light on key chromosomal regions and mutation gene networks. Am J Pathol 2017; 187: 2246–2258.

38.

Moffitt

Dave

. Clinical applications of the genomic landscape of aggressive non-Hodgkin lymphoma. J Clin Oncol 2017; 35: 955–962.

39.

Rosner

Markowitz

Chowkwanyun

. “Nondetected”: the politics of measurement of asbestos in talc, 1971-1976. Am J Public Health 2019; 109(7): 969–974.

40.

Pierce

Riordan

Miller

, et al. Evaluation of the presence of asbestos in cosmetic talcum products. Inhal Toxicol 2017; 29: 443–456.

41.

Noonan

. Environmental asbestos exposure and risk of mesothelioma. Ann Transl Med 2017; 5: 234.

42.

Sebastien

Bignon

Martin

. Indoor airborne asbestos pollution: from the ceiling and the floor. Science 1982; 216: 1410–1412.

43.

Abelmann

Glynn

Pierce

, et al. Historical ambient airborne asbestos concentrations in the United States—an analysis of published and unpublished literature (1960s-2000s). Inhal Toxicol 2015; 27: 754–766.

44.

Torre

Trabert

DeSantis

, et al. Ovarian cancer statistics, 2018. CA Cancer J Clin 2018; 68: 284–296.

45.

La Vecchia

. Ovarian cancer: epidemiology and risk factors. Eur J Cancer Prev 2017; 26: 55–62.

46.

Camargo

Stayner

Straif

, et al. Occupational exposure to asbestos and ovarian cancer: a meta-analysis. Environ Health Perspect 2011; 119: 1211–1217.

47.

Reid

de Klerk

Musk

. Does exposure to asbestos cause ovarian cancer? A systematic literature review and meta-analysis. Cancer Epidemiol Biomarkers Prev 2011; 20: 1287–1295.

48.

Hemminki

Sundquist

Brandt

. Incidence and mortality in epithelial ovarian cancer by family history of any cancer. Cancer 2011; 117: 3972–3980.

49.

Jervis

Song

Lee

, et al. Ovarian cancer familial relative risks by tumour subtypes and by known ovarian cancer genetic susceptibility variants. J Med Genet 2014; 51: 108–113.

50.

Terry

Karageorgi

Shvetsov

, et al. Genital powder use and risk of ovarian cancer: a pooled analysis of 8,525 cases and 9,859 controls. Cancer Prev Res (Phila) 2013; 6: 811–821.

51.

Berge

Mundt

Luu

, et al. Genital use of talc and risk of ovarian cancer: a meta-analysis. Eur J Cancer Prev 2018; 27: 248–257.

52.

O’Brien

Tworoger

Harris

, et al. Association of powder use in the genital area with risk of ovarian cancer. JAMA 2020; 323: 49–59.

53.

Antoniou

Pharoah

Narod

, et al. Average risks of breast and ovarian cancer associated with BRCA1 or BRCA2 mutations detected in case series unselected for family history: a combined analysis of 22 studies. Am J Hum Genet 2003; 72: 1117–1130.

54.

Liu

Zhang

Zhao

, et al. The association between overweight, obesity and ovarian cancer: a meta-analysis. Jpn J Clin Oncol 2015; 45: 1107–1115.

55.

Havrilesky

Moorman

Lowery

, et al. Oral contraceptive pills as primary prevention for ovarian cancer: a systematic review and meta-analysis. Obstet Gynecol 2013; 122: 139–147.

56.

Straif

Benbrahim-Tallaa

Baan

, et al. A review of human carcinogens—Part C: metals, arsenic, dusts, and fibres. Lancet Oncol 2009; 10: 453–454.

57.

Baan

Straif

Grosse

, et al. Carcinogenicity of carbon black, titanium dioxide, and talc. Lancet Oncol 2006; 7: 295–296.

58.

IARC Working Group on the Evaluation of Carcinogenic Risks, to Humans. Carbon black, titanium dioxide, and talc. IARC monographs on the evaluation of carcinogenic risks to humans 2010; 93: 1–413.

59.

Penninkilampi

Eslick

. Perineal talc use and ovarian cancer: a systematic review and meta-analysis. Epidemiology 2018; 29: 41–49.

60.

Gonzalez

O’Brien

D’Aloisio

, et al. Douching, talc use, and risk of ovarian cancer. Epidemiology 2016; 27: 797–802.

61.

Tossavainen

Karjalainen

Karhunen

. Retention of asbestos fibers in the human body. Environ Health Perspect 1994; 102(Suppl 5): 253–255.

62.

Sayan

Mossman

. The NLRP3 inflammasome in pathogenic particle and fibre-associated lung inflammation and diseases. Part Fibre Toxicol 2016; 13: 51.

63.

Donaldson

Murphy

Duffin

, et al. Asbestos, carbon nanotubes and the pleural mesothelium: a review of the hypothesis regarding the role of long fibre retention in the parietal pleura, inflammation and mesothelioma. Part Fibre Toxicol 2010; 7: 5.

64.

Heller

Gordon

Westhoff

, et al. Asbestos exposure and ovarian fiber burden. Am J Ind Med 1996; 29: 435–439.

65.

Snider

Wood

. Obesity induces ovarian inflammation and reduces oocyte quality. Reproduction 2019; 158(3): R79–R90.

66.

National Toxicology Program. NTP lifetime carcinogenesis studies of amosite asbestos (CAS No. 12172-73-5) in Syrian Golden Hamsters (Feed Studies). Natl Toxicol Program Tech Rep Ser 1983; 249: 1–81.

67.

National Toxicology Program. NTP toxicology and carcinogenesis studies of crocidolite asbestos (CAS No. 12001-28-4) In F344/N Rats (Feed Studies). Natl Toxicol Program Tech Rep Ser 1988; 280: 1–178.

68.

National Toxicology Program. NTP Toxicology and carcinogenesis studies of amosite asbestos (CAS No. 12172-73-5) in F344/N Rats (Feed Studies). Natl Toxicol Program Tech Rep Ser 1990; 279: 1–341.

69.

Swerdlow

Campo

Pileri

, et al. The 2016 revision of the World Health Organization classification of lymphoid neoplasms. Blood 2016; 127: 2375–2390.

70.

Teras

DeSantis

Cerhan

, et al. 2016 US lymphoid malignancy statistics by World Health Organization subtypes. CA Cancer J Clin 2016; 66: 443–459.

71.

Tramacere

Pelucchi

Bonifazi

, et al. Alcohol drinking and non-Hodgkin lymphoma risk: a systematic review and a meta-analysis. Ann Oncol 2012; 23: 2791–2798.

72.

Wang

Vajdic

Linet

, et al. Associations of non-Hodgkin lymphoma (NHL) risk with autoimmune conditions according to putative NHL loci. Am J Epidemiol 2015; 181: 406–421.

73.

Tio

Cox

Eslick

. Meta-analysis: coeliac disease and the risk of all-cause mortality, any malignancy and lymphoid malignancy. Aliment Pharmacol Ther 2012; 35: 540–551.

74.

Wang

Zhang

. Dairy product consumption and risk of non-hodgkin lymphoma: a meta-analysis. Nutrients 2016; 8: 120.

75.

Castillo

Mull

Reagan

, et al. Increased incidence of non-Hodgkin lymphoma, leukemia, and myeloma in patients with diabetes mellitus type 2: a meta-analysis of observational studies. Blood 2012; 119: 4845–4850.

76.

Han

Luan

, et al. Dietary fat consumption and non-Hodgkin’s lymphoma risk: a meta-analysis. Nutr Cancer 2017; 69: 221–228.

77.

Conde

Halperin

Akers

, et al. Genome-wide association study of follicular lymphoma identifies a risk locus at 6p21.32. Nat Genet 2010; 42: 661–664.

78.

Zhang

Rana

Shaffer

, et al. Exposure to glyphosate-based herbicides and risk for non-Hodgkin lymphoma: a meta-analysis and supporting evidence. Mutat Res 2019; 781: 186–206.

79.

Pahwa

Beane Freeman

Spinelli

, et al. Glyphosate use and associations with non-Hodgkin lymphoma major histological sub-types: findings from the North American Pooled Project. Scand J Work Environ Health 2019; 45: 600–609.

80.

Kane

Newton

Roman

. Non-Hodgkin lymphoma and gluten-sensitive enteropathy: estimate of risk using meta-analyses. Cancer Causes Control 2011; 22: 1435–1444.

81.

Gan

Fan

, et al. Hepatitis B virus and risk of non-Hodgkin lymphoma: an updated meta-analysis of 58 studies. J Viral Hepat 2018; 25: 894–903.

82.

Zhu

Jing

Hepatitis C virus infection is a risk factor for non-Hodgkin lymphoma: a moose-compliant meta-analysis. Medicine (Baltimore) 2019; 98: e14755.

83.

‘t Mannetje

De Roos

Boffetta

, et al. Occupation and risk of non-Hodgkin lymphoma and its subtypes: a pooled analysis from the interlymph consortium. Environ Health Perspect 2016; 124: 396–405.

84.

Caini

Masala

Gnagnarella

, et al. Food of animal origin and risk of non-Hodgkin lymphoma and multiple myeloma: a review of the literature and meta-analysis. Crit Rev Oncol Hematol 2016; 100: 16–24.

85.

Shankland

Armitage

Hancock

. Non-Hodgkin lymphoma. Lancet 2012; 380: 848–857.

86.

Cerhan

Slager

. Familial predisposition and genetic risk factors for lymphoma. Blood 2015; 126: 2265–2273.

87.

IARC. Glyphosate. In: Some organophosphate insecticides and herbicides. Lyon, France: IARC Monographs on the Evaluation of Carcinogenic Risks to Humans, 2017, p. 1.

88.

EFSA (European Food Safety Authority). Conclusion on the peer review of the pesticide risk assessment of the active substance glyphosate. EFSA J 2015; 13: 4302.

89.

EPA. United States Environmental Protection Agency. Glyphosate. Available at: https://www.epa.gov/ingredients-used-pesticide-products/glyphosate (accessed 18 November 2019).

90.

Health Canada. Government of Canada. Re-evaluation Decision RVD2017-01, Glyphosate. Pest Management Regulatory Agency. ISSN: 1925-1025 (PDF version). Available at: https://www.canada.ca/en/health-canada/services/consumer-product-safety/reports-publications/pesticides-pest- management/decisions-updates/registration-decision/2017/glyphosate-rvd-2017-01.html (2017, accessed 18 November 2019).

91.

JMPR. Glyphosate. Pesticide Residues in Food 2019 Extra Joint FAO/WHO Meeting on Pesticide Residues 2019: 79–82.

92.

Committee for Risk Assessment RAC. Annex 1. Background document to the Opinion proposing harmonised classification and labelling at EU level of glyphosate (ISO); N-(phosphonomethyl)glycine. EC Number: 213-997-4. CAS Number: 1071-83-6. 2017: 1–261.

93.

Chang

Delzell

. Systematic review and meta-analysis of glyphosate exposure and risk of lymphohematopoietic cancers. J Environ Sci Health B 2016; 51: 402–434.

94.

Andreotti

Koutros

Hofmann

, et al. Glyphosate use and cancer incidence in the agricultural health study. J Natl Cancer Inst 2018; 110: 509–516.

95.

Leon

Schinasi

Lebailly

, et al. Pesticide use and risk of non-Hodgkin lymphoid malignancies in agricultural cohorts from France, Norway and the USA: a pooled analysis from the AGRICOH consortium. Int J Epidemiol 2019; 48(5): 1519–1535.

96.

Greim

Saltmiras

Mostert

, et al. Evaluation of carcinogenic potential of the herbicide glyphosate, drawing on tumor incidence data from fourteen chronic/carcinogenicity rodent studies. Crit Rev Toxicol 2015; 45: 185–208.

97.

Freedman

Leitzmann

Hollenbeck

, et al. Cigarette smoking and subsequent risk of lung cancer in men and women: analysis of a prospective cohort study. Lancet Oncol 2008; 9: 649–656.

98.

Pirie

Peto

Reeves

, et al. The 21st century hazards of smoking and benefits of stopping: a prospective study of one million women in the UK. Lancet 2013; 381: 133–141.

99.

Skalet

Gombos

Gallie

, et al. Screening children at risk for retinoblastoma: consensus report from the American association of ophthalmic oncologists and pathologists. Ophthalmology 2018; 125: 453–458.

100.

Ericson

Nilbert

Bladstr

, et al. Familial risk of tumors associated with hereditary non-polyposis colorectal cancer: a Swedish population-based study. Scand J Gastroenterol 2004; 39: 1259–1265.

101.

O’Connell

. Causation’s nuclear future: applying proportional liability to the Price-Anderson Act. Duke Law J 2014; 64: 333–376.