Research trends in computerized cognitive training contents with text network

2.1 Searching strategy

This study used four databases, viz. PubMed (NCBI), EMBASE, Cochrane Library, and Web of Science, to gather literature and analyzed the published study results up to December 2019. The literature collection used the search strategy by four types: study type, participant type, intervention type, and result measurement type. Three researchers independently reviewed and discussed the collected papers based on whether they should be included in or excluded from this study, and the contents for the study were extracted accordingly.

2.2 Inclusion & exclusion criteria

Studies: Only papers written in English were included, and those written in other languages were excluded. There was no limit in clinical trial design methods, and the studies on the development and design of CCT contents were excluded. Studies comparing the evaluation scores before and after CCT of MCI patients were included. Cases in which the clinical trial results were insufficient (i.e., the evaluation scores before and after computerized cognitive training being similar or decreased) were excluded, and only cases in which the results improved in more than 50% of the participants were included. Moreover, papers of which the full-text could not be collected were excluded.

Participants: The present study included papers that tested patients of age 60 years or older and diagnosed MCI based on Petersen’s diagnosis criteria [22]. All papers focused on people with mental or neurological disorders other than MCI were not considered. Additionally, the papers that did not present data on MCI patients or subjected patients who may have dementia were excluded. Studies of a sufficient number of participants were included to derive the statistically significant results, and only the studies of 10 or more participants were included [23].

Interventions: The present study put no restriction on the means of CCT, e.g. virtual reality, video game, computer, and tablet PC. However, it excluded the papers that used CCT and non-CCT(training using pencil and paper) since the outcomes can be confusing. Furthermore, papers that did not provide a detailed description of the contents of CCT through discussion were computerized, while the papers that did not explicitly state the targeted cognitive domain were excluded.

Outcomes: We included literature on training the memory, attention, and executive function domains based on the five major cognitive domains (orientation, memory, attention, executive function, and language) evaluated in the Mini Mental State Examination (MMSE) [24]. Since orientation was a sub-item of attention and memory, it was included as attention and memory domains, and the language domain was excluded because it can be less effective in judging MCI due to differences in language used in each country [25, 26]. Moreover, since orientation to time, orientation to place, calculation, recall, copying, and 3-step command among the MMSE subcategories are related to activities of daily living (ADL), we included the studies aiming at training in the four cognitive domains: memory, attention, executive function, and ADL [27]. The areas included in the four cognitive domains were grouped in their respective domains (for e.g. short-term memory was included in the memory domain). There was no limit on the number of cognitive domains studied. However, the studies on cognitive domains other than the four were excluded. The studies on multiple cognitive domains that included one or more than the four cognitive domains were considered within the four cognitive domains (i.e. cognitive training to improve memory and judgment was considered as the cognitive training with the goal of memory).

2.3 Keywords & study selection

Table 1 lists the keywords used for literature search categorized by each type according to the inclusion and exclusion criteria.

A total of 888 papers were identified in four databases after the initial literature search, and three additional papers were collected by manually searching for the references in the collected papers related to this study in Google Scholar. Figure 1 shows the literature collection process.

OPEN IN VIEWER

Fig. 1

Flow chart of research selection process.

Through the literature identification process, the duplicate papers were removed, while through the eligibility process the subject and abstract were reviewed and those that did not meet the inclusion criteria were removed. In this way, 82 papers were selected via the eligibility process. Those 82 studies used CCT in the treatment of MCI patients. The inclusion process reviewed the main text and removed the papers that did not meet the criteria of the study type, participant type, intervention type, and outcome type. The final 18 papers selected in the inclusion process signified that they used CCT to treat MCI patients and showed positive clinical outcomes. The current study analyzed the trend of CCT using the 82 papers identified via the eligibility process, and performed a meta-analysis using the 18 papers selected through the inclusion process.

Figure 2 shows the 82 papers on the study of CCT for MCI patients selected in the eligibility process and 18 papers on the study with verified clinical effect selected in the inclusion process. They are listed by the published year.

OPEN IN VIEWER

Fig. 2

Research trends by year of publication.

The selected 82 papers were all published in journals, and the study type consisted of design, clinical, and review. In the present work, the first identified paper on CCT was the study by Mate-Kole, C. C. et al. (2007) published in 2007, and since then, the interest on CCT is ever increasing. In 2019, in particular, there were 16 studies on CCT, and 6 of them verified the effectiveness of CCT through clinical trials. It showed that the digitalization of cognitive training increased with the advancement of the digital content technology, thereby verifying the effectiveness of CCT. Tables 2 and 3 show the top 10 publishers of papers on the selected 82 CCT cases and the top 10 journals of most cited papers, respectively, for the statistical analysis of journals. Impact Factor (2019), ranked no. 1 in Table 3, scored 7.755 points in the Cochrane database of systematic reviews, and Laver et al. (2017), ranked no. 1 in Table 2 was cited 941 times [28]. In particular, one of the study goals of Laver et al. (2017) was to examine the effect of VR depending on the presence of interventions for global motor function, cognitive function, and activity limitation. The review article analyzed 72 cases of 2,470 stroke patients. Its main result showed that there was a statistically significant difference between the participants when using virtual reality with daily and in ADL also (SMD 0.49, 0.21 to 0.77, 10 studies, 210 participants, low-quality evidence).

Table 4 lists the countries of the institutions of the principal author of the 82 selected papers to analyze in which countries the studies of CCT are active. In Table 4, if the number of published papers is two or more, the institution that the principal author is affiliated with is categorized, and only the country without the affiliated institution is listed if the number of papers is one. Suppose the institution of the principal author is a college. In that case, it is included in the university category, and the medical institutions, such as hospital, clinical, and rehabilitation center, is included in the medical center category. As Table 4 shows, the United States had the most institutions affiliated with the principal authors, 17 in number, followed by Italy and China, with 6 in number. Of the 82 selected papers on CCT, the universities were the most active institutions, 60 in number, followed by medical centers, 13 in number, and research institutes, 9 in number.

The text network can understand the characteristics of clustered keywords groups through location and relationship analysis using the visual expression of keywords [29]. Network clustering can build networks with various approaches, such as co-citation, bibliographic combination, co-author, and co-word analysis, according to the analysis target and purpose [30]. The analysis based on the simultaneous occurrence of keywords among these network construction methods has an advantage in understanding the conceptual structure and trend of a specific field by measuring the relationship strength between representative keywords in related papers [31, 32]. Therefore, it was used to connect the keywords with edges and links and express the connection relationship between keywords visually. Density, average distance, centrality analysis, and component analysis are widely used to examine the characteristics of connected keywords. The most used index is centrality, which has the advantage of quantifying the frequency and connection strength according to various types of centrality by analyzing keywords that play an important role in the network connection.

This involves text network analysis using Quanteda package of R (version 3.6.2) statistical software to investigate the transfer effect of CCT [33]. The present study calculated the Degree Centrality (DC), Closeness Centrality (CC), and Betweenness Centrality (BC) to analyze the link strength between the keywords [34]. The DC refers to the number of keywords linked with a keyword. A higher centrality indicates that the keyword is linked with many other keywords. The calculation for the DC can be expressed as Equation (1). C d ( N i ) = ∑ j = 1 k - 1 x ij ( i ≠ j )(1) where C _d  (N _i ) refers to the DC of the i ^th keyword N, k refers to the total number of keywords, x _ij denotes an index representing the total number of keywords where the i ^th keyword has a connection with k-1 keywords except the i^th one. To note, the DC only shows the frequency of connection between the keywords; therefore, it becomes necessary to analyze the CC. The CC calculates the direct connections between the keywords and indirect connections to calculate the distance between a keyword and all other keywords to extract the influential keywords. In other words, the index represents how close a keyword is with another keyword. Therefore, a higher CC signifies a higher possibility of diffusion to the connected keyword. For example, we can infer that a higher CC between the memory and attention domains means that the CCT contents for memory training can be effective for attention training as well. The calculation for the CC can be expressed as Equation (2). C c ( N i ) = 1 ∑ j = 1 k d ( N i , N j )(2) where C _c  (N _i ) refers to the CC of the i^th keyword N, and d (N _i , N _j ) refers to the shortest path between the i ^th and j ^th keywords. The present study visualized the DC and CC indices between the keywords via thickness of the line or link that connected two keywords. Therefore, two keywords which are closer or have a thicker line between them mean that they have a more frequent or influential relationship. Additionally, we calculated the BC to evaluate to what extent a keyword can play a mediating role in the connections between other keywords in the network. In this study’s network of CCT contents, presence of a high BC means that the content can play a role in connecting other disconnected cognitive domains. In other words, it implied that the keyword in the middle position plays the role of a broker and connects other keywords. as the BC can be calculated using Equation (3). C b ( N i ) = ∑ j < k g jk ( N i ) g jk(3) where C _b  (N _i ) refers to the BC of the i ^th keyword N, g _jk refers to the number of the shortest paths between the j ^th and k ^th keywords, and g _jk  (N _i ) refers to the number of paths that include the keyword N _i between the j ^th and k ^th keywords. The BC represents the extent to which a keyword (playing the role of a broker) lies in the shortest path between two other keywords and indicates the number of keywords located in the path in a network. The current work developed and analyzed a network by connecting all the keywords used in the intervention manners and contents in a specific cognitive domain based on the co-occurrence matrix, which expressed frequencies of the keywords in an array using all indices described above. We also categorized the CCT contents which are widely used in a specific cognitive domain based on it and analyzed how effective the contents are in a cognitive domain.

3.1 Characteristics of included studies

Table 5 shows the number of participants and characteristics of the research methods in the 18 papers needed for the analysis of clinical trials among the collected 82 papers. The number of participants was more than 10. The age of participants in the study by Gagnon et al. (2012) and Ballesteros et al. (2017) was over 60, but the studies did not disclose the number of participants by gender. However, they mentioned that there was no statistical difference between the participants, and we included them in the analysis. The number of participants in the selected papers varied widely from 21 to 271 in number, and there was no statistical difference in gender. There was no restriction on criteria, and the studies used a variety of diagnostic criteria, including Mini-Mental State Examination (MMSE), Montreal cognitive assessment (MOCA), and Saint Louis University Mental Status Exam (SLUMS). In all the studies, the participants had no disease or neurological problem other than MCI and did not take any drug. There were also no clinical ethical problems. The participants had at least six years of education in most of the studies, and the studies stated the selection and study processes. The research design on the selected 18 papers included 14 papers on a randomized single-blinded controlled pilot trial, 2 papers on a single-blinded randomized controlled trial, 1 paper on a randomized controlled pilot trial, and 1 paper on single-blinded randomized controlled pilot trial.

Table 6 lists the intervention manner and contents of CCT in the 18 papers that reported the effectiveness of clinical trials. Computer was the most widely used tool for CCT, while video and virtual reality were used as the medium. As Table 6 shows, the comparative experimental studies that used computers and video games together as a training tool accounted for the most, and Nintendo and iPad were used as the media of the video game. All studies stated the process of computerized training to the participants and provided sufficient education and training. The training contents were either available or designed specifically matched with the study objective. Only the studies that reported significant effect of training were included. Memory accounted for most of the cognitive domain targeted by the training, and a variety of contents were performed for it. The most frequent memory training method used by the 18 studies was N-Back, and each study designed their contents based on the different memory areas, viz. occupational memory, short-term memory, and instantaneous memory. Since the lower-level recognition domains were divided according to the cognitive characteristics of the higher-level cognitive domains, the analysis included them in the higher-level cognitive domains of the target cognitive area [35]. The training period varied widely between 1 week and 26 weeks, and most studies compared the participant groups and observed the effect after training of one month or longer. The training on activities of daily living involved activities such as grocery shopping, cooking, and washing, which are routine activities of daily life. All studies performed them in a virtual reality environment. The training was conducted with one or two target cognitive areas to each content, and the training period varied widely between 1 week and 6 months.

Figure 3 visualizes the text network analysis of digitalization types, training contents, and targeted cognitive domains listed in Table 6. The main keywords in the text network are indicated in bold. Here, CCT_contents were connected to Memory_contents, Attention_contents, executive_contents, and ADL_contents. Separately, the Intention Type (IT) was connected to VR, Computer, and Video, and the Target cognitive domain (TCD) was connected to memory, attention, executive, and ADL. The connecting line between the keywords became thicker as the centrality index became higher, and a thicker line represented a stronger correlation between the keywords.

OPEN IN VIEWER

Fig. 3

Text network analysis of computerized cognitive training.

Table 7 lists the calculation results of the centrality indices shown in Fig. 3. In the DC index, memory was a CCT content having the highest frequency 33, followed by attention with frequency 16, activities in with frequency 3. Likewise, memory accounted as the largest training target domain with frequency 14, followed by attention with frequency 7, executive ability with frequency 2, and activities in daily living with frequency 1. The indices of the memory contents and the attention contents were relatively high in CC index, indicating that the studies focused more on memory and attention than executive ability and activities of daily living.

Although the frequency of memory training contents was more than twice that of the attention contents when analyzing only the DC, they showed similar scores of 0.387 and 0.381, respectively, in the CC index. It means that the semantic distance between the memory training contents and attention training contents is short and implied that performing memory training would be effective for the attention domain also. CoTras, which is the training contents targeting memory, is sometimes performed also for attention. However, it is segmented for training of each area and has not been analyzed for transfer effects. In the BC index, the memory training contents were the highest with frequency 64, indicating that it is likely to be used more widely. The BC increases as there are more target cognitive domains of the collected studies and is equal to the number of computerized cognitive training contents if there is only one domain. Moreover, the BC index of the video was higher than the other intervention manners, indicating that many studies tried to verify the effectiveness by comparing video with other intervention manners.