Global Long-Term Care Research: A Scientometric Review

Since the early 1960s, long-term care (LTC) has attracted a broad range of attention from public health practitioners and researchers worldwide and produced a large volume of literature. We conducted a comprehensive scientometric review based on 14,019 LTC articles retrieved from the Web of Science Core Collection database from 1963 to 2018, to explore the status and trends of global LTC research. Using CiteSpace software, we conducted collaboration analysis, document co-citation analysis, and keyword co-occurrence analysis. The results showed a rapid increase in annual LTC publications, while the annual citation counts exhibited an inverted U-shaped relationship with years. The most productive LTC research institutions and authors are located primarily in North American and European countries. A simultaneous analysis of both references and keywords revealed that common LTC hot topics include dementia care, quality of care, prevalence and risk factors, mortality, and randomized controlled trial. In addition, LTC research trends have shifted from the demand side to the supply side, and from basic studies to practical applications. The new research frontiers are frailty in elderly people and dementia care. This study provides an in-depth understanding of the current state, popular themes, trends, and future directions of LTC research worldwide.


Introduction
The aging population is growing at an unprecedented rate worldwide. According to the 2017 Revision of the World Population Prospects, the global proportion of people aged 65 years and over has risen from 5.08% in 1950 to 8.29% in 2015 and is expected to reach 15.82% by 2050 [1]. Elderly people are generally at substantial risk of functional limitations and physical disability, creating a great demand for long-term care (LTC) services. Within this context, developing an effective, equitable, and sustainable LTC system to meet the escalating demand becomes increasingly important. In the last half century, LTC has been a major public health issue in nearly all industrialized countries, and a variety of LTC systems have been established based on a mix of financing sources [2][3][4]. Although these systems have grown rapidly, reflecting their popularity among the elderly, they have also faced several challenges, including skyrocketing costs and workforce shortages [5,6]. Moreover, some developing countries have started to pay attention to LTC in recent years because their demands for it are rising dramatically at a rate that exceeds that experienced by industrialized countries [7][8][9].
Given that LTC is a matter of serious concern in many countries, researchers have shown considerable interest in this subject [10,11]. Extensive LTC studies have been conducted from the perspectives of geriatrics and gerontology; nursing; public, environmental, and occupational Despite the popularity of CiteSpace, to the best of our knowledge, it has not previously been used to map global LTC research. To provide a systematic review on LTC research and achieve the expected objectives, three types of scientometric techniques provided by CiteSpace were applied in this study: collaboration analysis, document co-citation analysis, and keyword co-occurrence analysis. Collaboration analysis takes authors' names, countries of affiliation, and institutional affiliation as the units of analysis and evaluates their publication contributions and academic influences by visualizing scientific collaboration networks [26,29,32]. Document co-citation analysis provides insights into the intellectual structures of a knowledge domain and identifies the quantity and authority of references cited by publications [23][24][25]28]. In the process of this analysis, cluster views and timeline views are performed to reveal the conceptual structures and the evolution of scientific activity. Keyword co-occurrence analysis tracks the research hotspots, frontiers, and trends over time by establishing a network of co-occurring keywords that provide information about the core content of articles [27,31]. Specifically, the research frontiers and evolution trends are identified by burst detection [23]. During execution, the parameters (e.g., time slice, node type, and pruning) in CiteSpace should be properly selected in accordance with the research objectives [29].

Data Collection
CiteSpace data collection required two steps. First a globally representative database was selected. Our review was based on the bibliographic records obtained from the Science Citation Index Expanded (SCI-E) and Social Sciences Citation Index (SSCI) in the Thomson Reuters Web of Science Core Collection (WoSCC) database, which is considered one of the most authoritative data sources for bibliometric investigations because it contains the leading global scholarly journals and is continuously dynamically updated [29]. The second step was to design an accurate retrieval strategy. The dataset was retrieved in this study by an LTC topic search, which is an ideal retrieval mode because it can characterize the article content from the perspectives of title, abstract, and keyword. We conducted a topic search for "long-term care" articles published between 1963 (the first article related to LTC was published in this year) and 2018 (downloaded June 27, 2018). The search was limited to journal articles and resulted in the 14,019-article dataset used in the scientometric analysis. In order to examine the effectiveness of the results, we assessed the relevance of the top 100 most cited articles to LTC. The results showed that 83 articles (83%) were closely related to LTC, indicating that our retrieval strategies and search term are appropriate.

Current Status of LTC Research
The 14,019 LTC research articles were published in 18 different languages, although English (13,479) was the predominant language, accounting for 96.15% of the total. Among non-English languages, German (277) and French (103) were the most widely used, contributing 1.98% and 0.74%, respectively. The current status of LTC research is depicted by the year-wise distribution of publications and citations shown in Figure 1. The green points represent the number of publications per year and exhibit a noticeable upward trend. The bar graphs illustrate the annual citation counts, showing a trend that initially increased then decreased. Additionally, two trendlines were identified by fitting a polynomial to the data, as revealed by the dotted lines. Three research stages were summarized according to publications, as follows: 1.
Low-speed development stage (1963)(1964)(1965)(1966)(1967)(1968)(1969)(1970)(1971)(1972)(1973)(1974)(1975). The annual number of publications increased slowly in this period and never exceeded 10. There was only one article in the years 1963, 1965, 1966, 1968, and 1972. Notably, 1975 was the most productive year with nine articles. In terms of annual citation counts, the dominant position was occupied by the year 1975 with 122 citations; there were fewer than 50 citations in all other years, probably because of a limited number of publications in those years.
High-speed development stage . Publication output increased rapidly from 145 in 1991 to 1149 in 2017, resulting in a nearly eightfold increase in the past 27 years. In this period, the polynomial trendline of publication growth showed a significant correlation between publication year and publication counts.  1. Low-speed development stage (1963)(1964)(1965)(1966)(1967)(1968)(1969)(1970)(1971)(1972)(1973)(1974)(1975). The annual number of publications increased slowly in this period and never exceeded 10. There was only one article in the years 1963, 1965, 1966, 1968, and 1972. Notably, 1975 was the most productive year with nine articles. In terms of annual citation counts, the dominant position was occupied by the year 1975 with 122 citations; there were fewer than 50 citations in all other years, probably because of a limited number of publications in those years. 2. Rapid development stage (1976)(1977)(1978)(1979)(1980)(1981)(1982)(1983)(1984)(1985)(1986)(1987)(1988)(1989)(1990). With a negligible amount of fluctuation, the number of publications per year increased steadily over this period. The most productive year was 1990 with 51 articles. Moreover, an increased number of citations per year were also observed over time. Articles published in 1988 and 1990 received many citations (each more than 1,000), and the greatest average number of citations per article (28.31) also belonged to 1990. 3. High-speed development stage . Publication output increased rapidly from 145 in 1991 to 1,149 in 2017, resulting in a nearly eightfold increase in the past 27 years. In this period, the polynomial trendline of publication growth showed a significant correlation between publication year and publication counts. Through curve fitting, the number of publications was estimated to reach 1,217 in 2018 (1,222 articles were retrieved from this year) and 1,302 in 2019, indicating that LTC research may remain active for many years. Moreover, three publication bursts were found in 1991, 2012, and 2015, increasing by 94,

Major Contributors to LTC Research
LTC has become a popular research subject, with varying degrees of concern across countries, institutions, and authors. Who are the predominant contributors? How much difference do they make? What research network do they have? These questions are fundamental to identifying the major contributors to LTC research, but they have not been fully explored in the research. We attempted to answer these questions by mapping three collaboration networks.

Country Collaboration Analysis
An analysis of collaboration among countries was conducted by focusing on the affiliation location of at least one author of each published article [29,31]. The following parameters in CiteSpace were used: (1) time slice from 1963 to 2018; (2) years per slice = 1; (3) term source = title /abstract/author keywords/keywords plus; (4) node type = country; (5) pruning = none; (6) select top 50 most cited articles per slice. After running CiteSpace, results revealed that 68 countries/ regions contributed to LTC publications. Table 1 shows the top 15 contributors, ranked by publication counts and centrality, respectively. The LTC contributions primarily originated from the US, Canada, and England. Specifically, the US ranked first by contributing 6308 (45%), followed by Canada with 1714 articles (12.23%) and England with 815 articles (5.81%). In terms of the large number of publications, it seems that these three contributors form a leading LTC research group. In Figure 2a, each node is a country or region and each link represents the relationship between two nodes. The figures indicate link-strength between the nodes, mainly ranging from 0.15 to 0.35. Among members of the leading group, the link strength of England-USA, England-Canada, and USA-Canada was 0.07, 0.07, and 0.05, respectively, showing limited collaboration. Additionally, the links between other contributors included in the network and the leading group (England-USA-Canada) were not strong, with the strength between 0.5 and 0.12. It is thus necessary to take this into account by adjusting research directions to further improve the level of collaboration among countries.
The collaborative relationships among the three group members were primarily established in the 1990s. Moreover, we found that developing countries/regions were still under-represented in the global LTC research network, although they were making great efforts to develop LTC. As shown in Table 1, China was the only developing country (Taiwan Province of China is treated as a developed region) among the top 15 most productive contributors.
Generally, nodes with high centrality (≥0.1) in the network are indicated by purple rings and connect more links. As shown in Table 1, the top six contributors in terms of centrality were England (0.16), Denmark (0.15), France (0.12), Scotland (0.12), Sweden (0.11), and Switzerland (0.1). Collaboration networks of four countries are shown in Figure 2b-e. England and Denmark, with high centrality had a dense network structure, showing that they occupied key positions on the critical paths in the network and played important roles in connecting with others. However, the limited number of publications significantly weakened Denmark's influences. Canada, with a centrality of 0.03, had a sparse network structure, demonstrating that it did not actively participate in collaborative research activities. Its power, however, cannot be ignored because Canada was the second most productive country. Taking publications and centrality into account simultaneously, the US stands out as the most important contributor to LTC research.
Among members of the leading group, the link strength of England-USA, England-Canada, and USA-Canada was 0.07, 0.07, and 0.05, respectively, showing limited collaboration. Additionally, the links between other contributors included in the network and the leading group (England-USA-Canada) were not strong, with the strength between 0.5 and 0.12. It is thus necessary to take this into account by adjusting research directions to further improve the level of collaboration among countries.   The collaborative relationships among the three group members were primarily established in the 1990s. Moreover, we found that developing countries/regions were still under-represented in the global LTC research network, although they were making great efforts to develop LTC. As shown in Table 1, China was the only developing country (Taiwan Province of China is treated as a developed region) among the top 15 most productive contributors.
Generally, nodes with high centrality (≥ 0.1) in the network are indicated by purple rings and connect more links. As shown in Table 1, the top six contributors in terms of centrality were England (0.16), Denmark (0.15), France (0.12), Scotland (0.12), Sweden (0.11), and Switzerland (0.1). Collaboration networks of four countries are shown in Figures 2b-e. England and Denmark, with high centrality had a dense network structure, showing that they occupied key positions on the critical paths in the network and played important roles in connecting with others. However, the limited number of publications significantly weakened Denmark's influences. Canada, with a centrality of 0.03, had a sparse network structure, demonstrating that it did not actively participate in collaborative research activities. Its power, however, cannot be ignored because Canada was the second most productive country. Taking publications and centrality into account simultaneously, the US stands out as the most important contributor to LTC research.

Institution Collaboration Analysis
In this section, the parameters in CiteSpace were kept the same, except for node type being changed from "Country" to "Institution". To visualize the influential institutions and the collaborations between them, a network comprising 438 institutions and 1,740 collaboration links is depicted in Figure 3a, showing that the top 15 most productive institutions were split into three major groups: clustering from nine American universities; five Canadian universities; and one Netherlandish university (see Table 2). The result further confirms previous findings and attests to

Institution Collaboration Analysis
In this section, the parameters in CiteSpace were kept the same, except for node type being changed from "Country" to "Institution". To visualize the influential institutions and the collaborations between them, a network comprising 438 institutions and 1740 collaboration links is depicted in Figure 3a, showing that the top 15 most productive institutions were split into three major groups: clustering from nine American universities; five Canadian universities; and one Netherlandish university (see Table 2). The result further confirms previous findings and attests to the importance of European and North American LTC contributions [2]. In terms of publications, 15 institutions issued 3047 articles, accounting for 21.74% of the total. Specifically, UT ranked first with 406 articles (2.9%), followed by UNC (233, 1.66%) and Harvard (231, 1.65%). One prominent institution in Europe was VU with 155 articles (1.11%), ranking 12th. Moreover, we noted that Asian institutions accelerated efforts to participate in LTC research, including National Yang-Ming University (90, 0.64%) and the University of Tokyo (75, 0.54%). Apart from the universities, several research centers also played important roles in LTC research. For example, the Centers for Disease Control and Prevention (CDC) in Atlanta GA, USA ranked 19th with a total of 118 articles.  The collaboration network of 15 highly productive institutions is mapped in Figure 3b. The link strength ranged from 0.07 to 0.58 with a mean of 0.18. Research collaboration was most active between McMaster and UA (0.58), followed by McMaster-UBC (0.55) and UNC-Duke (0.51). LTC relationships were established mainly at the beginning of the 21st century, while the UNC-Duke relationship was created in the early 1990s. As for centrality, Harvard (0.14) and UM (0.12) represented the major turning points, acting as bridges linking others in different phases. Moreover, nodes with red inner rings are institutions with strong bursts, i.e., strength. In Figure 3, bursts (strength>20) that reflected significant increases in publications over short periods of time occurred in Maastricht University (28.38, 2015-2018), King's College London (23.13, 2014-2018), VU (21.58, 2013-2018), and the National Center for Geriatrics and Gerontology (20.95, 2015-2018), indicating that the articles from these four institutions attracted special attention over the past five years (2014-2018) and strongly contributed to LTC research development.

Author Collaboration Analysis
A scientific co-authorship network provided information on the core authors and potential collaborators and helped researchers to establish collaborative relationships. The parameters in CiteSpace were remained the same except node type being changed from "Institution" to "Author". The distribution of authors is depicted in Figure 4a. This network consisted of 1,759 nodes and 3,186 The collaboration network of 15 highly productive institutions is mapped in Figure 3b. The link strength ranged from 0.07 to 0.58 with a mean of 0.18. Research collaboration was most active between McMaster and UA (0.58), followed by McMaster-UBC (0.55) and UNC-Duke (0.51). LTC relationships were established mainly at the beginning of the 21st century, while the UNC-Duke relationship was created in the early 1990s. As for centrality, Harvard (0.14) and UM (0.12) represented the major turning points, acting as bridges linking others in different phases. Moreover, nodes with red inner rings are institutions with strong bursts, i.e., strength. In

Author Collaboration Analysis
A scientific co-authorship network provided information on the core authors and potential collaborators and helped researchers to establish collaborative relationships. The parameters in CiteSpace were remained the same except node type being changed from "Institution" to "Author". The distribution of authors is depicted in Figure 4a. This network consisted of 1759 nodes and 3186 collaboration links. Notably, a small group of highly productive authors contributed to a significant share of publications. For example, 2.84% of authors (top 50 authors) were responsible for 10.6% (1486) of the publications. Among all the authors, Mor was the most productive with 98 articles, followed by Zimmerman with 89 articles (see Table 3). Additionally, all authors had low centrality (<0.1), revealing that collaboration among them was insufficient. In Figure 4, bursts (strength ≥ 10), as indicated by red inner rings, occurred in Vellas (13.74, 2014-2018), Lapane (12.69, 1998-2002), Williams (11.8, 2005-2008), Van der Steen (10.83, 2011-2015), and Suzuki (10,(2015)(2016)(2017)(2018). Given that bursts are often accompanied by subsequent increasing trends, research output from Vellas and Suzuki, who both had strong bursts in the last three years (2016-2018), was expected to continue to increase. (1,486) of the publications. Among all the authors, Mor was the most productive with 98 articles, followed by Zimmerman with 89 articles (see Table 3). Additionally, all authors had low centrality (< 0.1), revealing that collaboration among them was insufficient. In   Figures 4b and 4c show the collaboration networks of Mor and Zimmerman, respectively. The authors tended to collaborate with a single, highly productive author, forming co-author clusters. For example, Mor was a central author of one research community, which included Grabowski, Bernabei, Lapane, etc. As compared with Figure 4c, Figure 4b depicts a much denser network structure; Mor connected more links than Zimmerman, indicating that Mor seemed more active and maybe played a more important role in LTC research. Interestingly, no direct collaborative relationship was established between the two major authors. As shown in Figure 4d, Grabowski (37, Harvard University) and Ribbe (16, VU University Medical Center) were the only two authors who made a great contribution to bridging the collaborative gap between Mor and Zimmerman.
To further identify core authors' research contributions, we listed the top six most cited articles published by Mor and Zimmerman, respectively (see Table 3). Mor, a well-known professor in the Brown University School of Public Health, has a long history of conducting research projects on the quality of nursing home care, the determinants of hospitalization, and racial discrimination in health care treatment. During the past 30 years, Professor Mor frequently adopted an integrated research approach that combined quantitative and qualitative data analyses in program evaluations, such as nursing home resident assessment. Moreover, he is one of the authors of the congressionally mandated Minimum Data Set (MDS) for nursing home resident assessment which was designed to assess elders' functional status and care needs. Overall, the above contributions are fully embodied in the top six most cited articles from Mor.   Figure 4b depicts a much denser network structure; Mor connected more links than Zimmerman, indicating that Mor seemed more active and maybe played a more important role in LTC research. Interestingly, no direct collaborative relationship was established between the two major authors. As shown in Figure 4d, Grabowski (37, Harvard University) and Ribbe (16, VU University Medical Center) were the only two authors who made a great contribution to bridging the collaborative gap between Mor and Zimmerman.
To further identify core authors' research contributions, we listed the top six most cited articles published by Mor and Zimmerman, respectively (see Table 3). Mor, a well-known professor in the Brown University School of Public Health, has a long history of conducting research projects on the quality of nursing home care, the determinants of hospitalization, and racial discrimination in health care treatment. During the past 30 years, Professor Mor frequently adopted an integrated research approach that combined quantitative and qualitative data analyses in program evaluations, such as nursing home resident assessment. Moreover, he is one of the authors of the congressionally mandated Minimum Data Set (MDS) for nursing home resident assessment which was designed to assess elders' functional status and care needs. Overall, the above contributions are fully embodied in the top six most cited articles from Mor. Similarly, Zimmerman was a Kenan Distinguished Professor at UNC-Chapel Hill and a highly representative scholar in geriatrics and gerontology. Over the last 16 years (2002-2017), she was the most productive scholar in LTC research, and has conducted groundbreaking research and intervention studies in nursing homes and assisted living communities and served as a clinician in a range of LTC settings. The six landmark articles from Zimmerman show that she has paid close attention to LTC recipients with dementia.

Intellectual Landscape of LTC Field
In this section, we adopted a rigorous analytical framework of intellectual landscape to identify hot topics, research frontiers, and evolution trends in LTC research based on a dual perspective of references and keywords [23,33]. Two results gained from the analysis were compared and corroborated each other; thus, this study's findings were more comprehensive and definite than those stemming from the single perspective of references or keywords used in previous studies. The analyses were conducted as follows:

Document Co-citation Analysis
Generally, this analysis clusters the related references into groups according to their link strengths; then, the hot research topics can be identified through analysis of the articles in each cluster. The main features of the references' intellectual landscape used in this investigation were: highly cited references; references with high centrality; and references with strong citation bursts (i.e., burst references). The highly cited references characterize the long hotspots of a given research field [29,33]. References with high centrality are considered anchors for the evolutionary path and research root of a discipline because they are major intellectual turning points [30]. Burst references that reflect a strong citation count surge over a short period of time provide predictive indicators of research frontiers and trends [26].
We conducted a co-citation analysis to identify LTC knowledge clusters. The following parameters in CiteSpace were used: (1) time slice from 2000 to 2017; (2) years per slice=1; (3) term source=title/abstract/author keywords/keywords plus; (4) node type=cited reference; (5) pruning=none; (6) select top 50 most cited articles per slice. After running CiteSpace, a total of 273,837 valid references cited in 11,115 articles over an 18-year time span were extracted, and a network consisting of 725 nodes and 2316 links was visualized. We used a short time span because of the sparse research output in the 37-year period from 1963 to 1999, and incomplete records in 2018. The network was divided into 71 clusters, which were automatically labeled by choosing title terms as the labelling source and log-likelihood ratio (LLR) as the standard algorithm. Among these, we focused on the top 10 largest clusters shown in Figure 5. Each node represents one cited reference, and each link indicates the co-citation relationship. Moreover, cluster size measured by the number of references is sequenced in descending order of the cluster numberings. The network was divided into 71 clusters, which were automatically labeled by choosing title terms as the labelling source and log-likelihood ratio (LLR) as the standard algorithm. Among these, we focused on the top 10 largest clusters shown in Figure 5. Each node represents one cited reference, and each link indicates the co-citation relationship. Moreover, cluster size measured by the number of references is sequenced in descending order of the cluster numberings. According to the three features mentioned above, the top 25 most cited references are listed in Table 4, from No. 1 to No. 25, assigned to 10 clusters. To better characterize the interrelationships among the clusters, 13 references with high centrality were presented as No. 1 and from No. 26 to No. 37. Sixteen references (i.e., No. 1, 3-5, 10, 14-16, 24, 25, 32-34, 38-40) with the strongest bursts in the group of references that started to burst at the same time can be adopted to disclose the LTC research frontiers and trends. Detailed descriptions of 40 representative references are shown in Table 4.  According to the three features mentioned above, the top 25 most cited references are listed in Table 4, from No. 1 to No. 25, assigned to 10 clusters. To better characterize the interrelationships among the clusters, 13 references with high centrality were presented as No. 1 and from No. 26 to No. 37. Sixteen references (i.e., No. 1, 3-5, 10, 14-16, 24, 25, 32-34, 38-40) with the strongest bursts in the group of references that started to burst at the same time can be adopted to disclose the LTC research frontiers and trends. Detailed descriptions of 40 representative references are shown in Table 4.  Cluster #0, long-term care, ranked first in cluster size, containing 87 references that were mostly published around 1998 [12,48,50,60]. In the last decades, topics related to LTC changed significantly and numerous research themes developed rapidly. Nonetheless, studies shared a common aim: to meet care needs of residents in LTC facilities. The two most active citers to this cluster were references No. 23 and No. 25. Wallace et al.'s [48] arguments have received much attention, with results suggesting that any expansion of community-based LTC should take into account older minorities' need patterns and potential access barriers. Manton and Gu [50] assessed changes in the prevalence of chronic disability in the United States black and nonblack elderly population based on a 1999 National LTC Survey. Reference No. 35 focusing on the patterns of functional decline at the end of life showed high centrality, linking cluster #0 to clusters #4 and #5.
The second largest cluster (#1) contained 86 references focused mainly on nursing home-related topics published around 2009 [1, 13,14,21,39,40,45,63]. A nursing home (NH) is defined as a facility with a domestic-styled environment that provides 24-hour functional support and care for people who require assistance with activities of daily living and who often have complex health needs and increased vulnerability [64]. There are six highly cited references in this cluster. Among them, three references had a common interest in dementia care [13,14,40]. The report from WHO [40] regarded dementia as a public health priority. Recently there has also been interest in improving quality of care for NH residents by proposing practical programs, like the NH culture-change movement and the interRAI instrument for LTC facilities [21,39,45].
Cluster #2, nursing assistant, ranked third in cluster size, including 84 references that were mostly published around 2004 [4,17,19,53,56,62]. In the NHs, nursing assistants (NAs) are responsible for a considerable amount of direct patient care and resident handling. Recent evidence has shown that the present health care workforce is small and is ill-prepared for the future surge in LTC needs, although NAs are increasingly employed to provide high-quality care for residents [19]. To address this, public health nurses were formally integrated into LTC insurance in Japan [4]. Three references with high centrality concentrated on quality indicators used to measure quality of care, and its relationship to NH profit status and NH staff turnover [17,53,56]. Obviously, quality of care has become a hot theme in LTC research.
There are other clusters worth mentioning. Cluster #3 focused on the SAGE database and its applications [49,52]. The most active citer to cluster #4 [10,57,61] was Kane [10] who wanted to bring LTC and a good quality of life closer together. Four references in cluster #5 reflected a common theme-the prevention and control of common infections in LTC facilities [46,55,58,59]. Cluster #6, ageing society, has been widely used as a context for research on NH admission [35,38], and providing and paying for LTC [5,6,47]. The two most active citers to cluster #7 examined the effects of caregiver influenza vaccination on mortality in elderly people [34,36]. Cluster #8, New York State was commonly chosen as a typical example to examine the frequency, reasons, and costs of potentially avoidable hospitalization or rehospitalization [41,44,51,54], because it has the highest countrywide Medicaid reimbursement rate and generous bed-hold policies [65]. The newly formed cluster #9 with mean publication year 2011 contained three representative references with high citation counts, which revealed a common research theme-frailty in elderly people [37,42,43].
Overall, the focus of these 10 clusters can be divided into 12 hot topics, including health status, mortality, care needs, dementia care, quality of care, formal and informal caregivers, database application, infection control, NH admission, providing and paying for LTC, potentially avoidable hospitalization or rehospitalization, and frailty in elderly people. Moreover, some analytic techniques, including randomized controlled trials and meta-analysis, received increased attention.
The research frontiers in a certain field can be identified by references with strong citation bursts. A citation burst indicates the likelihood that the scientific community has paid or is paying special attention to the underlying contribution [25,27]. In investigating these references, we focused mainly on discussing 16 burst references (i.e., No. 1, 3-5, 10, 14-16, 24, 25, 32-34, 38-40) in Table 4. Moreover, the largest 10 emerging clusters were considered when detecting LTC research frontiers because each cluster represented a thematic concentration in the bibliographic landscape [25,33]. A timeline view of 10 clusters and 16 burst references with their respective research foci is shown in Figure 6. The LTC research trends at different times were revealed as follows: in the early stage from 2000 to 2005, research focused on health status and care needs among elderly LTC residents; in the second stage from 2006 to 2010, focus shifted to aspects of the LTC workforce, such as nurse turnover, vaccination of care home staff, and nurse training; in the third stage from 2011 to 2015, some practical programs aimed at improving quality of care, such as the NH culture-change movement and clinical practice guidelines, received increased attention. In short, LTC research trends have shifted from the demand side (care demanders) to the supply side (caregivers), and from basic studies to practical applications.
In this study, we considered references that underwent significant bursts within the past three years (2015-2017) to represent the newest LTC research frontiers. Among 16 burst references in Figure 6, some recent bursts, such as references No. 4, No. 14, No. 15, and No. 16 were identified; bursts of these four references are expected to continue to lengthen in the future and may be deserving of more attention, because bursts are often followed by subsequent increasing trends. Among 10 clusters, cluster #9 with mean publication year 2011 emerged as the newest research frontier. Moreover, to analyze the research frontiers more accurately, other references with strong recent bursts, shown in Table 4, were also identified as predictive indicators, such as references No. 9, No. 11, and No. 17. The new research frontiers in the LTC field can be summarized as follows: frailty in elderly people (#9,No. 4,No. 16,and No. 17), the NH culture-change movement (No.9 and No. 14), dementia care (No. 11), and LTC workforce and LTC financing (No. 15). second stage from 2006 to 2010, focus shifted to aspects of the LTC workforce, such as nurse turnover, vaccination of care home staff, and nurse training; in the third stage from 2011 to 2015, some practical programs aimed at improving quality of care, such as the NH culture-change movement and clinical practice guidelines, received increased attention. In short, LTC research trends have shifted from the demand side (care demanders) to the supply side (caregivers), and from basic studies to practical applications. In this study, we considered references that underwent significant bursts within the past three years (2015-2017) to represent the newest LTC research frontiers. Among 16 burst references in Figure 6, some recent bursts, such as references No. 4, No. 14, No. 15, and No. 16 were identified; bursts of these four references are expected to continue to lengthen in the future and may be deserving of more attention, because bursts are often followed by subsequent increasing trends. Among 10 clusters, cluster #9 with mean publication year 2011 emerged as the newest research frontier. Moreover, to analyze the research frontiers more accurately, other references with strong recent bursts, shown in Table 4, were also identified as predictive indicators, such as references No.

Keyword Co-Occurrence Analysis
The main features of the intellectual landscape of LTC keywords were three-fold: high-frequency keywords; keywords with high centrality; and keywords with strong citation bursts (i.e., burst keywords). Specifically, we conjectured the hot topics by extracting high-frequency keywords [29,33]. Keywords with high centrality represent major intellectual turning points connecting other keywords, while burst keywords represent new research frontiers [25,27,30].
To illustrate the hotspots in LTC research, we conducted a keywords co-occurrence analysis. The parameters in CiteSpace were remained the same except node type being changed from "Cited Reference" to "Keyword". Keywords used in our analysis included "Author Keywords" which were supplied by authors, and "Keywords Plus" which were supplied by the journals. To avoid potential misunderstandings, some similar keywords were combined. For example, "long-term-care" and "long term care" were merged into "long-term care." Notably, we also used a short time span (2000-2017) to construct the network of co-occurring keywords. The network consisting of 138 nodes and 841 links is shown in Figure 7. Each node represents one keyword; bigger nodes reflect higher co-occurrence frequency and each "keyword to keyword" link indicates the co-occurrence relationship. The main features of the intellectual landscape of LTC keywords were three-fold: high-frequency keywords; keywords with high centrality; and keywords with strong citation bursts (i.e., burst keywords). Specifically, we conjectured the hot topics by extracting high-frequency keywords [29,33]. Keywords with high centrality represent major intellectual turning points connecting other keywords, while burst keywords represent new research frontiers [25,27,30].
To illustrate the hotspots in LTC research, we conducted a keywords co-occurrence analysis. The parameters in CiteSpace were remained the same except node type being changed from "Cited Reference" to "Keyword". Keywords used in our analysis included "Author Keywords" which were supplied by authors, and "Keywords Plus" which were supplied by the journals. To avoid potential misunderstandings, some similar keywords were combined. For example, "long-term-care" and "long term care" were merged into "long-term care." Notably, we also used a short time span (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017) to construct the network of co-occurring keywords. The network consisting of 138 nodes and 841 links is shown in Figure 7. Each node represents one keyword; bigger nodes reflect higher co-occurrence frequency and each "keyword to keyword" link indicates the co-occurrence relationship.
Similarly, we identified 34 representative keywords in terms of the three features mentioned above. The top 17 keywords with a co-occurrence frequency over 550 are listed in Table 5 Additionally, 17 keywords with the strongest burst in the group of keywords that started to burst at the same time are also summarized in Table 5  The top two keywords in terms of co-occurrence frequency were "long-term care" (4,837) and "elderly people" (2,351). Accordingly, elderly people were the main subjects of LTC research. Keywords with high centrality were observed in "long-term care," "elderly people," "nursing home," "dementia," and "nursing home resident," which represented major intellectual turning Similarly, we identified 34 representative keywords in terms of the three features mentioned above. The top 17 keywords with a co-occurrence frequency over 550 are listed in Table 5  The top two keywords in terms of co-occurrence frequency were "long-term care" (4837) and "elderly people" (2351). Accordingly, elderly people were the main subjects of LTC research. Keywords with high centrality were observed in "long-term care," "elderly people," "nursing home," "dementia," and "nursing home resident," which represented major intellectual turning points linking different keywords with significantly influenced LTC research development.
As for these 17 high-frequency keywords, according to previous scientometric studies, they can be directly regarded as LTC research hotspots. However, we believe that the hot topics identified in this way are too broad and macro to focus clearly on the major issues. To address the gap, we identified the hot research topics by integrating 17 high-frequency keywords and considering the co-occurring keywords shown in Figure 7. The resulting five main hotspots were as follows: 1.
Dementia care was extracted using four keywords "long-term care," "elderly people," "dementia," and "long-term care facility." With an increasing number of people suffering from dementia, considerable attention is currently focused on improving quality of life for people with dementia [66]. Patients and caregivers are the central stakeholders in dementia care and there has been much research on the health status and care needs of elderly patients with dementia [13,14,67,68]. In terms of dementia caregivers, especially informal caregivers, compelling evidence has suggested that they suffer a high level of care burden [40,66]. Some care models for people with dementia have received increased attention, such as the ABLE model [67], person-centered care model [68], and personhood model [69].

2.
Quality of care was identified using six keywords "long-term care," "nursing home," "health care," "quality of life," "quality," and "outcomes." The quality of care for NH residents has been a major concern internationally. Despite great improvements in NHs, LTC quality in the context of population ageing remains challenging. Within this theme, three major aspects were addressed: (1) the indicators for measuring quality of care, such as depression symptoms, psychotropic drug use, physical restraints, pain management, pressure ulcers, fall incidents, and mortality rate [53,56]; (2) factors influencing quality of care, such as nurse-staffing levels, NH profit status, and the education level and training of nursing staff [40,56]; and (3) the strategies for improving quality of care, such as implementing health information technology, enhancing quality monitoring systems, strengthening the caregiving workforce, and implementing culture-change practices in NHs [5,19,21].

3.
Prevalence and risk factors were extracted using the keywords "prevalence," "risk factor," and "infection." Psychiatric disorders are common among elderly people. In an effort to better understand and treat the diseases, some researchers have investigated the prevalence of and risk factors for common psychiatric disorders in LTC facilities, such as dementia, depression, bipolar disorder, anxiety disorders, schizophrenia, and alcohol use disorders [14]. Moreover, much evidence indicates a marked increase in the incidence of healthcare-associated infections caused by multidrug-resistant organisms (MDROs). Hence, prevention and control of MDROs has become a public health priority. In recent years, considerable attention has been paid to determining the prevalence of and risk factors for MDROs, such as methicillin-resistant staphylococcus aureus (MRSA), multidrug-resistant gram-negative bacilli (GNB), and vancomycin-resistant enterococci (VRE) [70].

4.
Mortality comprised five representative keywords "long-term care," "elderly people," "mortality," "people," and "population." Patient death becomes an increasingly common occurrence in LTC literature. Identifying mortality and its predictors is important for implementing of therapeutic management for high-risk patients, with the goal of improving survival. In an LTC context, the focus of this theme can be summarized into two main aspects: (1) mortality trends and differences between different groups, such as mortality of residents with regard to age, gender, and distribution of care levels under home-based or institutional care [71]; and (2) the factors associated with mortality, such as demographic characteristics, functional and cognitive status, specific diseases (cancer and heart disease), antipsychotic drug use, social support, and influenza vaccination of health-care workers [34,72].

5.
Randomized controlled trial was extracted using one keyword "randomized controlled trial." As a form of scientific experiment, it aims to evaluate the effectiveness of various types of medical interventions [34,36]. For example, one such trial was designed to determine whether vaccination of health-care workers can lower mortality and the frequency of laboratory-proven influenza infection in elderly patients in long-term care hospitals [34].
Moreover, 17 burst keywords in Table 5 were considered as indicators of research frontiers. Figure 8 shows their evolution paths. To analyze the trends more accurately, some new keywords from each year were also listed in the figure because their publication time can reflect the research evolution trends [29]. Forty-one new keywords from each year were identified. In terms of the time evolution of these keywords, LTC research trends can be categorized into three phases: in the first phase (2000)(2001)(2002)(2003)(2004)(2005), the keywords reflected the topics on resident-related health or functional status and care demand; in the second phase (2006-2010), the focus shifts to caregiving-related workforce, such as keywords "nurse," "nursing," and "family caregiver;" and in the last phase (2011-2016), the keywords indicated an increased focus on practical exploration of the LTC field.
The keywords showing occurrence bursts within the past three years (2015-2017) were cognitive impairment (16.94, 2014-2017), frailty (46.40, 2015-2017), association (36.48, 2015-2017), and women (9.18, 2015-2017); these findings indicated that a much research attention was directed to these areas. Thus, the emerging LTC research frontiers were summarized as follows: cognitive impairment, frailty in elderly people, association analysis, and LTC for older women. evolution of these keywords, LTC research trends can be categorized into three phases: in the first phase (2000)(2001)(2002)(2003)(2004)(2005), the keywords reflected the topics on resident-related health or functional status and care demand; in the second phase (2006-2010), the focus shifts to caregiving-related workforce, such as keywords "nurse," "nursing," and "family caregiver;" and in the last phase (2011-2016), the keywords indicated an increased focus on practical exploration of the LTC field.  (36.48, 2015-2017), and women (9.18, 2015-2017); these findings indicated that a much research attention was directed to these areas. Thus, the emerging LTC research frontiers were summarized as follows: cognitive impairment, frailty in elderly people, association analysis, and LTC for older women.

Conclusions
To the best of our knowledge, this study is the first scientometric analysis of global LTC research over the past decades. The overall investigation included: (1) the scientific output and citations of LTC research were depicted to examine its current status and development; (2) the contributions of major countries/regions, institutions, and authors were identified by visualizing collaboration networks; and (3) the hot topics, new frontiers, and evolution trends in LTC research were explored through integrating document co-citation analysis and keyword co-occurrence analysis. The remarkable findings were as follows: A noticeable upward trend in LTC research publications over time suggested that increasing attention has been directed toward LTC. The number of publications was estimated to reach 1217 in 2018 and 1302 in 2019, confirming that LTC research is likely to remain active in the next few years. The polynomial trendline of citations indicated an inverse U-shaped relationship, especially in the period 1991-2018.
LTC research scientific publications primarily originated from the US, Canada, and England, and these three contributors formed a leading research group. Interestingly, we found that the collaborative relationships among them were not strong. To our minds, their weak collaboration might result from the reason from different LTC systems. Moreover, developing countries were still under-represented in the global research network. The top 15 most productive institutions all came from North America and Europe, confirming the importance of their contributions to LTC. Harvard University was the most powerful institution because it received high publication counts and high centrality simultaneously. Meanwhile, we noted that institutions from Asian countries have accelerated their efforts to participate in LTC research. Mor and Zimmerman were the most productive and high-impact authors, although there was no direct collaboration between them. Grabowski and Ribbe played an important role in bridging the collaborative gap between Mor and Zimmerman.
The results of visualizing the intellectual landscape of references were consistent with those of visualizing the intellectual landscape of keywords. Simultaneous consideration of both revealed that the common LTC research hot topics in the 2000-2017 period were dementia care, quality of care, prevalence and risk factors, mortality, and randomized controlled trial. The LTC evolution trends showed three stages: an early stage (2000)(2001)(2002)(2003)(2004)(2005), where research was primarily focused on functional, cognitive and health status, and care demand among older LTC residents; a second stage (2006)(2007)(2008)(2009)(2010), where the focus shifted to caregiving-related workforce factors, such as nurse turnover, nurse training, and home care staff vaccinations; and a third stage (2011)(2012)(2013)(2014)(2015), where several practical explorations aimed at improving quality of care and clinical practice guidelines received considerable attention, such as the NH culture-change movement. This reflects a shift in LTC research trends from the demand side (care demanders) to the supply side (caregivers), and from basic studies to practical applications. In summary, LTC research has become substantially broader and deeper. The common research frontiers were frailty in elderly people and dementia care (closely related to cognitive impairment).
These results provide valuable information to LTC researchers and practitioners. A variety of visualized networks offer an in-depth understanding of the major countries/regions, institutions, researchers, hot topics, evolution trends, and new research frontiers. Moreover, for LTC practitioners, this study presents accurate information regarding the key authors and institutions best suited to assist in developing LTC policies. This scientometric review method can also be used to visualize the status and trends of other research topics. However, there was also a limitation in terms of our scientometric analysis. The retrieval strategy, i.e., treating "long-term care" as the only search term, still has room to be improved. While this study is a good starting point for reviewing the literature on LTC, for researchers who want to deeply delve into the field, they are suggested using search terms other than "long-term care" and connecting them with a Boolean operation to LTC.