In assessing the controversy around the adequacy of EBM, Martini (2020) asserts that disagreement partly results from “ambiguity about the concept of evidence”, which often goes largely “unanalysed” and therefore “contains the contradictions that allow both camps to defend their positions and charge their adversaries”. Similar concerns about the lack of critical reflection on the meanings of evidence and evidence-based have been raised in relation to other areas of evidence-based practice, where analogous controversies have arisen. For instance, Kvernbekk (2011, p. 515) notes that evidence-based has become “a buzzword in contemporary educational debates (and also in medicine and policymaking, among other areas)”, and resorts to philosophy to unpack the meanings of the two components of the term, namely evidence and based. Rather than pursue this discussion in the realm of philosophical debate, our proposed methodology involves analysing attested uses of the two concepts that feed into different interpretations of evidence-based medicine—evidence and based—in a large electronic corpus of texts written by non-specialists using evidence, -based, and evidence-based in arguments about everyday topics such as climate change and human rights, as well as specialists on various areas of Evidence-based Practice, including EBM. Understanding the variety of meanings that basic concepts such as evidence assume in general discourse and in different contexts may afford us some insight into how they came to be condensed and streamlined to support a specific conception of medical practice associated with the EBM model and dominant among some, though by no means all members of the medical community.
Evidence in the GoK Internet corpus
In the GoK Internet corpus, there are 4402 occurrences of the keyword evidence, placing this basic concept in a very high (118th) position in the general frequency list. Evidence occurs with higher frequency in the corpus than lexical items that are very common throughout the English language, such as same and since, and is also more frequent than many other terms that encode basic concepts characteristic of our thematically designed corpus, such as policy and society. At N − 1, the position one word to the left of the keyword, the most frequent collocates are the (719), of (359) and no (229). The and of are the two most frequent words in the overall corpus, so their high frequency in relation to evidence is not immediately informative. In contrast, no generally occurs much less frequently, and thus the pair no evidence makes for an interesting collocation, as can be seen in Fig. 2.
Many instances of the collocation no evidence form part of the larger pattern no evidence of (64), a phrase that in turn occurs in several longer sequences, as in the following examples:
There is no evidence of a link between autisms and vaccines. (Hoofnagle and Hoofnagle 2008; Denialism Blog)
There is no evidence of a link between glyphosate in the food chain and autism, diabetes and obesity. (Zaruk 2016; The Risk Monger)
Both examples are taken from individual blogs concerned with a broad range of scientific matters, yet remarkably, the same exact phrasing occurs twice 8 years apart, and both times in an argument against purported connections between chemicals and health disorders. The first example speaks of a supposed link between autism and vaccination, and while no relevant causality has been established in the medical literature, belief in this relation seems to have only grown over the past decade and developed among large parts of the global population into a general aversion to vaccination. Indeed, while there are currently several vaccines available for COVID-19, this does not necessarily mean that the problem of ‘vaccine hesitancy’ is easily overcome (Harrison and Wu 2020). A central problem in this regard is that it may be feasible in the short term to provide evidence for the efficacy of a vaccine, but it is much harder to establish its safety. Indeed, variants of the common expression ‘absence of evidence is not evidence of absence’ occur several times in the corpus. The expression illustrates a logic on which the sceptic can always fall back when defending a suspicious position in relation to the long-term health effects of vaccines, or to any other hypothetical consequence of multi-faceted bodily or social interventions.
The second example intervenes in the glyphosate controversy, the debate about whether the use of this herbicide is harmful to humans. Glyphosate is domestically known for being the main active ingredient in Roundup, but is also used on a vast scale in agriculture. Studies on its effects are so far inconclusive. Chemical interactions produce complex effects, and while studies declaring the herbicide safe are often considered questionable for being produced by commercial stakeholders, several studies condemning the use of glyphosate are criticized for being “unsupported by evidence” (Mesnage and Antoniou 2017, p. 4). In this respect, providing evidence would mean incontestably showing that glyphosate in the environment causes disease. However, proponents of a stricter regulation of pesticide and herbicide use regularly invoke the “precautionary principle”: when in a state of “scientific uncertainty”, do not take risks (Kudsk and Mathiassen 2020, p. 216). The argument cannot easily be resolved, as both sides come to rely on the rhetorical fallacy termed argumentum ad ignorantiam (Walton 1999). Those critical of glyphosate argue that, since there is no proof of safety, the chemical is unsafe, whereas their opponents, in the absence of proven harmful effects, declare the herbicide safe. Thus, the burden of proof is passed back and forth, and a consensus cannot be reached. Consequently, further oppositional entrenchment can be expected, and notably, both articles from which the above examples are derived ultimately resort to name-calling: Hoofnagle and Hoofnagle (2008) frame the discourse of their opponents as “denialist claptrap”, and Zaruk (2016) seeks to confront what he calls “chemophobic propaganda”. Thus, the dichotomy established by the common pattern no evidence is indicative of a potential deadlock in scientific argumentation. If the idea of evidence requires absolute verifiability, its scope shrinks considerably. The fact that the collocates with evidence is also informative in this regard. The definite article, in contrast to the zero marker, implies certainty and singularity. Evidence suggests is not as forceful and monolithic as the evidence suggests, a phrase the high frequency of which signals, like no evidence, that the concept of evidence is often invoked when dealing in absolutes. One may also note, in this respect, a purported equivalence between science and evidence in parallel expressions of absolute certainty, as in the corpus examples the science is clear (7 instances) and the evidence is clear (9).
Other common collocates of evidence sketch a more nuanced picture, not restricted to the yes–no or present–absent binary: there can be much (17 instances) or enough (17) evidence, but also insufficient (17) or little (38) of it. The fact that the quantifiers expressing sufficiency are often negated, as in not much evidence or not enough evidence, while the main negative ones tend not to be reversed, suggests that the quantification of evidence in most cases implies that it falls short of confirming or supporting a given claim (Fig. 3).
Not only can evidence be quantified, but it can also be qualified in a variety of ways: fossil evidence (22) identifies a particular object of study, whereas empirical evidence (38) identifies a certain method of observation. New evidence (51) is uncovered in relation to earlier information on the same topic, and supporting evidence (11) is expected to back up claims to truth. An interesting category of qualifying adjectives concerns the specification of an affective response, as in compelling (29) and convincing evidence (13). Convincing is often modified by negation, as in no convincing evidence, whereas compelling and other intensifying qualifiers such as ample (14) and damning (7) are usually employed affirmatively, as in:
There is compelling evidence that children’s perceptions of cigarettes are influenced by branding and that branding detracts from the impact of health warnings on packs. (Abbott 2013; Left Foot Forward)
Note that the indefinite ‘there is’ does not specify where the evidence originates. While such specification may occur elsewhere in the article, a more impactful omission is that of the supposedly compelled subject—in the absence of a subject, the qualification of the evidence as compelling is presented as an objective characteristic, thus fabricating a tacit agreement between the author, their sources, and the implied reader. The implication is that the evidence ‘speaks for itself’. The representation of actors and participants is a central issue in critical discourse analysis (e.g. van Leeuwen 2008, pp. 23–54). In scientific as well as political reports, the responsibility of human agents is often elided, while “agency is shifted to abstract processes and entities” (Fairclough 2003, p. 138). In this regard, it is also important to take stock of the verbs of persuasion commonly found at the N + 1 position of evidence: the evidence shows (40), indicates (15) and supports (11). Together, these two patterns (the absence of a subject and the proliferation of verbs of persuasion in position N + 1) suggest that perspectives and positions are seemingly automatically validated by the presence of evidence, with no need for a human mediator to be specified as the agent responsible for interpreting the information at hand. The most common collocate on the right-hand side of evidence, namely suggests (59), forms part of this widespread grammatical pattern. Often, the broader pattern includes strong markers of emphasis that reinstate the yes–no dichotomy at the far ends of quantification: there are several variants of all the evidence suggests as well as no evidence suggests. In these cases, too, the majority of instances do not specify potentially relevant participants in the situation of evidence assessment.Footnote 5
Rhetorically charged quantifications and qualifications such as abundant and compelling frequently precede evidence, but by far the most common qualifier modifying evidence is the seemingly more neutral adjective scientific (197). What exactly makes evidence scientific is seldom specified, but in some cases a number of expected characteristics are explicated:
This is the first scientific evidence, published in Environmental Research Letters, that confirms the numerous anecdotal accounts. (Albert et al. 2016; Desmog)
Here, scientific evidence is partly defined by the communicative setting: an academic outlet is required to present it, and it is contrasted to anecdotal accounts. The latter contrast implies, once again, demands of quantity and quality, as anecdotal evidence (14 instances) is generally understood to be gathered unsystematically and in limited amounts. Furthermore, anecdotes are told rather than observed, meaning that the scientific mode is presented in opposition to the narrative mode of exposition. While such schemas can in some cases be derived from the concordance lines, often the collocation scientific evidence is simply used as a means of emphasis, as further illustrated by the fact that the collocation is frequently preceded by additional adjectives of persuasion. Indeed, the most frequent pattern associated with scientific evidence is overwhelming scientific evidence (10 occurrences). Overwhelming encodes a rhetorical mixture of quantity and suggested affective response, and thus a merger of the features of adjectives such as much and compelling, discussed earlier. In addition, the phrase overwhelming scientific evidence is in the majority of cases used to set up a contrast, as in:
Despite the overwhelming scientific evidence linking fine soot particles to premature death, Honeycutt testified before Congress that “some studies even suggest PM makes you live longer.” (Negin 2017; UCSUSA)
Collocational patterns do not merely report a state of affairs; they express a communicative purpose which is of pragmatic rather than semantic importance (Stubbs 2009, pp. 124–125). In other words, recurrent phrases are likely to have an attitudinal or evaluative function that extends beyond their constative value. This function is termed semantic prosody in corpus linguistics, and in the case of the contrastive pattern despite the overwhelming scientific evidence, which occurs three times in the corpus, the function is not primarily to qualify the evidence in question but to suggest that one should be surprised at the existence of an opinion opposed to the author’s. Arguably, the various patterns of linguistic modification characteristic of evidence do not merely, or even primarily, serve to affirm, negate, qualify or quantify, but mainly to convince the implied reader of a given stance by rhetorical measures that conflate persuasive and epistemic modes of expression.
Importantly for our overall argument in this paper, when the GoK corpus is narrowed down to include only texts that make explicit mention of evidence-based medicine, the overall collocational patterns associated with evidence remain largely unchanged. Thus, the use of rhetorical adjectives such as compelling and overwhelming is not restricted to discussions outside the disciplinary scope of EBM, and this broad rhetorical inflection indicates that common discursive patterns of persuasion are shared across varied settings of practice. The boundaries of different spheres of knowledge production are porous, and basic concepts that display highly frequent patterns of usage alongside more specialist applications are particularly likely to generate heated disagreements through mutual misunderstanding. In the following section, we show how the meaning of -based is as elusive as that of evidence, further contributing to the fact that, while the concept of evidence-based medicine may purportedly apply to a restricted set of principles operational within a circumscribed context, this does not mean that it is necessarily presented and understood as such by different interlocutors.
Based in the GoK Internet corpus
One of the central arguments put forward by corpus linguists is that meaning does not reside in individual words, but is realised in extended lexical elements. This means, as already shown in the previous section, that evidence has to be examined in its context of use. It also means that seemingly secondary elements, such as based, are bound to contribute heavily to the reception and interpretation of the concept and practice of evidence-based medicine. A search for based in the GoK Internet corpus returns 2891 concordance lines. Based is a multi-faceted lexical element that can fulfil a number of distinct linguistic functions. In order to focus on usages that structurally correspond to the phrase evidence-based medicine, we limit our analysis to its occurrence as part of a hyphenated compound adjective (as in evidence-based), rather than as a verb (as in based on evidence). However, the two variants are often closely related, and where relevant attested verbal uses will contribute to the argument.
For compound adjectives containing based, the Oxford English Dictionary (OED 2020) lists four main uses. A first common use indicates a relation to a specific location, as in London-based (36 instances) or UK-based (33). As these examples illustrate, such locations tend to be nations or cities. A second general use indicates a relation to a particular material, often described in chemical terms, and exemplified in the corpus by compounds such as carbon-based (10) and chlorine-based (1). The OED (2020) lists as a third option the use of based to indicate “a foundation, fundamental principle, or underlying basis”. This is by far the most frequent use in the corpus, not only because of the high frequency of evidence-based (218) and science-based (95), but also because of the widespread use of based in relation to social categorizations, as in class-based (36), community-based (30), or faith-based (25). The final main use listed by the OED concerns a specification of the base in question, as in broad-based (20). The fact that instances of all main categories of use can be easily retrieved from the concordance output suggests that the corpus provides an adequate general overview of the variety of meanings expressed by based.
The meaning of based cannot always be intuitively grasped by restricting one’s view to the various compounds in which it occurs, given that these compounds themselves form an adjectival unit that typically modifies a noun that follows. The co-text to the right of the compound has to be taken into account to arrive at an interpretation. A first observation is that when the adjectival compound itself contains a concrete material substance, the full phrase may still establish an abstract relation, as in oil-based economy (1) or carbon-based capitalism (1). The meaning of based is not equivalent in carbon-based capitalism and carbon-based vegetation (1), with the former expressing a relation of dependency and the latter one of composition. Thus, the interpretation of each adjective remains highly context dependent. An observation related to this tendency towards the abstract commonly expressed by based is the adjective’s frequent involvement in the description of large, intangible systems, which may be more or less regulated. The words system and systems immediately follow based fifteen times, and more specific examples of systems attested in this position include taxation (7), politics (8) and organi[zs]ation(s) (17). Social systems are composed of regular patterns of practice, and there is a marked tendency in the corpus for adjectival compounds containing based to be followed by nominal compounds containing making, such as rulemaking (follows based 7 times), policymaking (19) and decision-making (16, counts include hyphenated and unhyphenated spelling variants). This is a strong collocational pattern, as the gerundial form making seldom occurs in the corpus without the preceding words policy or decision. In fact, the expression evidence-based decision-making has become so cliché that, as the following example from our corpus indicates, it can be toyed with for humorous purposes:
Tom Slater has written that in order to give ‘scientific’ credibility to extending the neo-liberal state and deepening social inequality, think-tanks such as the CSJ have “mastered the craft of decision-based evidence making, tailored to the needs of policy elites and politicians on the lookout for accessible catchphrases to woo a jaded electorate.” (Silver et al. 2014; Discover Society)
Typically, nouns referring to the outcome of regulatory procedures such as policymaking can also follow based; there are, for instance, no fewer than 55 occurrences of policy or policies immediately following based, the bulk of which are part of the larger pattern evidence-based policy and science-based policy. Alternatives to policy that occur within similar patterns include approach(es) (21) and practice(s) (8). Thus, based very frequently, and often quite explicitly, functions to express the relation between a productive field of practice and the principles that guide it. This interpretation also holds for evidence-based medicine, yet what linguistically separates this particular phrase from its more general variants is the frequent use of an acronym, namely EBM (38 occurrences across several articles). There is, for instance, no acronym EBA, which could hypothetically refer to evidence-based approaches. The acronym EBP occurs only 17 times in the corpus, and its use is restricted to a single article (in which the P stands for policy, while outside the corpus the same abbreviation will often refer to practice). The use of acronyms is extremely common in scientific writing across various disciplines, even though in the health sciences it has been criticised as a source of “irritation, misunderstanding, and even alienation” (Moris 2020, p. 1274). Even writers who are seemingly well-acquainted with the vocabulary involved may produce formulations that suggest a lack of transparency, as the following example from the corpus illustrates:
One example is the implementation of a new procedure to treat asthma at the Royal Children’s Hospital in Victoria, Australia. The EBM-based approach was able to achieve an unusually high success rate of 95.5 percent during the first three months of transition. (Mawby and Harris 2016; OpenDemocracy)
In the above example, the abbreviation EBM itself contains the word based. Consequently, the example contains both a pleonastic and a tautological element, with the full formula EBM taking up the position normally reserved for evidence. The entire framework comes to serve as a substitute for the raw material: in other words, the wording may suggest that EBM occupies a singular, unambiguous conceptual space. This recalls our discussion of phrasings such as the evidence suggests in the previous section. The evidence suggests implies access to a fully transparent, univocal dataset, while the EBM-based approach implies access to a fully coherent, set framework, but in both cases the assumption of a rigid outline may be evaluated either positively or negatively. The author of the example above favours EBM and sees it as a model practice that should guide developments in other sectors. At the other end of the spectrum, some critics of the model characterise EBM’s rigid univocity as distinctly malignant, as the following example illustrates:
“[EBM is not] medicine based on evidence, but the equivalent in the field of medicine of a cult with its unique dogma, high priest … and fervent disciples,” says Dr. John Service, editor-in-chief of Endocrine Practice. (Hoofnagle 2007; Denialism)
The critique presented in this quote characterises EBM as an oppressive and disciplinary apparatus. Similar critiques were raised in other publications, including scholarly journals, around the same time the Denialism blog post was written. Holmes et al. (2006), for instance, argue that a focus on EBM “eliminates some ways of knowing”, a process which they ultimately describe as “microfascism”. This article too is quoted several times in the corpus. Importantly, in the quote above the speaker feels the need to unpack what he sees as the linguistic dishonesty of EBM by questioning the link to evidence as well as based: “[EBM is not] medicine based on evidence”. In effect he is suggesting that the verbal and the adjectival form of based should essentially express the same relation. Yet, in our corpus, the phrase based on evidence is largely reserved for unique events, decisions or explanations, and only seldom employed with the broad disciplinary quality of the related adjectival compounds, such as evidence-based. Disagreement on the proper application of terms such as based is thus at the heart of the argument. The quotes from both scholarly publications and more informal statements by specialists confirm that evidence-based medicine has met with strong resistance for over a decade now, while at the same time it is still presented as a model practice for other sectors to emulate. What some see as helpful instructions to guarantee best practice, others interpret as undue restrictions, and importantly, based elicits both meanings across the corpus, not just in discussions about the health sciences. Consider the following examples:
His book left scope for alternative views of meritocracy including positive associations of merit-based systems with social justice. (Gallinat 2018; Discover Society)
The corruption of visa work programs might soon expand under Trump’s immigration-reform agenda: His administration seeks to expand “merit based” migration of “high skill” and professional workers, while cutting humanitarian and family-reunification programs. (Chen 2017; The Nation)
In the first example, based connects a practice to its justification, while in the second it sceptically connects a practice to its limitations. The phrases are formally equivalent while their evaluative purposes—partially signalled through the use of scare quotes—are functionally opposed, as they are in different interpretations of evidence-based medicine. Thus, researchers seeking to untangle controversies in the health sciences may benefit from examining discursive patterns that cut across a broad range of disciplines and fields of practice. The continued rejection of the term evidence-based in the context of healthcare has of course not gone unnoticed, and in recent years alternative terms have come to circulate widely. The World Health Organization (n.d.), for instance, explicitly states that evidence-based implies a restricted viewpoint, and that the institution now prefers the term evidence-informed. It is not only based that in some contexts has fallen into disrepute. In Norwegian, the term evidence-based is often translated using kunnskap, usually an equivalent of knowledge rather than evidence, in part precisely to dissociate the discourse from the “positivist connotations that are tied to the English expression” (Bondevik and Engebretsen 2018). Yet, in all these examples, it is not so much the terms evidence or based that independently fall out of favour, but rather their specific combination, as the following example poignantly illustrates in the choice of different terms in parallel positions:
Ending AIDS requires evidence-informed, rights-based global leadership (Fried 2016; OpenDemocracy)
Yet are shifts in wording, such as the one from evidence-based to evidence-informed, indicative of a correspondent shift in practice? Positions on the subject will vary, but our corpus suggests that while medicine may attract a variety of modifiers such as alternative (27 occurrences before medicine), conventional (8), modern (8) and real (8), none offer rival paradigms to evidence-based. They may refer to a competing outlook, but not to a competing set of principles and correspondent practices. The rigid view associated with EBM thus partly consolidates its continued dominance, at least as the standard against which to argue.