Research Trends on Living Donors for Liver Transplantation: A Text Network Analysis and Topic Modeling
Article information
Abstract
Purpose
This study aimed to identify research topics and trends on living liver donors over time through text network analysis and topic modeling.
Methods
Five electronic databases (PubMed, CINAHL, Embase, Web of Science, and PsycINFO) were reviewed for studies published through September 2023, and 392 studies were included. Text network analysis was used to identify the basic characteristics and centrality of the network. The topics were named after extracting meaningful topics through topic modeling using latent Dirichlet allocation.
Results
A total of 1,111 keywords were extracted from the abstracts of 392 selected studies, among which “length of stay,” “morbidity,” “mortality,” “pain,” and “quality of life” showed high frequency and centrality. Through topic modeling analysis, the following four topics were derived: objective health indicators (topic 1), subjective health indicators (topic 2), hepatobiliary-related indicators (topic 3), and early health indicators (topic 4). An analysis of trends in these topics over time showed that the proportion of topics 1, 3, and 4 increased or remained stable. In contrast, there was no significant change in topic 2, representing subjective health indicators.
Conclusion
This study explored research trends on living liver donors using text network analysis and topic modeling. Based on the main topics derived, research on postoperative outcomes for living liver donors has focused on objective health indicators, hepatobiliary-related indicators, and early health indicators compared to subjective health indicators. We suggest that future studies utilize integrated indicators of physical and psychosocial aspects.
INTRODUCTION
Living donor liver transplantation (LDLT) is an alter-native treatment crucial for the survival of patients with end-stage liver disease, and the number of LDLT cases in Korea is steadily increasing [1]. Recipients benefit from optimizing their health status for scheduled surgery [2,3], and advancements in treatment methods have the poten-tial to yield more positive outcomes for recipients after transplantation [3,4]. However, concerns remain regarding donors developing physical complications [5,6] and psychosocial problems after surgery [6,7]. These problems indicate that donors may incur additional follow-up and societal costs [8]. Therefore, continued attention should be paid to the donors’ postoperative health status and needs, and an understanding of these aspects is required.
Donors experience a complex journey that requires change and adaptation to their physical and psychosocial status throughout the donation process, which includes decision-making, donor suitability assessments, postoperative liver regeneration, and functional recovery [9,10]. Therefore, to comprehensively understand the physical and psychosocial aspects of postoperative donors and meet their needs, it is necessary to review the research on donors to improve understanding and expand knowledge. Previous studies have used systematic reviews to understand donors’ physical and psychosocial outcomes after surgery [11-14]. Nevertheless, the results for postoperative physical function, complications [13,14], and well-being [11,12] using existing methods have limitations in estab-lishing relationships between topics and identifying keywords that have not been revealed in previous studies. Therefore, new research methodologies that reflect meth-odological flexibility and epistemological diversity are needed to analyze the postoperative outcomes of donors comprehensively.
As big data analysis methods, text network analysis and topic modeling present novel approaches for identifying research trends in donors. Text network analysis facili-tates an intuitive understanding of context by identifying influential keywords in extensive texts and illustrating their relationships [15]. Topic modeling offers the advant-age of extracting key topics from data and analyzing their relevance, thereby providing comprehensive information for identifying topics and topic trends over time from a macro perspective [16,17]. To our knowledge, overall research trends have not been explored using text network analysis and topic modeling for living liver donors.
Therefore, this study aimed to identify 1) keywords based on network centrality indicators in donor-related studies using text network analysis and 2) topic trends in donor-related studies using topic modeling. These methods provide basic data that can be used from a new perspective in future studies.
METHODS
1. Research Design
This study was designed to extract keywords and identify topic trends using text network analysis and topic modeling, focusing on the abstracts of studies related to living liver donors.
2. Data Collection
A literature search was conducted for studies published up to September 2023 in five electronic databases (Pub-Med, CINAHL, EMBASE, Web of Science, and PsycINFO) using the keywords “ liver transplantation” and “ living liver donor.” The initial search yielded 6,299 studies. After removing 1,018 duplicates, the titles and abstracts of 5,281 studies were reviewed. Studies that did not meet the eligibility criteria were excluded (n=4,889), and 392 studies were included in the final analysis. The screening process excluded studies for the following reasons: studies involving adult-to-child (pediatric) LDLT (n=697), focusing on donor suitability evaluation (n=138), other organ type (e.g., kidney; n=63), not targeting living liver donors (n=2,067), focusing on surgical procedures or methods (n=146), involving deceased donor liver transplantation (n=1,143), literature reviews, dissertations, abstracts (n=635). In the screening process, the researchers (SC and WS) independently screened titles and abstracts based on the eligibility criteria. At each step, the researchers discussed the resolution of any disagreements and reached a consensus regarding the eligibility of each study.
3. Data Analysis
A total of 392 studies were analyzed using NetMiner version 4.5 [18]. The abstracts of each study were inserted into NetMiner in an Excel spreadsheet format for analysis. The analysis process comprises data preprocessing and dictionary construction, followed by the extracting of top keywords, identifying important keywords within documents, text network analysis, and topic modeling.
1) Data preprocessing
For data preprocessing, the words in the abstracts were extracted after converting them to lowercase letters, and the parts of speech of the extracted words were designated as nouns to identify the main concepts. Two researchers (SC and WS) independently reviewed the extracted words, and through discussion, a dictionary comprising defined words, a thesaurus, and stopwords was constructed to extract the words for analysis. In the dictionary with the defined words, phrases comprising two or more words or clauses that convey a unified meaning as a single phrase, such as “ length of stay” and “ quality of life,” are recognized as a single word. In the dictionary with the thesaurus, “ quality of life” and its abbreviation “ QoL” were specified as synonymous. Finally, the dictionary with the stopwords includes words related to research method-ology, such as “ background” and “ method,” which are commonly used in abstracts, and “ donor” and “ liver,” which were used as search terms.
2) Top keywords
Keywords were extracted from the abstracts included in the final analysis based on dictionaries constructed during data preprocessing to obtain the term frequency and term frequency-inverse document frequency (TF-IDF), and the top 30 words were extracted for each. TF-IDF divides the frequency of a particular word in a document based on the frequency of its occurrence in all documents containing the word. The more frequently a word is used across all documents, the closer its value is to zero; its value in-creases when used in fewer documents [19]. Therefore, TF-IDF can determine whether a word carries significant meaning within a document, and a higher value indicates that it is an important keyword within the document [20].
3) Text network analysis
A text network analysis was used to identify the basic characteristics and centrality of the network. For the text network analysis, the 2-mode network of “ document-word” was converted into a 1-mode network of “ keyword-keyword.” We identified the density, average degree, and average distance as the basic characteristics of the network. Density is the degree of linkage between nodes, and a higher density indicates a higher degree of linkage [21]. The average degree represents the number of words connected to a word; a greater number of connected words indicates greater word influence [22]. The average distance is the average value of the shortest distance between words, and a shorter average distance indicates a faster spread [23].
Centrality refers to the degree of importance within a network [24], and closeness, between, and degree centralities are analyzed. The closeness centrality measures how close a word is to other words, with a higher value indicating that it is located at the center of the network [25]. Between centrality is the degree to which a word is located between other words. The more central it is, the more influential it is in controlling the flow of information [25]. Finally, degree centrality refers to the degree to which a word co-occurs with other keywords. Words with high degree centrality values are located at the core of the network and represent topics [25].
4) Topic modeling
Topic modeling is used to estimate the probability of a topic occurring within an unstructured document [26]. In this study, the topic modeling method of latent Dirichlet allocation (LDA) was employed to identify the probability distribution of keywords highly relevant to a topic [26]. Topic modeling used the following parameters: ⍺=0.01, beta=0.01, and 1,000 iterations. Multiple simulations were performed to characterize each topic. The number of topics was determined by consensus, and each topic was named after reviewing its categorization.
RESEARCH FINDINGS
A total of 392 studies were included in the final analysis: 4 studies from before 2000, 34 studies from 2001-2005, 56 studies from 2006-2010, 132 studies from 2011-2015, 105 studies from 2016-2020, and 61 studies from 2021-2023. Notably, the number of donor-related studies increased since 2011.
1. Keyword and Frequency Analysis of the Study
A total of 1,111 keywords were extracted from the abstracts of the studies, and the top 30 keywords with the highest term frequency and TF-IDF were identified (Table 1). The most frequent terms were “ morbidity,” “ pain,” “ mortality,” “ length of stay,” and “ quality of life,” while the highest TF-IDF values were associated with “ length of stay,” “ morbidity,” “ mortality,” “ quality of life,” and “ pain,” in that order. Differences were observed in the keyword rankings extracted by term frequency and term frequency-inverse TF-IDF.
Top 30 Keywords of the Included Studies
2. Text Network Analysis
The network analysis revealed 1,111 keywords and 23,479 links. The network density was 0.04, with an average degree of 42.27, and an average distance of 2.21. Table 2 shows the top 30 keywords for centrality in the network. The seven keywords with the highest centrality - closeness centrality, between centrality, and degree centrality - in the network are “ length of stay,” “ morbidity,” “ mortality,” “ pain,” “ need,” “ recovery,” and “ quality of life,” in that order. Words excluding these seven keywords had different rankings according to their centrality.
High-Ranked Keywords by Closeness Centrality, Between Centrality, and Degree Centrality using Network Analysis
3. Topic Modeling
Topic modeling using LDA revealed four significant topics (Table 3). The researchers named each of these extracted topics based on the association between the main keywords, context, and purpose of this study.
Results of Topic Modeling
Topic 1 comprised the highest proportion (38.5%) and included keywords such as “ morbidity,” “ mortality,” “ length of stay,” “ steatosis,” and “ bilirubin.” Considering the relevance and context of these keywords, the topic was named “ objective health indicators” as it pertained to postoperative outcomes reported in statistical values. Topic 2 included keywords such as “ quality of life,” “ pain,” “ health,” “ satisfaction,” and “ recovery,” and it focused on the psychosocial outcomes of donors. This was named the “ subjective health indicators.” Topic 3, representing the second highest proportion (30.9%), included keywords such as “ reconstruction,” “ stricture,” “ biliary,” “ liver regeneration,” and “ length of stay,” indicating struc-tural characteristics of the hepatobiliary system. Therefore, it was categorized as a hepatobiliary-related indicators.” Topic 4 had the lowest proportion (9.2%) and included keywords such as “ pain,” “ length of stay,” “ morphine,” “ recovery,” and “ platelet,” referring to short-term con-ditions related to surgical site pain. This topic was named “ early health indicators”(Figures 1, Figure 2-A). “ Length of stay” was a common keyword across topics 1, 3, and 4, while “ recovery” and “ pain” were common across topics 2 and 4 (Figure 1).
Topic network of main keywords.
Trends of topic.
The relative trend for each topic over time showed that the proportions of topics 1 (objective health indicators) and 3 (hepatobiliary-related indicators) either increased or remained constant. Topic 4 (early health indicators) first appeared after 2001-2005, and consistently accounted for the smallest proportion in each period. Topic 2(subjective health indicators) remained relatively constant throughout the study period. Notably, topic 3 (hepatobil-iary-related indicators) had the highest proportion of recent studies (Figure 2-B).
DISCUSSION
This study aimed to explore research trends in living liver donors using text network analysis and topic modeling. According to our findings, keywords such as “ length of stay,” “ morbidity,” “ mortality,” “ pain,” “ need,” “ recovery,” and “ quality of life” played a significant role in research on living liver donors. The four topics identified in each period represent indicators related to postoperative outcomes, and in particular, postoperative complications accounted for more than half of the topics in each period. These findings offered valuable evidence for research on trends and topics related to postoperative outcomes for living liver donors.
The analysis of the 392 studies included in this review suggests that the increase in and steady publication of donor-related articles since 2011 is likely due to the increased number of LDLT performed [1,27], leading to an increased interest in postoperative outcomes for donors. Although there were differences in the ranking of topics in each period, “ objective health indicators” and “ hepatobiliary-re-lated indicators” were the most dominant topics overall, with “ early health indicators” emerging as a topic with a temporary but noticeable increase in recent years. Con-versely, subjective outcomes have received relatively less attention. The relative importance of the dominant topics changes over time but suggests that more attention and research have been focused on medical and physical outcomes. These findings emphasize the need to further in-vestigate the various aspects of donor postoperative care from a nursing perspective.
Among the keywords identified by the network analysis, “ length of stay” showed a relatively higher frequency and centrality. Donor hepatectomy is a major abdominal surgery, the length of stay is an important indicator for monitoring the patient's physical and psychological status and planning any necessary additional medical support or treatment before discharge so that they can return to their daily life [28-32]. Despite donors’ low rates of severe complications and relatively good physical condition after surgery [5,6], emphasis on this keyword highlights the neces-sity of contextualized and individualized treatment and management of donors during their transition to becoming patients. Therefore, it is important to consider the postoperative length of stay as an important factor in improving the approach to donor postoperative management and optimizing postoperative nursing care.
As a result of topic modeling, “ objective health indicators” and “ hepatobiliary-related indicators” were found to be related to the postoperative complications of the donors, accounting for 69% of the total. According to results reported using the Clavien-Dindo classification, the frequency of life-threatening complications was minimal [27,33], and hepatobiliary complications were reported in ap-proximately 2∼18% of cases [27,34]. Nevertheless, some donors may require medical treatment or prolonged hospitalization [35-37], which can reduce health-related qual-ity of life [38] and threaten mental health, such as the de-velopment of anxiety or alcohol use disorders [7]. Donor safety after surgery is of utmost importance [39]. Therefore, it is important to ensure the long-term well-being of donors by identifying postoperative complications and monitoring the risk factors for postoperative outcomes.
The “ subjective health indicators” focused on quality of life showed consistent publication rates over time, with no notable fluctuations compared with other topics. According to a meta-analysis study, donor quality of life did not differ significantly before and after surgery but differed at each time point [11]. These findings showed that donors’ quality of life changed during a specific period after surgery and indicated that quality of life evaluation and management are necessary longitudinally. Additionally, the donor's postoperative quality of life is influenced by donation-related characteristics, such as decision-making [41], emergency surgery [38], relationship with the recipient, and the recipient's postoperative outcome [12,42,43]. None-theless, only a few long-term studies on donor quality of life have been conducted, and these studies have predom-inantly relied on generic quality-of-life assessment tools [8,40], which do not fully capture the unique circum-stances of donors. Therefore, the donors’ postoperative quality of life requires cautious interpretation, and it is necessary to emphasize the need for further investigations at various postoperative time points.
The “ early health indicators,” which mainly deal with pain, demonstrated through the topic network that pain can be related to subjective outcome indicators after surgery as well as indicators related to postoperative complications. Postoperative pain is considered an important indicator of early recovery, which influences the length of hospitalization [44]. Recent studies have increasingly recognized it as an important factor in early health indicators. Postoperative pain is related to several factors, including the surgical procedure, pain control method, and postoperative complications [44]. However, the donor's pain response may vary depending on donation-related con-cerns, decision-making motivation, and the recipient's postoperative health status [45]. These characteristics suggest that pain management may not be effective for donors in patients undergoing other surgical procedures may not be effective in donors [45]. As the contextual and psychosocial characteristics of the donor may influence the response to pain and physical recovery, early postoperative pain management should reflect a multifaceted assessment that includes medical aspects and the contextual and psychosocial characteristics of the donor [45,46].
This study had several limitations. First, only abstracts collected using keywords determined by the researchers were analyzed. Therefore, the abstracts included in the analysis may have been limited by the search terms. Second, the results should be generalized cautiously because this study focused on high-frequency and centrality keywords. Finally, topic modeling requires caution in interpretation because subjective standards based on researchers’ evaluations may be reflected during keyword refining and post-analysis interpretation processes.
CONCLUSION
This study explored the research trends in living liver donors using text network analysis and topic modeling. The knowledge structure was identified through keywords with high frequency and centrality, and the need for research on psychosocial health, which was relatively insufficient, was emphasized through derived topics. These findings provide an integrated understanding of donors in research and clinical practice and insight into individualized management strategies after surgery. Future research prospectively explores the postoperative evaluation of donors, including multifaceted factors such as physical and psychosocial aspects, and these data are expected to be used as basic data for future intervention develop-ments.
Notes
CONFLICTS OF INTEREST
The authors declared no conflict of interest.
AUTHORSHIP
Conceptualization - Choi S and Kim M; Design - Choi S, Kim M and Seo WJ; Resources and acquisition of data - Seo WJ; Analysis and interpretation of data - Choi S, Kim M and Seo WJ; Writing-original draft preparation - Choi S and Seo WJ; Writing-review and editing - Kim M; Supervision - Kim M.
DATA AVAILABILITY
The data that support the findings of this study are avail able from the corresponding author upon reasonable request.
ACKNOWLEDGMENTS
This research was supported by the Brain Korea 21 FOUR Project funded by the National Research Foundation (N RF) of Korea, Yonsei University College of Nursing.