Our findings reveal wide inconsistencies in both environmentally attributable risk estimates and definitions of the environment over the last 50 years. As expected, publications defining the environment in the narrow sense (air, water, food, and soil pollutants) tended to have smaller attributable risk estimates, whereas publications referring to the environment in the broadest sense (including lifestyle factors, occupational exposures, and pollutants) had consistently larger risk estimates. The majority of publications referred to the environment in the broadest sense, although a trend was observed where the proportion of publications using the narrow definition appeared to be increasing.
The findings from the current review emphasize the caution researchers must use when interpreting environmentally attributable risk estimates, particularly in relation to cancer. These results also contribute to the existing body of knowledge on the vast range of interpretations for the term “environment.” Recent reviews have urged researchers to clearly define what concepts of the environment they are referring to when reporting environmentally attributable risk estimates (Belpomme, 2007
; Saracci and Vineis, 2007
). The need for a clear definition of the environment was consistently observed throughout the conduct of the current review, as it was often difficult, and perhaps meaningless, to interpret an environmentally attributable risk estimate without a clear understanding of the researcher’s interpretation of the “environment.” Boffetta et al. (2007)
emphasizes this fundamental concept by stating that “such ambiguous terminology should be avoided and caution should be exercised in providing and interpreting estimates of attributable and avoidable numbers of cases.” Thus, unless careful definitions for environmentally attributable risk estimates are provided to and understood by the reader, the intended purpose of the reported risk estimate is often undermined by public confusion.
An attempt to identify definitional differences in order to explain corresponding differences in risk estimates has been conducted (Saracci and Vineis, 2007
). It was only when authors explicitly stated what factors they included in their definition of the environment could the appropriate comparisons be made. Saracci and Vineis (2007)
compared Prüss-Üstün and Corvalan’s (World Health Organization) (2006)
19% environmentally attributable risk estimate with Boffetta et al.’s (2007)
1–3% estimate. The researchers were able to compare Prüss-Üstün and Corvalan’s definition (including eight classes of environmental agents) to Boffetta et al.’s (air, water, soil, or food pollutants) since each publication explicitly reported what components of the environment were included in their definition. From their analysis, the authors conclude that “population attributable proportions of a disease are summary indexes attractive for their simplicity but subject to severe limitations.”
Inconsistency in the use of language to describe risk estimates was observed frequently in the current review. Authors reported cancer outcomes as cancer cases, incidence, prevalence, or deaths. Each of these concepts convey different connotations, further complicating the task of comparing attributable risk estimates. Furthermore, the use of the terms “influenced,” “due to,” “caused by,” “associated with,” “attributed to,” “linked to” and “determined by” were all used interchangeably to describe risk estimates; however each have different implications. In addition, many authors refer to the environment in the broadest sense, when in reality they are strictly referring to pollutants.
The current review has several limitations. The literature search focused on publications defining the environment in relation to cancer, thus the main databases utilized were medical in nature. It is therefore possible that we may have discovered additional relevant definitions if we broadened our search to include other sources, such as environmental databases. Using Google Scholar and manually scanning all reference lists of our included publications did not indicate that it is likely any records relating to the current topic were omitted. Furthermore, separating the definition of the environment into narrow and broad categories may have put contrived restrictions on the term “environment.” Nonetheless, we felt that it was an adequate representation for the purpose of the current analysis, as it allowed us to systematically and graphically assess the different definitions. Cancer is a diverse and complex set of diseases with a multifactorial etiology. Environmental exposures also act on this complexity and have differential effects on the diversity of cancer subtypes. In the current review we excluded studies that were limited to relationships between specific environmental exposures and specific cancer subtypes as adding too much reporting complexity with minimal additional information. This decision may slightly underestimate the total environmentally attributable risk estimate, but does not affect our conclusions.
Despite these limitations, the current review emphasizes a few key points. Several publications cited Higginson’s 80–90% estimate when referring to their definition of the environment (Higginson, 1967
). Caution must be exercised when interpreting this estimate. To explain, Higginson’s environmentally attributable risk estimate stemmed from his and Oettle’s early research in the 1950’s on an African cohort residing in a rural setting. Thus, his results should be generalized more broadly with caution. Authors, nonetheless, consistently apply his estimate on an international scale. Doll and Peto’s risk estimate is another widely quoted estimate in reference to the environmental etiology of cancer (Doll and Peto, 1981
; Colditz, 2009
). Their estimate, however, was directed strictly towards the United States, yet is again frequently reported on a global level.
A fundamental concern found in the reviewed publications was the misclassification and underestimation of risk when investigating environmental exposures (Vineis, 2004
; Vineis, 2004
; Wild, 2005
; Vineis and Berwick, 2006
). The role of the environment is frequently underestimated, particularly in relation to gene-environment interactions, due to the low sensitivity of environmental exposure estimates (Brunekreef, 2008
). Misclassification is often apparent since environmental exposures are constantly fluctuating, both in the internal and external environment (Rappaport and Smith, 2010
). Moreover, many exposures and their corresponding effects are measured at low levels, but have a ubiquitous presence in the environment, thus making their true etiological contribution difficult to interpret (Wild, 2009
). Consequently, in order to reduce this common misclassification issue in measuring environmental exposures, Chris Wild coined the term “exposome” which measures all external sources of radiation, psycho-social stress, lifestyle, infections, drugs, diet and pollution and the resulting internal chemical environment (Wild, 2005
; Wild, 2009
; Rappaport and Smith, 2010
). The exposome may offer a way to reduce exposure misclassification and more accurately measure what proportion of cancers are attributable to environmental exposures.
In consideration of possible misclassification and both underestimation and overestimation of environmental exposures, environmentally attributable risks should be used with caution. The term “attributable” implies causation between environmental exposures and cancer cases, yet often this causation is difficult to explicitly establish. It is a rare case that a single environmental agent is necessary or sufficient, let alone necessary and sufficient, for explaining cancer. In the strictest, though not necessarily the most useful, sense all disease is attributable to the environment given that the human genetic code must interact in some way with what surrounds it to develop any phenotypic outcome, including cancers. Interactions among different exposures as well as between exposures and genes are in effect. When one adds to these issues the emerging scientific data from epigenetic mechanisms, explicitly attributing risk is further complicated.
Another way of examining what portion of cancer risk is attributable to the environment involves asking what portion is not attributable to the environment. As definitions of what constitutes the environment become more and more inclusive they approach a possibly oversimplified, but still useful concept where the environmentally attributable risk = 1 – the genetically attributable risk. Irrespective of how one accounts for epigenetic changes in this concept, it is interesting to note that despite dramatic increases in genetic knowledge that began with a draft sequenced human genome in 2000 through today, the estimates of the genetic contribution to cancer incidence have not much changed. Vogelstein and Kinzler (1998)
provided in 1998 an estimate that high-penetrance genes accounted for less than 5% of cancers. Few changes have occurred since then (Vineis, 2004
), in fact it has been reported that only three new high-penetrance cancer susceptibility genes had been discovered between 2001 and 2007 with few expected in the future (Eisinger and Horsman, 2007
). Additionally, from 2007 through 2011 genome-wide association studies have suggested no high-penetrance susceptibility alleles and taken in total genome-wide association study findings continue to explain only a small amount of the heritability of cancer. This all suggests there is a small fixed portion of risk attributable to genetics leaving the large remainder of risk attributable to epigenetic and environmental factors. During this same time period human activity has continued to substantially change the types of environmental exposures we experience and how they are distributed (for example Climate change, ubiquity of endocrine disrupters, shifting tobacco-smoking demographics, changing food environment, the Deepwater Horizon oil spill etc).
Taken together these observations suggest that the numerous definitions of the environment we observe are simply dividing up the same large and relatively fixed portion of the environmentally attributable “risk pie” in different ways, often appropriately reflecting a changing environment. Even Doll acknowledged that the “pie” includes many overlapping and interacting slices and the true proportions may total hundreds of percents (Doll, 1998
). This also suggests that time and energy devoted to establishing a consensus on the environmentally attributable cancer risk may not be well-spent. Rather, the best way forward may prove to be a clear disclosure of the conceptual model and perspective used by the observer that is strongly tethered to the environmentally attributable risk estimate they have provided.