Current Issue

A systematic search of relevant databases was undertaken to locate journal articles that investigated communication disorders during the last decade. The databases searched for this study were Web of Science, PsycINFO, and CINAHL, in which research articles relevant to communication disorders were most likely to be published.

Searches were restricted to peer-reviewed journal papers in English published between 2000 and 2019. The limit resulted in excluding bibliographies of original articles, review articles, book chapters, opinion papers, and conference proceedings. A combination of terms that refer to communication disorders was used in an attempt to identify the most relevant papers. The incorporated terms were those commonly used to refer to CSD, such as disorder, impairment, and therapy. Example terms are “speech-language pathology,” “speech-language pathologists,” “speech pathology,” “speech therapy,” “language disorder,” and “language impairment.” In addition, truncation was used to include all possible morphological differences (i.e., speech-language pathology*, speech pathology*, speech impair*, speech disorder*, language disorder*, language impair*). These search terms were limited and combined using a Boolean operator ‘OR.’ The title and abstract of the identified articles from the search were analyzed in the current study.

A total of 47,681 journal articles were collected and analyzed from the databases after removing duplications. The initial search yielded a total of 64,312 articles which consisted of 35,008 hits from the Web of Science, 17,715 hits from PsycINFO, and 11,589 hits from CINAHL were yielded. Of those, 16,631 articles were found as duplications and dropped from the analysis.

Text mining was carried out to extract relevant terms and trace relationships among dementia and other communication disorders examined in the published research. Text mining is a computer-supported process of extracting useful information and discovering latent patterns from unstructured textual data by utilizing various information retrieval, information extraction, and natural language processing techniques and algorithms (Hotho et al., 2005). This relatively new set of analytic techniques enables researchers to process scaled literature and facilitate the systematic and efficient reviews with analytic techniques such as identification of relevant studies, rapid description, categorization, summarization, and visualization (Malheiros et al., 2007; Thomas et al., 2011; O’Mara-Eves et al., 2015). A growing body of research using text mining techniques in a variety of research areas from health care and medicine (Abbe et al., 2016; Korhonen et al., 2012) to the economy (Abbas et al., 2014) and business (Moro et al., 2015). KH coder3 (Higuchi, 2018), an open text mining analysis software, used in the current review.

Three phases of analytics were conducted according to three general steps of text analytics: preprocessing, text representation, and knowledge discovery. In the first phase, the preprocessing phase was segmented into three steps: a) stemming which cuts the end of words to make it its original form (e.g., ‘says’ and ‘saying’ to ‘say,’), b) dropping stop words (e.g., ‘of’, ‘kg,’ and ‘%’) and generic terms (e.g., ‘abstract,’ person,’ and ‘general’), and c) extracting and selecting compound words. The compound terms were extracted by the researchers of this study through review on the extracted individual terms to derive more meaningful text representations. The researchers of this study also reviewed associated sentences to reflect the contextual meaning of individual compounding terms. For example, the term ‘quality of life’ was extracted as a single term, rather than extracting two terms ‘life’ and ‘quality.’ The reviewed and selected compounding terms were imported to the KH coder 3 software and extracted as single terms or text representations.

In the following phase, text representations were extracted, resulting in 134,288 terms. The researchers of this study independently reviewed and identified terms about dementia (e.g., ‘Alzheimer,’ ‘Lewy bodies,’ and ‘Huntington’) and other CSD conditions (e.g., ‘hearing impairment,’ ‘autism spectrum disorders,’ and ‘aphasia’). Synonyms, abbreviations, and substrings were merged as a single text representation. For example, for autism spectrum disorders, terms such as ‘ASD’ ‘autistic spectrum disorder,’ and ‘autistic spectrum’ were grouped as a single term in the analysis. To select the CSD related terms, we referenced Communication Disorders Taxonomy (Government of Malta, 2019). Medical Subject Headings (MeSH) 2020 (U.S. National Library of Medicine, 2020) and its tree structures were also referenced to group dementia. The disagreement was solved by discussion. Hierarchical relationships among the identified terms were ignored because text mining analyses are based on text representation such as term, it is regarded as an appearance of the particular concept in a dataset. This study, accordingly, assumed that generic condition name such as dementia was found, we regarded it as a representation of the research topic or subjects in the collected publication. For example, dementia can include Alzheimer’s disease (AD), Huntington’s diseases (HD), Lewy Bodies dementia (LBD) and others, but in case of studies did not specify what type(s) of dementia they examined, we simply extracted and used the term ‘dementia’ in the analyses. The selection resulted in 43 terms regarding CSD conditions, ten terms of dementia. Appendix A provides the list of the terms included in the analyses.

Once term selection was completed, research productivity by conditions and by year of publication, and the relationships among studied conditions were computed to answer research questions about the current investigation. The following techniques and indexes were employed: Document frequency (df) was used to compute the productivity of research by conditions in total. Document frequency refers to the number of unique published papers containing individual terms. In the current study, df indicates how many unique studies were conducted regarding individual conditions. The percentage and number of articles to identify dementia and other CSD conditions concerning the year of publication were also computed using the crosstab command embedded in the software. Finally, co-occurrence network analysis was applied to discover and visually present the associations among dementia and other CSD conditions, which answers the research question (2). Co-occurrence, a network analysis technique, represents the association among the terms, assessing how frequently the paired terms co-occur in the same document (Darkes and Goldman, 1998; Wachs-Lopes and Rodrigues, 2016). Among different network analysis techniques, this study adopted modularity, which represents a network property to detect sub-group or division of that network into the community (Clauset et al., 2004). The edge of the network, or the degree of associations between paired terms in a network, was evaluated using the cosine coefficient. The index calculates co-occurrence patterns based on normalized frequency rather than raw frequency, which prevents the length of the document from affecting the evaluation. The score ranges from -1 meaning maximum dissimilarity to 1 indicating maximum similarity, and 0 meaning decorrelation. Figure 1 shows the analytic pipeline.

Figure 1. 
Analytic pipeline

A total of 53 conditions, which consisted of 43 CSD conditions excluding dementia and ten types of dementia, were identified in the dataset. Table 1 shows the top 20 most frequently found conditions. Among the other CSD conditions, language impairment was found most commonly, appearing in slightly over 5,900 (12.38% out of 47,681, df=5,908) journal articles. It was closely followed by autism. When considering language impairment is an umbrella term describing disorders of language that interfere with communication such as aphasia, it can be concluded that autism found in over 5,400 articles (11.52%, df=5,492), is the most commonly studied CSD condition. Of those identified conditions, dementia ranked fourth. This rank appeared relatively high. However, the quantity of related studies was only approximately half of those about language impairment or autistic spectrum disorder (ASD). Besides, the number of articles regarding dementia in Table 1, indicates the total amount of publications of ten types of detected dementia. As shown in Table 2, the percentages of publications on individual dementia conditions were 4% or lower, which were significantly smaller than other CSD conditions. Refer to Appendix A for detailed information about CSD conditions found in the dataset.

Among different types of dementia, broader dementia without a specified cause (unspecified dementia hereafter) was found most common in the dataset (see Table 2). Despite its comparatively higher number of articles, the proportion of studies regarding general dementia is less than 4% (df=1,700, 3.57% out of 47,681). It was followed by AD (2.85%). The proportions of studies regarding other types of dementia was found to be less than 1%, which is less than one-third of unspecified dementia and AD studies. To illustrate, studies pertaining to frontotemporal dementia (FTD) and primary progressive aphasia (PPA) were only 0.98% (df=465) and 0.92% (df=440) respectively. Particularly, three types of dementia have been significantly understudied: only 33 studies (0.07%) appeared to concern neurofibrillary tangles, 16 studies were about Creutzfeldt-Jakob dementia (0.03%), and four studies regarded Klüver-Bucy syndrome (0.01%).

Examination on the publication productivity by year revealed that research into ASD has increased gradually and has become the most popular study subject since 2010. To illustrate, in 2000 only 4.56% (n=34) out of the 746 CSD publications concerned ASD. This proportion of studies on ASD increased to over 10% in 2010 and reached slightly less than 15% in 2018. Contrastingly, studies on aphasia, specific language impairment (SLI), and stuttering gained attention by researchers in the field of CSD before 2010 and the percentages of regarded studies have gradually reduced. The proportion of studies regarding unspecified language impairment had been consistently over 10% during the two decades. However, in comparison to over 17% of studies published in 2000, 12% of CSD research was about unspecified language impairment in 2019. The proportions of SLI and stuttering studies showed similar patterns, but the number of studies regarding these CSD conditions had been much less than that of unspecified language impairment, having been reduced by slightly over 5% to 1%. In other words, in the early 21st century, a relatively large portion of CSD studies focused on general language impairment issues, but recently the focus of CSD researchers had moved onto autism issues. Figure 2 visually shows the changes in the study proportions of the detected CSD conditions including dementia. See Table 3 for details of the top ten most frequently studied CSD conditions.

Figure 2. 
A bubble map of the identified dementia and other conditions in communication sciences and disorders by year

The productivity of studies on dementia had gradually increased but not significantly. When combined with all types of dementia included in this study, only 61 studies (8.18%) out of 746 articles in 2000 studied dementia. The quantity and proportion of studies on dementia had grown to 149 (10.3% out of 1,447) in 2008, 293 (9.53% out of 3,076) in 2012, and 454 (9.42% out of 4,820) in 2018 (see Table 3). These proportions and the number of studies were still relatively smaller than those of other popularly studied CSD conditions such as language impairment or ASD. When examining individual dementia conditions, the proportions of studies about unspecified dementia and AD have been consistently higher than other types of dementia across the last two decades (see Figure 2). The proportion of unspecified dementia studies have ranged from approximately three to four percent (see Figure 2 and Table 3). Studies regarding AD had ranged from slightly over two to three percent. Despite the relatively consistent proportions of studies regarding these two conditions, the quantities of studies had been increased. To illustrate, the number of AD studies had increased from 19 to 155, and unspecified dementia studies have increased from 22 to 174 during the last two decades (see Table 4). Aside from AD and unspecified dementia, the number of studies regarding PPA and FTD recently started to increase as shown in Figure 3.

Figure 3. 
A graph of study productivity regarding different dementia conditions by year

To discover which CSD conditions shared their areas with dementia or vice versa, co-occurrence analysis was conducted. Figure 4 visually represents the associations among the sub-group structure of dementia and other CSD conditions. Conditions in a sub-group, which were associated with lines and presented in the same color, indicates a significant amount of research regarded the paired conditions in the same study. A total of seven groups of CSD conditions were detected. Overall, the group of dementia and other CSD conditions appeared to have weak relationships in the network. The group with ASD, language impairment, and aphasia were found to consist of the largest groups of researched conditions. A majority of dementia conditions, which highlighted in yellow in Figure 4, were grouped separately from other CSD conditions. This finding implies that major study topics, such as ASD, language impairment, and aphasia, in the field of CSD had not actively involved dementia.

Figure 4. 
Co-occurrence network of conditions involved in communication sciences and disorders based on modularity (node=34, edge=40, D=.071)

Only a few specific CSD conditions were discussed in connection with dementia. PPA was found to be the only dementia condition that was associated with the dementia group, showing relatively stronger associations with AD, unspecified dementia, and frontotemporal dementia. This result refers that unlike other dementia conditions PPA is relatively often studied in relation with CSD conditions. At the same time, PPA is also more frequently addressed with the three dementia studies. Other CSD conditions that appeared to have been closely discussed regarding dementia were amyotrophic lateral sclerosis (ALS) and mild cognitive impairment (MCI). Although studies regarding these two conditions were not significant in amount, these conditions were closely discussed in connection with dementia in the field of CSD.