Our data were retrieved from the Web of Science Core Collection (WOSCC), which is the specified article data source for CiteSpace as it contains citation information. An initial topic search for ‘visual hallucination’ resulted in 3178 records published between 1985 and 2020. We filtered out conference abstracts and proceedings and corrigendum documents, which were less representative[31]. We believe that original research papers can better represent the state of the research field compared with other types of documents. Review papers can attach additional importance to the representative papers selected by domain experts[32]. A total of 2641 publications which consist of 2176 articles and 465 reviews, were selected as a database to be used in subsequent analysis (537 were excluded). Then we downloaded raw data, which included full records and cited references, from WOSCC in the form of plain text files.

CiteSpace V based on Java was utilized for information visualization analysis, which provides insights into VH research and makes it easy to effectively follow the progress of information[30,32,33]. In this study, we selected the top 50 most cited or occurring articles or items to create a visualized network with a 1-year interval. In the document co-citation analysis stage, we performed clustering analysis on co-cited references, in which similar references were combined to determine related research fields. Moreover, log likelihood tests (LLRs) typically provide the unique and best results that consider all of the contents of a cluster; thus, we extracted noun phrases from the keywords of articles that cited a cluster on the basis of LLR to characterize the nature of the cluster[34]. Office Excel 2019 was also applied to our study.

To determine the general trend of VH research, we summarized the publications of original articles and reviews over the years. The earliest record we found in WOSCC was published in 1999. The results in Figure 1 show that the publication outputs are mainly in a fluctuating growth trend, with an increase from 61 in 1999 to 182 in 2020.

Open in New Tab   Full Size Figure   Download Figure

Figure 1 Publication output performance during 1999-2020.

Dual-map overlays in CiteSpace can help reveal the trends of the scientific portfolio under a background of a global map of scientific literature. The background has two base maps, the left part shows a base map of citing journals and the right part shows the cited journals, each containing a network of over 10000 journals. Similar journals form a cluster, which is labeled on the basis of the terms in the journal titles of the cluster. The reference relationships between the left and right parts are connected by colored curves that indicate how a current research obtains inspiration from previous works[33]. The vertical and horizontal axes of the ellipse in the left part respectively indicate the number of articles and authors published in journals. The number of citations determines the size of the ellipse in the right part[34]. Figure 2 shows a dual-map overlay visualization of the citing and cited papers with regard to the topic search on VH. Four threads of citations stand out. They originate from four clusters in the citing base map: the orange threads from the cluster of molecular, biology and immunology, the pink threads from the cluster of neurology sports and ophthalmology, the blue threads from the cluster of psychology, education and health, and the green threads from the cluster of medicine, medical and clinical. These threads generally point to two clusters in the cited base map. One is the cluster of molecular, biology and genetics; the other is the cluster of psychology, education and social. New developments are highlighted in red from the publication point of view. The new progression of VH is in the field of mathematics, systems and mathematical, which is worth further research.

Open in New Tab   Full Size Figure   Download Figure

Figure 2 Dual-map overlay for final document storage. The nature of each region is labeled by the journal of the corresponding region. The colored curve represents the reference path from the citing base map on the left to the cited base map on the right. The number of articles and authors published in journals determines the size of the ellipse in the graph.

The total number of publications by country was analyzed to understand their geographic distribution. We checked the names of all countries and merged some regions into corresponding countries. The analysis is presented in Figure 3. The larger the published quantity, the deeper the color. The United States, England, and Japan were the top three countries.

Open in New Tab   Full Size Figure   Download Figure

Figure 3 Geographic distribution of the countries. The figures represent the corresponding country’s total published quantity on visual hallucination between 1999 and 2020. The larger the published quantity, the deeper the color.

Co-authorship was analyzed to detect active authors and their cooperation in the field. For accuracy and objectivity, we reviewed all of the authors’ names to reduce misidentification. The result is displayed in Figure 4. The font size of each author’s name corresponds to the number of articles by each author, which represents the contribution of the author to this field. The color of the tree ring stands for the year in which the author published his or her articles. The thickness of the tree ring represents the number of his or her articles in a particular year. Collaborative intensity between authors is indicated by the thickness of a connecting line. Chronological order information is included in the color of the lines that appear together between nodes: blue represents the oldest, green the middle, and red the newest. In summary, top-productivity authors have greatly contributed to this field, and a relatively stable cooperation has formed among many authors. More than 485 authors have made contributions to the research on VH. Among these authors, Taylor JP (40 articles) ranked first, followed by Aarsland D (38 articles), O'Brien JT (25 articles) and McKeith IG (24 articles).

Open in New Tab   Full Size Figure   Download Figure

Figure 4 Map of active authors in the visual hallucination field. The font size of each author’s name represents the article counts of each author. Different colors inside the circle represent different time intervals. The collaborative intensity between authors can be seen from the thickness of the connecting line.

The topics involved in VH can be described by the keywords extracted from each article in the dataset[31]; however, new insights into this field should be determined. The burst patterns of keywords reveal research hotspots in the field of VH because a burst of a keyword is a sharp increase of the keyword that is likely to have a great influence[32]. Among the 340 selected keywords, 104 have the strongest strength of burst during 1999-2020. Figure 5 lists the top 104 keywords with strongest burst. The blue line represents the timeline from 1999 to 2020, and the red line stands for the years when a keyword has burst. Among the top 104 keywords, particular attention was paid to those keywords that remain to have a burst until 2020, such as ‘neuropsychiatric symptom’ and ‘functional connectivity’. Other burst keywords included ‘auditory verbal hallucination, diagnostic criteria, functional magnetic resonance imaging (fMRI), meta-analysis, management, impulse control disorder, sleep behavior disorder, and default mode network’.

Open in New Tab   Full Size Figure   Download Figure

Figure 5 The top 104 burst keywords. The blue line represents the timeline from 1999 to 2020, while the red line stands for the years when a keyword has burst.

The most outstanding function in CiteSpace is co-citation analysis, which was used to analyze all references of the 2641 documents downloaded from WOSCC, and the top 50 most-cited references each year was select. As a result, 776 nodes of references were generated and automatically linked in the visual interface. Only if two references were cited by the same document could they be connected. If two references were often cited by documents together, they tended to have a close relationship, so they were classified to the same cluster. In our analysis, the smallest clusters were filed out, and 69 clusters were left. The information on all references formed the intellectual base of this field, and 776 highly cited references were classic documents[32]. By mining the most classic references in each cluster, we can understand the general development process and research frontiers of the VH research field. An overview of a co-cited reference network is shown in Figure 6. The overall structure can be divided into three major parts: the upper left part of the nodes and links, which represent the co-citation of the first 7 years from 1999 to 2005, is essentially in blue. The central part of the network is mainly in green and yellow, which indicates that the relationship was probably constructed between 2006 and 2012. The bottom right part is predominantly in red, and connections are formed credibly in the most recent 8 years.

Open in New Tab   Full Size Figure   Download Figure

Figure 6 A network of 776 co-cited references representing citation patterns of top 50 articles per year between 1999 and 2020. The overall structure can be divided into three major parts: the upper left part of the network is essentially in blue, which represent the co-citation between 1999 and 2005. The central part of the network is mainly in green and yellow, which indicates the relationship is probably constructed in the middle 7 yr from 2006 to 2012. The bottom right part is predominantly in red and connections were formed credibly in the most recent 8 yr, that is 2013-2020.

The quantity of the clusters can be measured via two indexes: Modularity Q value and Silhouettre value. Modularity Q value is a network modularization index with a value range of 0-1. Q > 0.3 indicates that the structure of a certain cluster is significant. Silhouettre value indicates the homogeneity or consistency of the cluster. The closer the S value is to 1, the better the homogeneity of the cluster is. When S > 0.5, the clustering result can be considered reasonable; when S > 0.7, the clustering result is efficient. The number of references constituting the cluster must be greater than 10[35]. Table 1 lists the major clusters of co-cited references selected from 69 clusters. In general, the 10 clusters in the table represent 10 research directions in the field.

The earliest formed cluster is Cluster #1 senile dementia, whose average publication year is 1997. It has more than 100 references as its members. A common theme of this cluster is identifying DLB from Alzheimer's disease, which are both belong to senile dementia[36-38]. The 82 members of Cluster #2 are evenly published in 2001, mainly involve PD with hallucination and focus on phenomenology[39]. Cluster #5 Charles bonnet syndrome, #3 incident dementia, and #6 schizophrenia introduce CBS, DLB and schizophrenia, respectively. These diseases all have VH as a hallmark. The research foundation in the three fields was mostly established in 2001, 2005, and 2014. Their high silhouette values indicate a high homogeneity of the clusters. Importantly, Cluster #2 and #3 are very close to each other in Figure 6; this may be partly because both PD and DLB are Lewy bodies (LB) diseases[40], and persistent VH is related to the spread of LBs[29].

The largest cluster is Cluster #0 labeled body disease, which includes 154 complete references. These references’ average publication year is 2011. Cluster #4 Parkinson’s disease dementia (PDD), which has 70 group members with an average publication year of 2014. These two clusters constitute the main body of VH research in the last decade and are closely connected in Figure 6 because they include research that introduces PD with VH. Cluster #0 focuses on the evidence of changes in brain structure and function in PD with VH[41,42], while Cluster #4 is mainly about the comparison and management of VH in PDD and DLB[43,44].

Other clusters, such as Cluster #7 impulse control disorder and Cluster #9 REM sleep behavior disorder (RBD) are formed more recently. Impulse control disorder and RBD are both included in the non-motor symptoms of PD. The co-cited references in these two clusters demonstrate the relationship between these two symptoms and VH respectively. Cluster #10 5-HT2A involves the investigation of serotonin 2A receptor in PD with psychosis (PDP)[45,46]. It includes publications averagely in 2010.

To detect emerging trends in the field, we further investigated the top high-quality co-cited references. In this part, we used the first author’s name plus the publication year as a notion to refer to the articles extracted from 776 co-cited articles. Top three references are shown in Table 2 as ranked by citation counts, centrality, burst and sigma respectively.

Given the groundbreaking contributions, the most cited articles in the research field are often considered the landmarks[31]. Mckeith IG (2005) and Harding AJ (2002) are at the top of the list. Both of them are in Cluster #3. Mckeith IG (1996) is the third most cited article and belongs to Cluster #1. Notably, Mckeith IG is a pioneer in the field of DLB and has published a series of DLB clinical guidelines in Neurology. Mckeith IG (1996) proposed the first consensus guideline for the clinicopathological diagnosis of DLB[36]. In 2005, Mckeith et al[69] published the third revised clinical diagnostic and treatment criteria for DLB, which included management in the criteria for the first time. These two guidelines authoritatively summarized the clinicopathological diagnosis of DLB and showed the direction of DLB treatment at that time. The article by Harding et al[70] entitled Visual hallucinations in Lewy body disease relate to Lewy bodies in the temporal lobe advanced the finding that temporal lobe LB is strikingly associated with VH given the distribution of LB in the brain. This result is a remarkable finding that links the clinical and pathological features of DLB together[70].

Pivot articles often refer to gateway articles between two densely connected subfields with a unique position. This type of articles can provide insights into emerging trends. The top ranked article by centrality is Aarsland D (2002) in Cluster #1, which has a centrality of 0.09. This double-blind, randomized, placebo-controlled study included 14 PDD patients to study the safety and effectiveness of the cholinesterase inhibitor donepezil in the treatment of PDD. This study lasted for 20 wk and finally they found that donepezil is safe and effective, also it does not worsen motor symptoms of PD[71]. This outstanding discovery has important therapeutic implications for PDD. The second one is Goetz CG (2011) in Cluster #0, with centrality of 0.08. Goetz et al[72] followed up 60 patients with PD but without hallucinations at baseline for more than 10 years. VH was found to dominate in early hallucination profile. This discovery revealed the outstanding position of VH in PDP. The third is Cummings J (2014) in Cluster #0, with centrality of 0.07. Cummings et al[46] conducted a randomized controlled double-blind trial on patients with PDP and concluded that this population can achieve benefit from Pimavanserin, a selective serotonin 5-HT2A inverse agonist, which is the only U.S. Food and Drug Administration-approved medication for PDP[23]. This study was a phase 3 trial that marks a critical breakthrough in the treatment of PDP, in which VH is a common symptom[73].

The importance of burst cannot be overemphasized. Through burst testing of all cited articles, we easily found that Mckeith IG (2017), Mckeith IG (1996), and Mckeith IG (2005) are on the top of the diagram. They are the milestones in relation to DLB. We discussed the last two articles in front part. The top one was Mckeith IG (2017) in Cluster #4. Mckeith et al[74] renewed the consensus report of DLB in 2017. Compared with the 2005 edition, the 2017 edition clearly distinguished clinical features and biological markers. According to different clinical features and biological markers, the diagnosis was divided into probable and possible DLB. Consistency of the diagnostic criteria for DLB will be more conducive to further research on DLB.

It is valuable to examine the citation burst in more recent articles while eliminating the overshadow burst of those landmark articles. The top 10 references that retained citation burst until 2020 are shown in Figure 7, which anticipates emerging trends in the future. In general, a new emerging trend involves PD with VH. Waters F (2014) is a relatively early article comparing VH in different diseases[59]. This article opened up a new dimension in VH research. Although many hypotheses have been proposed to explain the mechanism of VH, and many imaging and electrophysiological studies have partly proved these hypotheses; it is still unclear whether VH in different diseases has the same mechanism, and further research is needed. Pagonabarraga J (2016) is the only article that had citation burst as soon as its publication. PD minor hallucination is not a newly found phenomenon in PD but is underestimated because previous research on PDP focused on the study of well-structured VH. Pagonabarraga et al[29] reported that minor hallucination is the most frequent symptom in PDP and may even occur before the onset of parkinsonism. The third paper was Onofrj M (2013). This paper reviews the hypothetical mechanisms of VH in PD and DLB. To date, three predominant mechanistic models have been presented: a disturbance between top-down and bottom-up aspects of visual perception[1,75]; chronic deafferentation causing hyperexcitability to cortical structures involved in vision[3,76]; and the misattribution of internal imagery[59,77,78]. The authors tend to identify with the attention deficit network model as mentioned before.

Open in New Tab   Full Size Figure   Download Figure

Figure 7 The top 10 references that retained citation burst until 2020.

Sigma is defined as: (centrality + 1)^burstness, which can simultaneously measure a cited reference’s structural centrality and citation burstness. Teunisse RJ (1996) is at the top of the list, which objectively describes the characteristics of CBS by using a semi-structured interview[79]. The advanced nature of this article lies not only in the formation of the prototype semi-structured interview of VH but also in the occurrence of CBS, which is partly due to sensory deprivation and low arousal. The second was Cummings J (2014) in Cluster #4, Mckeith IG (2005) ranked third in this part and their importance is self-evident.

Visual hallucination (VH) refers to a spontaneous visual perception without corresponding external stimuli and often occurs in ophthalmological and neuropsychiatric disorders. It is associated with poor life quality, increased patient hospitalization, and nursing home admission.

To date, there is a lack of scientometric analysis of the research on VH.

To objectively summarize the features of VH research and gain insights into the emerging trends for research on VH.

CiteSpace V was used in this article. Publication outputs, document types, geographic distributions, co-authorship status, research hotspots, and co-citation status were analyzed. A total of 2176 original articles and 465 reviews were included in the database downloaded from the Web of Science Core Collection.

The results showed that most publications can be classified into neurology, sports and ophthalmology studies. In addition, North America, Europe, Asia, and Australia published the most documents. Some well-known authors have always had a leading role in this field; meanwhile, new authors keep emerging. A relatively stable cooperation has been formed among many authors. Furthermore, neuropsychiatric symptom and functional connectivity are the top hotspots. Research on VH in dementia with Lewy bodies and Parkinson’s disease (PD) have received much attention.

Studies on VH in PD are likely to be the new emerging trends in the future, especially the mechanism of VH.

More large-scale clinical and in-depth basic research studies are required to better understand the mechanisms underlie VH, which will contribute to our understanding of pathophysiology and therapy in VH.