Practitioners in the field of mental and behavioral health have become very familiar with how anxiety presents itself in the clinical space. Since the COVID-19 pandemic, the surge of anxiety throughout the lifespan has been observed extensively across disciplines. With language associated with anxiety becoming more prevalent, there is a shift in how anxiety has been defined, and thus, has influenced the way mental illness is discussed among those impacted, regardless of which side of the therapy room someone is on. Currently, clinicians grapple with understanding the way anxiety has changed over time, especially objectively, which has a bi-directional influence through using the Diagnostic and Statistical Manual of Mental Disorders (DSM).

Fortunately, there is a way to better understand the changes associated with the semantics of meaning of anxiety over time. The context and associations with anxiety can be further understood through an examination of the linguistic dimensions and features based on a technology-informed methodology. Through this approach, the field can be better informed of both the evolutionary development of the word anxiety in the most prominent text associated with its classification and diagnosis.

Rationale

Since the inception of this research, the data on anxiety and their prevalence in the United States have grown exponentially. With the DSM remaining the manual for diagnosing in the field of psychiatry and behavioral health, it is undeniably both influencing and influenced by culture, so much so that seven versions have been created since the publication of the original DSM in 1952 [1]. Only one diachronic study with a foundation in corpus linguistics has examined the language across versions of the DSM [2]. With anxiety being the most common mental illness in the United States, the need to understand the language surrounding its diagnostic properties is imperative, especially to fully grasp the aforementioned influence of this diagnosis in the field, the United States culture, and world.

Four topics were identified in a review of the research on anxiety and specifically how the concept of anxiety has been defined over time: (a) the prevalence of anxiety in the United States, (b) anxiety as a diagnosis in the DSM, (c) the cultural influence of anxiety, and (d) the intersection of mental health and corpus linguistics.

??According to the Anxiety and Depression Association of America [3], over 19% of the U.S. adult population is affected by an anxiety disorder, making it the most common mental illness in this country. Over 9% (5.8 million) of children carry a diagnosis of anxiety [4], which, when left untreated, is associated with depression, school failure, and substance use disorders [5]. Despite anxiety disorders being treatable, only about 37% of those with these diagnoses receive treatment. With the rise in mental health issues across the United States, a need exists to continue to study anxiety disorders and how the diagnoses of these disorders have evolved and will continue to evolve.

Anxiety first appeared within a diagnostic category in the DSM-I but within the chapter of “psychoneurotic disorders” and under the classification “anxiety reaction” [1]. It was not until the DSM-II [6] was released that anxiety became a characteristic of neuroses, presented under the diagnostic category “anxiety neurosis.” Generalized anxiety disorder (GAD) is the most familiar and cited anxiety disorder in the United States and was recognized as a disorder in the release of the DSM-III [7], when the word “neuroses” was eliminated from the language in this version. Since the DSM-III, the manual and classification of anxiety disorders has continued to evolve, as has the collaboration with the APA in contributing to this evolution. Despite these transformations, the granular study of language as it relates to anxiety disorders remains absent.

The DSM is used by practitioners throughout both the United States and internationally. As it relates to the influence of culture on the DSM, an international review team is specifically designed to examine the implications of diagnosis and culture to incorporate findings in the section on culture-related diagnostic issues [8]. However, it is important to note that with anxiety, specifically, cross-cultural comparisons fail to disaggregate diverse cultural groups [9]. Because linguistics can be studied diachronically, new opportunities to examine the evolution of language associated with the DSM and anxiety can be accomplished.

Corpus linguistics is a methodology that uses computer programs to analyze large sets of data and their underlying structure, meaning, and patterns. Corpus linguistics began intersecting the field of behavioral health starting in the mid-1990’s, with increasing numbers of studies over the past decade. With the issues among the mental health field being so vast, many corpus linguistic studies have analyzed anxiety through the sphere of social media [10,11] rather than professional literature. Still, studies such as these are synchronic in nature, and a diachronic approach to analyze the phenomenon of anxiety would help to understand its evolution over time.

Within corpus linguistics is the ability to study language from a variety of perspectives. One of which is through collocation network analysis. Collocations are, in essence, words that repeatedly co-occur in text [12]. Of the many benefits of studying collocations is the ability to examine the relationships of words through tables, as well as through visual summaries such as networks/graphs. Sophisticated programs such as #Lancsbox assist in converting the numerical values of collocations to more layered analyses. The rationale for using collocations analyses is the greater opportunity to examine both words in isolation as well as the habitual co-occurrence of words together and how those co-occurring words have changed over time.

The purpose of this study is to explore how language relating to the diagnosis of anxiety disorders has changed since the inception of the DSM. Anxiety disorders remain the most common mental illness in the DSM, and yet an understanding of how language is used in diagnosing since the first DSM has yet to be studied. Using corpus linguistics, specifically through the lens of collocations, we hope to contribute to the helping professional field and facilitate a greater understanding of how much anxiety has changed or remained static over the course of the development of the DSM.

Three research questions were developed to guide this study:

RQ1: What is the frequency of the word “anxiety” in comparison to total word count in each version of the DSM?

RQ2: What are the top 15 collocates of anxiety in each version of the DSM?

RQ3: How have the collocates of anxiety changed over time?

Design

This study utilized a diachronic corpus linguistic design [13] specifically focusing on collocations. The variables were the node word ‘anxiety’, and any collocations of the node word within a five-word span to the left and right.

Corpus

Register, scope, and sources: The sources of the corpus in this study included versions of the DSM starting with the first edition and ending with the recently released DSM-5-TR [14]. Two specific manuals in this sequence were excluded: the DSM-I Special Supplemental [15] and the DSM-II 6th Printing Change [16] as the first is not a DSM diagnostic manual, and the second is a retroactive publication that is irrelevant to the development of linguistic features of anxiety words.

For the DSM-I through the DSM-IV-TR, PDF versions of the manual were utilized. For the DSM-5 and DSM-5-TR the electronic versions through psychiatry.org were utilized. The DSM-5-TR was accessed in April 2022 shortly after its release. This is significant because there have been text updates to the DSM-5-TR since then, but with a minimal impact on the anxiety disorder subsections.

Because classification and grouping of anxiety over time has changed, it was best to examine the specific subsections of each manual that seemed to address anxiety specifically. Thus, for the DSM-I, all material under “Psychoneurotic Disorders” was included since descriptions of these disorders identify anxiety as the chief symptom [1]. From the DSM-II, all material under the heading “Neuroses” was included due to these disorders being marked by anxiety [6]. In the DSM-III, there is a shift to identifying anxiety disorder as its own diagnostic category, which has continued through the DSM-V-TR [7,14]. Thus, for the DSM-III, DSM-III-R, DSM-IV, DSM-IV-TR. DSM-5, and DSM-5-TR the material under the diagnostic category “Anxiety Disorders” was included [7,14,17-20].

Corpus preparation: All files were converted to .txt format for compatibility with the analysis software. The .txt files were separately reviewed by each researcher to ensure there were no errors in the conversion. Another linguistic study of the DSM has taken a more aggressive approach to preprocessing due to the nature of their analysis [2]. However, this study focused on collocations and thus did not require the same level of preprocessing for the analysis to occur. The only preprocessing was done on the DSM-5 and DSM-5-TR. Both manuals in their electronic forms include in-text citations and references in each subsection. Because the analysis software included authors’ names in counts regarding collocations and because the DSM-I through the DSM-IV-TR did not include in text citations or references it was appropriate to remove in-text citations and the references section to keep anxiety collocations consistently counted across all manuals.

Measures

Query words: Query words are the specific node words used to identify collocations. Possible candidates for query words in this study being words with a high semantic or conceptual overlap with anxiety. As, the entire point of the study was to identify what the meaning of anxiety is over time; therefore, using semantically similar words would have been to put the cart before the horse. Therefore, the query term for the study was anxiety.

Log Dice: Log Dice computes the harmonic mean of two ratios that convey the tendency of two words to appear together relative to the individual frequencies of the words in the text corpus [21]. As a standardized metric it characterized the assessment and comparison of collocations extracted from corpora of varying sizes [22].

Apparatus: Two programs were utilized in this analysis: Antconc [23] and Lancsbox [24]. Antconc is a language processing program that can perform various analyses on a corpus, but for this study it was utilized in the text conversion from PDF to .txt and then used to check for errors in the conversion.

Lancsbox is a similar corpus analysis program and was used in this study to identify the specific collocates with the Log Dice.

Data analysis

In terms of RQ1, raw count frequency and the percentage of the word anxiety were calculated out of the total words or tokens reported for each of the eight DSM versions using #Lancsbox X.

Regarding RQs 2–3, the minimum Log Dice for inclusion was statistical cutoff value: 6.0: and minimum collocation frequency: 5 [24]. Regarding the maximum number of collocations to show from each query, the parameters or non-shared collocates per query was 30, and the shared collocates per query was also 30. The parameters for L and R span (L5–R5), minimum collocate frequency, and minimum collocation frequency were drawn from Brezina [24]. The higher the Log Dice score, the more the two words appear exclusively [21]. The top 15 words that collocated with anxiety are reported in Table 2. Stop words [25] were not included Table 1.

Regarding RQ1, the frequency of the node word anxiety in proportion to the total word count in the anxiety-related disorder chapter of each DSM can be found in Table 1. It is worth highlighting that the highest percentage of anxiety in proportion to the total word count was found in DSM-I (1.44%) until six versions later with the DSM-5 (2.39%). While the frequency of anxiety leveled off on both versions of the DSM-IV, an over 1% increase was observed in the DSM-5, which was maintained in the text revised version that followed.

As for RQ2, the top 15 collocates of anxiety and their associated Log Dice measure can be found in Tables 2 and 3. The fixed maximum value of Log Dice is 14.

In response to RQ3, many of the collocates have remained consistent, but many other collocates have changed significantly. Additionally, the relative strength or weakness of the collocations have changed as well. The specific collocates can be viewed in Tables 2 and 3.

The results present an initial picture of both static elements of anxiety yet also dynamic elements that are in flux. Regarding RQ1, we note that overall word count of anxiety in contrast to total word count is relatively stable with an initially high count in the DSM-I that drops off in the DSM-II, picks up in the DSM-III, has a slight dip in the DSM-III-TR, and then continues to increase generally with each new manual. One trend is that in each revised manual (DSM-III-R, DSM-IV-TR, and DSM-5-TR) the usage of anxiety in proportion to total word count is less than its predecessor.

Along these same lines, the total word count of the DSM-IV-TR and DSM-5-TR are relatively close, which suggests that either the concept of anxiety is reaching a limit or the appropriate bounds to what can be said about it in the context of a diagnostic manual. An alternative explanation is that the register is constricting what is being written about. Given that Rensi and Dykeman [2] noted continued increase in word count in the substance use sections, it is more likely that the first hypothesis is the case.

Regarding RQ2, the top 15 collocates in each version of the DSM have changed. Although the changes from new versions of the DSM to the text revisions appear to be less striking, each new version of the DSM has different collocates. This has two possible explanations. First, the concept of anxiety is actively in flux. Second, the writing style of the DSM is actively in flux. Between these two explanations the first appears most likely as two previous studies have found the writing style of the DSM to be relatively stable or at least constrained by the specific register of the DSM [2,26].

Regarding RQ3, there are several ways that the collocates have changed. First, the concept of anxiety as a disorder has been firmly established since the DSM-III. This pattern has not changed except in the fact that more and more anxiety is being linked with the concept of disorder; the log score has increased over time to the point where disorder is almost exclusively used in collocation with anxiety (13.5 log dice where the maximum is 14). Thus, the concept of anxiety as a disorder is both enduring and increasing in strength.

In contrast to the relative stability of the concept of anxiety as a disorder, the focus on specific diagnoses appears to be in flux. The term “generalized” has a high Log Dice score and relative place in the DSM-III through the DSM-IV-TR but then drops off in the DSM-5. The term substance-induced is present in the DSM-IV and DSM-IV-TR but then disappears from the DSM-5. The term “medical” is in the DSM-IV, missing from the DSM-IV-TR, and present in increased strength in the DSM-5 and DSM-5-TR. The term “separation” is present in the DSM-III, missing from the DSM-III-TR, back in the DSM-IV, and increased in strength in the DSM-5 and DSM-5-TR. The term “panic” nearly follows this course except that it does not reappear in the DSM-IV or DSM-IV-TR. However, it does show up in the DSM-5 and DSM-5-TR but at a lesser strength than the DSM-III. Lastly, the term “social” appears first in the DSM-IV and increases in strength in the DSM-5 and DSM-5-TR.

There are two possible explanations for this, first, it could be that the diagnoses and symptoms themselves are in flux. Alternatively, instead of the diagnoses or symptoms being in flux, the emphasis within the concept of anxiety is changing. Given that no new diagnoses have been added to the anxiety disorder section since the DSM-IV, this appears to be a subtle shifting of focus from one diagnosis to another. The idea is not that the diagnoses themselves have changed, but the emphasis has. Which leads to the next observation.

The concept of anxiety is increasingly becoming tied to interpersonal interactions. The term separation first appeared in the DSM-III, drops off in the DSM-III-R, returns in the DSM-IV, and ends in the DSM-5-TR at the strongest collocation strength. Along this line the term social first appeared in the DSM-IV and increased in strength until the DSM-5-TR in which it is second in collocations only to disorder. Broadly speaking, the DSM-I has no collocates that are tied to environmental/relational factors, where the DSM-5-TR has two of its top four collocates being social words, and the other two collocates are disorder and disorders. Thus, anxiety is increasingly being used in the context of human relationships or lack thereof.

The first explanation for this is that from a clinical perspective, anxiety is increasingly being viewed as more of a relational disorder rather than an individual disorder. Alternatively, this could be explained by great disconnection in the public; increased use of technology, combined with social media and the COVID-19 pandemic have impacted people’s ability to interact socially, and thus anxiety is increasingly being seen in the context of social interactions. Of these two explanations, the second appears more likely due to two major factors: the first explanation simply deals with an observation of clinical response while the second explanation deals with the causes of the shift; additionally, the second explanation is in line with other research on anxiety [27].

Lastly is the shift in authorial stance, or the relative confidence/certainty or hesitancy of the authors. In corpus linguistics these two concepts are called hedges (hesitancy or lack of certainty) and boosters (certainty or confidence). In the DSM-III-TR we see the word “invariably,” which is a booster, while it immediately drops off in the DSM-IV, while we see a concurrent use of the word “may,” which is a hedge, from the DSM-IV onward. There are two explanations for this. First, the authors of the DSM from the DSM-IV onward are signaling that they are less confident about anxiety in general. In contrast, it may be that this is simply a chance observation and would not stand up to a rigorous research examination. Given that other research has shown boosters and hedges to be relatively stable in the DSM [26] it is more likely that the second option is correct, but additional research would be necessary to completely confirm this theory.

Limitations

There are three limitations to this study that should be considered when interpreting the findings. First, limitations exist in the sample size, as the anxiety section from each DSM represents a relatively small sample size for analysis, especially within earlier additions, which could limit the statistical power and generalizability of the results. Small sample sizes result in increased variability in frequency and collocation statistics, so trend analysis across editions may be harder to detect in a significant and meaningful way.

Second, anxiety disorder compositions, categories, and names have changed across DSM editions. The tracking of diagnostic terms over time is complicated by these changes and affecting the ability to compare collocation frequencies and strengths across DSM editions, which could be less accurate without properly accounting for major taxonomy changes across editions. An example being posttraumatic stress disorder and obsessive-compulsive disorder have been added to the anxiety sections and then moved to their own sections. Other terms such as anxiety neurosis have been subtracted, and others have been renamed.

Third, not all collocates for each node word of anxiety were used across each version of the DSM. Our corpus included only the text for the specific description of anxiety disorder in each edition of the DSM (DSM-I through DSM-5-TR) and not the use of the word anxiety in the entire DSM text edition. This may have an impact on our findings because the word anxiety may have different collocations or accompanied words when it is used in the context of different parts of the text—for example, when used to describe a different disorder.

Implications

Based on the results there are a couple of key implications. The first implication is that the concept of anxiety is increasingly being tied to social interactions. This is significant because treatments for anxiety focus on individual therapy or treat anxiety as an individual disorder. The APA Division 12 website lists two treatments for GAD: cognitive behavioral therapy (CBT) and Mom Power [28]. CBT is an individual treatment that typically does not include interpersonal foci but rather intrapersonal processes, and Mom Power is a combined group and individual treatment for a specific population, mothers. This is an example of how the DSM is moving more towards a social construction, while treatments are still focused on individuals. Researchers and clinicians may want to focus on interventions that take a relational or systemic approach to treating anxiety.

The second implication is that the concept of anxiety is changing over time. This has implications for the validity of research and current therapy models. Evidence-based practices that rely on studies conducted before 1980 (when the DSM-III was published) should be reconsidered. The concept of anxiety before the DSM-III is significantly different from the current understanding of anxiety. Additionally, studies conducted under the DSM-III and DSM-III-TR conception of anxiety may be seen as less valid due to the changes in how anxiety has been conceptualized since then. For example, the meta-analyses of CBT as a treatment for GAD were all conducted in the early 2000’s, almost a decade before the DSM-5, and they mostly review research under the DSM-III, DSM-III-TR, and DSM-IV [29-33]. Researchers should consider replicating previous findings around treating anxiety to determine whether previous models of treatment continue to be as efficacious.

Lastly, clinicians should be cautious in how they discuss anxiety with clients. Increasingly, anxiety is being associated with a disorder, and this may have a negative effect in two main ways. First, it may over-pathologize a common experience; this is in line with Shorter [34] echoing Chadoff’s [35] fears that common human experiences would become disorders in the DSM as the DSM increases in size and scope. Second, it casts anxiety in a wholly negative light. To have no anxiety is not a helpful condition. The student who feels no anxiety about exams and thus does not study, the hiker who feels no anxiety about snakes and thus sticks their hand in every hole, and the pilot who feels no anxiety about flying and thus does not perform the checks they should are all not served well by a lack of anxiety. Having no anxiety is unhelpful; rather, anxiety needs to be proportional, yet not debilitating.

We have no conflicts of interest to disclose.