A Metrology Informatics Investigation of Conversion Therapy in Hepatocellular Carcinoma: 2014–2023

Abstract

Background:

With advancements in drug therapy, local treatments, and evolving concepts, conversion therapy has shown benefits for patients initially diagnosed with unresectable hepatocellular carcinoma (HCC). Over the past 10 years, the conversion therapy field has accumulated a vast amount of underutilized data, necessitating in-depth bibliometric evaluation.

Methods:

This cross-sectional retrospective study collected a substantial amount of research on conversion therapy published between 2014 and 2023, adhering to strict search criteria. The primary outcomes were publication volume, citation count, and interstudy relationships. Comprehensive analysis was conducted using unsupervised hierarchical clustering, spatiotemporal analysis, regression model, and the Walktrap algorithm.

Results:

Over the past decade, conversion therapy has demonstrated significant progress, with an annual growth rate of 23.0%. Post-2020, these metrics saw a marked increase, reaching 116 publications and 1943 citations by 2023. Cluster analysis grouped 244 authors into 17 clusters, highlighting early and sustained contributions from Western authors compared with later-emerging Eastern authors. Research characteristics in HCC conversion therapy were classified into 5 clusters, with Cluster 2 (Target Therapy and Immunotherapy) emerging as a new focus. Thematic analysis categorized research characteristics into 4 quadrants, identifying “immune checkpoint inhibitor” and “combination therapy” as highly relevant and rapidly developing themes, while “hepatic arterial infusion chemotherapy” and “radioembolization” show high potential for future research.

Conclusions:

This study highlights key contributors and emerging trends and provides important predictions for future research directions. To achieve effective conversion therapy for HCC, researchers may prioritize immunotherapy and locoregional treatments such as hepatic arterial infusion chemotherapy or radioembolization.

INTRODUCTION

Hepatocellular carcinoma (HCC) is a primary malignancy of the liver and one of the most common cancers worldwide.¹ Although surgical resection remains the primary route to long-term survival for HCC patients, most are diagnosed at an advanced stage, with less than 30% eligible for surgery.² For patients with advanced, unresectable HCC, the overall prognosis is poor, with a 5-year survival rate of less than 20%.³ Conversion therapy aims to downstage initially unresectable HCC to a resectable state through various treatment modalities, offering survival benefits for patients initially deemed ineligible for surgery.⁴

Conversion therapy for HCC is an evolving field, with ongoing research aimed at improving techniques, optimizing therapy combinations, and identifying biomarkers for better patient stratification.⁵ Over a decade, with advancements in drug therapy, local treatments, and evolving concepts, this field has accumulated a vast amount of unstructured data, which continues to grow annually.⁶ This increasing volume of diverse information makes it challenging for scholars to quickly understand the interconnections and evolutions within the field, identify key research questions, and guide future developments. Bibliometric analysis evaluates the impact and trends of academic research by quantitatively examining published literature, but presently, no bibliometric analysis has been conducted on conversion therapy for HCC. Therefore, comprehensive informatics analysis and comparison of key research characteristics will be valuable for subsequent preclinical and clinical studies.⁷

From a machine learning-based bibliometric evaluation perspective, this cross-sectional study aims to present fundamental bibliometric data, intrinsic relationships among studies, evolutionary processes, and interconnections of key research elements. This approach will help identify critical research features and provide important predictions for future clinical studies related to HCC conversion therapy.

METHODS

Patients were not involved in this research; therefore, informed consent was waived. No ethics approval was needed for this study as it solely involved the analysis of deidentified publicly available data.

Data Source

In the current biomedical field, popular literature databases include Web of Science, PubMed, Scopus, Embase, MEDLINE, OVID, and Cochrane.⁸ These databases often contain a significant amount of duplicate data. Although data cleaning can mitigate this issue to some extent, including data from various sources may introduce unnecessary confounding factors and reduce data quality. Using a single high-quality, authoritative database as the data source can not only represent the entire research field but also ensure minimal confounding factors and optimal data quality. Due to its comprehensive coverage, high authority, and excellent traceability, this informatics analysis study used the Web of Science Core Collection as the data source.

Data Retrieval and Collection

Using the Web of Science Core Collection, data filtering was conducted via its advanced search function. The search formula was “TS=(“hepatocellular carcinoma” OR “hepatocellular cancer” OR “liver cancer” OR “hepatic neoplasm” OR “HCC”) AND TS=(“conversion therapy” OR “downstaging” OR “downstage” OR “preoperative treatment” OR “preoperative therapy”)”. The data covered the period from January 1, 2014, to December 31, 2023. This study only included English-language literature. The raw data were exported on April 1, 2024. The exported raw data file was in plain text format, containing all records for further analysis.

Quality control was a process that ensured the reliability of research data and results, effectively addressing potential errors and biases in medical research. A homogenization process was applied during data collection and processing. In addition, the standardization and repetition of the data collection process significantly reduced biases in the final results. One of the researchers was responsible for collecting the raw data, ensuring adherence to standardized procedures and methods, and recording all necessary parameters. Subsequently, 2 other individuals reviewed the collected data, verifying its format, range, and consistency.

Hierarchical Clustering and Visualization

Unsupervised hierarchical clustering, a machine learning method, was applied using VOSviewer in this study. The basic principles and steps are as follows: in the dataset, specific objects frequently co-occur, indicating co-occurrence relationships. The co-occurrence frequency of 1 object with another represents the connection strength between these 2 objects. The total co-occurrence frequency of an object with all other objects is the total connection strength of that object. The co-occurrence relationships between every 2 objects in the dataset can be converted into a co-occurrence matrix, where each element represents the connection strength between 2 objects. Hierarchical clustering algorithms are applied to cluster the co-occurrence matrix, categorizing data objects with similar co-occurrence patterns into the same cluster.⁹ Keyword Plus was selected for analysis to obtain more extended information. Research hotspots with a frequency of less than 4 occurrences in the keywords (including author keywords and Keyword Plus) were filtered out to generate objects for hierarchical clustering. The overall semantic information in the clusters annotated the top 5 major clusters. Finally, different visual representations of the data were achieved using VOSviewer by setting various parameters (eg, average publication year, total connection strength, and occurrence frequency [OF]).

Burst Analysis

Burst analysis provides information on the concentration of research topics over continuous periods.⁹ Quantitative processing for this analysis used the R package “bibliometrix.” Some necessary parameter settings are as follows: field = author keywords; time span = 2014–2023; minimum frequency of words = 4; number of words per year = 4.

Regression Analysis

Keyword Plus in the Web of Science database is automatically generated by computer technology based on the titles of cited references. For accurate statistical results, this analysis included only author keywords, filtering out research topics with a frequency of less than 4 occurrences as they do not achieve statistical significance and would increase unnecessary workload. Quantitative processing for this analysis used the R package “bibliometrix.” Regression curve plotting and statistical analysis were conducted using GraphPad Prism 9.0. “s” represents the slope of the fitted curve, and “R²” represents the correlation between the 2 variables.

Walktrap Algorithm and Research Frontier Identification

The Walktrap algorithm is a spectral clustering algorithm based on the random walk strategy for community network analysis. First, data points are converted to nodes on a graph; then, the similarity matrix and Laplacian matrix are calculated to determine the similarity and distance between nodes. Finally, community classification is obtained through eigen decomposition of the Laplacian matrix. The Walktrap algorithm in this study was executed by the R package “bibliometrix.” Four quadrants were obtained through the Walktrap algorithm. The horizontal axis represents the centrality or importance of the research topic within the entire network, with higher centrality values indicating greater importance or influence in the field. The vertical axis represents the development level of the research topic, with higher density values indicating more active status or higher development in research. Quadrant I (high centrality, high density): located in the upper right, represents research topics closely related to the field and fully developed. Quadrant II (low centrality, high density): located in the upper left, represents relatively important research topics with lower centrality in the field, including some specific subtopics or small research areas. Quadrant III (low centrality, low density): located in the lower left, represents relatively unimportant and less developed research topics, possibly marginal themes or less relevant concepts. Quadrant IV (high centrality and low density): located in the lower right, represents research topics of considerable importance in the field but still underdeveloped. Notably, topics in Quadrant IV have the most research potential.⁹

Statistical Analysis

VOSviewer was used for hierarchical clustering and visualization.¹⁰ Quantitative information on research hotspots in the 5 clusters was obtained from VOSviewer. R packages “bibliometrix” and “ggplot2” were used for quantitative processing, hotspot burst analysis, research prospect discovery, and data visualization.¹¹ GraphPad Prism 9.0 was used for regression curve analysis and plotting. A P-value of less than 0.05 was considered statistically significant.

RESULTS

Overview of Publication Information

The field of conversion therapy for HCC has shown promising progress over the past decade, with a growth rate of 23.0% annually and global collaboration increasing by 13.2% (Table S1, see http://links.lww.com/AOSO/A482). The period from 2014 to 2023 can be divided into 2 stages: a stable development stage and a rapid growth stage. From 2014 to 2020, the annual publication and citation counts for HCC conversion therapy remained below 50 and 1000, respectively (Fig. 1A). Subsequently, both metrics exceeded these values and exhibited rapid growth. By 2023, the annual publication and citation counts reached 116 and 1943, respectively (Fig. 1B). Over the past decade, the predictive function for annual publications was y = 10.0x − 20189 (R² = 0.789, P = 0.0006), and for annual citations, it was y = 197.2x − 397356 (R² = 0.8656, P < 0.0001). The highest global citation (n = 313) and local citation (n = 86) for HCC conversion therapy was the classic paper by Yao FY et al in 2015, titled “Reassessing the boundaries of liver transplantation for hepatocellular carcinoma: Where do we stand with tumor down-staging?” (Fig. 1C and Table S2, see http://links.lww.com/AOSO/A482). The annual citation trends for the top 10 cited classic papers in HCC conversion therapy show an upward trajectory (Fig. 1D).

FIGURE 1.

Visual overview of bibliometric information in the field of conversion therapy for HCC. (A) The annual publication volume shows an increasing trend. (B) The annual citation volume shows a year-by-year increasing trend. (C) The top 10 global and local citations in the field. (D) Z-score normalized citations of the top 10 cited papers, showing an increasing trend over the years.

Prolific Scholars

Mehta N is regarded as the most comprehensive scholar, ranking first in all bibliometric indicators and achieving the highest cumulative publication (n = 13) and citation (n = 149) counts by 2023 (Figs. 2A, B). Along with Mehta N, Zhou J, Fan J, Lu SC, Sun HC, Tabrizian P, Zhao HT, Chen CL, and Li B also published 2 or more papers in 2023 (Fig. 2C). Of 2944 scholars, 465 remained after filtering for those with at least 2 appearances, and 244 were identified for optimal display. Hierarchical clustering analysis grouped these 244 authors into 17 clusters based on their connections (Fig. 2D and Table S3, see http://links.lww.com/AOSO/A482). The timeline distribution of author clusters indicates that authors from Western countries appeared earlier and sustained longer development, whereas Chinese authors emerged later (Fig. 2E). The publication density among authors is visualized in Figure 2F, and the citation density among authors is shown in Figure 2G. Results highlight the outstanding academic performance of Mehta N and other scholars, suggesting they should be followed in future studies for insights into frontier developments in this field.

FIGURE 2.

Bibliometric survey of prolific scholars in conversion therapy for HCC. (A) Total publication volume of prolific scholars. (B) Total citation volume of prolific scholars. (C) The annual publication volume of prolific scholars from 2014 to 2023. (D) Hierarchical clustering analysis categorizing authors based on their connections. (E) Timeline distribution of author clusters. Purple nodes represent earlier publications, whereas yellow nodes represent more recent publications. (F) Visualization of publication density among authors. (G) Visualization of citation density among authors.

Spatiotemporal Distribution and Interaction of Prolific Countries and Institutions

China (685), the USA (429), and Italy (127) are the leading countries in terms of publication volume (Figs. 3A, C, E). In addition, the USA (2118), China (1611), and Italy (702) lead in citation counts (Figs. 3B, D). For the top 10 countries, the publication volume in HCC conversion therapy has shown an upward trend. After filtering institutions with at least 4 appearances, 53 of 712 institutions remained, and 23 were identified for optimal display (Fig. 3F and Table S4, see http://links.lww.com/AOSO/A482). The average publication year of institutions is shown in Figure 3G. The visualization of publication density and citation density among institutions is depicted in Figures 3H, I, with detailed information available in Table S4, see http://links.lww.com/AOSO/A482.

FIGURE 3.

Spatiotemporal distribution and interaction of prolific countries and institutions in conversion therapy for HCC. (A) Total publication volume of prolific countries. (B) Total citation volume of prolific countries. (C) Top 10 countries by publication volume. (D) Top 10 countries by total citation volume. (E) Annual publication volume of prolific countries from 2014 to 2023. (F) Spatial interaction of prolific institutions based on hierarchical clustering analysis. (G) Timeline distribution of institutions. Purple nodes represent earlier publications, whereas yellow nodes represent more recent publications. (H) Visualization of publication density among institutions. (I) Visualization of citation density among institutions.

Spatiotemporal Network of Research Characteristics

Unsupervised learning categorized research characteristics into 5 clusters: Cluster 1 (Internal Radiation Therapy), Cluster 2 (Target Therapy and Immunotherapy), Cluster 3 (Treatment Outcome), Cluster 4 (Local-regional Therapy), and Cluster 5 (Radiotherapy) (Fig. 4A and Table S5, see http://links.lww.com/AOSO/A482). The timeline distribution of research characteristics indicates that Cluster 2 (Target Therapy and Immunotherapy) is an emerging research cluster, with an average publication year of 2021.47 ± 1.04 (Fig. 4B). The core nodes of Clusters 1 to 5 are radioembolization (total link strength [TLS] = 391, OF = 42), sorafenib (TLS = 671, OF = 84), survival (TLS = 710, OF = 92), transarterial chemoembolization (TACE) (TLS = 1,124, OF = 130), and radiation therapy (TLS = 182, OF = 21), respectively (Figs. 4C, D).

FIGURE 4.

Spatiotemporal network of research characteristics in the field of conversion therapy for HCC. (A) Unsupervised hierarchical clustering of research characteristics, resulting in 5 clusters. (B) Temporal distribution pattern of research characteristics from 2014 to 2023. (C) Spatial density network of research characteristics based on link frequency. (D) Spatial density network of research characteristics based on occurrence frequency.

Burst, Temporal Evolution, and Regression Analysis of Research Characteristics

Thirteen research characteristics exhibited burst intensities of 2 or more (Fig. 5A). The top 3 are “Transarterial chemoembolization” (burst intensity = 34), “Downstaging therapy” (burst intensity = 54), and “Locoregional therapy” (burst intensity = 27), indicating their sustained high burst intensity over the past decade. These 13 characteristics are categorized into 4 main types, all showing significant upward trends. Category 1 includes “conversion therapy” (s = 3.061, R² = 0.653, P = 0.005) and “downstaging therapy” (s = 1.133, R² = 0.588, P = 0.010). Category 2 includes “local-regional therapy” (s = 0.830, R² = 0.587, P = 0.010) and “transarterial chemoembolization” (s = 1.370, R² = 0.709, P = 0.002). Category 3 includes “target therapy” (s = 0.170, R² = 0.540, P = 0.016), “tyrosine kinase inhibitor” (s = 0.236, R² = 0.542, P = 0.015), and “lenvatinib” (s = 0.891, R² = 0.720, P = 0.002). Category 4 includes “tumor microenvironment” (s = 0.188, R² = 0.448, P = 0.034), “immunotherapy” (s = 0.970, R² = 0.773, P = 0.001), “immune checkpoint inhibitors” (s = 0.636, R² = 0.689, P = 0.003), “PD-1 inhibitors” (s = 0.158, R² = 0.466, P = 0.030), “atezolizumab” (s = 0.352, R² = 0.432, P = 0.039), and “camrelizumab” (s = 0.236, R² = 0.542, P = 0.015).

FIGURE 5.

Burst status, temporal evolution, and regression curve of research characteristics in conversion therapy for HCC. (A) Burst status and temporal evolution of research characteristics. (B) Regression analysis of annual occurrence frequency of research characteristics. Results show an increasing trend in conversion therapy (a), local therapy (b), targeted therapy (c), and immunotherapy (d).

Potential Analysis of Research Characteristics

Based on relevance degree and development degree, all research characteristics were divided into 4 quadrants (Fig. 6). “immune checkpoint inhibitor” (relevance degree = 16, development degree = 17) and “combination therapy” (relevance degree = 18, development degree = 16) were identified as highly relevant and rapidly developing themes in the motor themes community. In the basic themes community, “hepatic arterial infusion chemotherapy” (HAIC) (relevance degree = 21, development degree = 3) and “radioembolization” (relevance degree = 25, development degree = 8) were identified as broadly relevant but underdeveloped themes. Clearly, the research characteristics in Quadrant IV are highly relevant to the field but remain underdeveloped, indicating their significant potential and worth for further exploration.

FIGURE 6.

Potential analysis of research characteristics in conversion therapy for HCC. Potential findings of research characteristics categorized into 4 quadrants: Quadrant I (motor themes, representing highly relevant and rapidly developing themes), Quadrant II (niche themes, representing relatively small but particularly important areas), Quadrant III (emerging or declining themes, representing emerging or declining topics), and Quadrant IV (basic themes, representing broadly relevant but underdeveloped themes). Notably, Quadrant IV (lower right) represents research themes that are significant but underdeveloped, indicating high potential for future research.

DISCUSSION

To our knowledge, this cross-sectional study is the first to use machine learning-based algorithms to analyze and explore the global scientific landscape of HCC conversion therapy. This includes basic bibliometric data, intrinsic relationships among studies, evolutionary processes, and interconnections of key research elements (cited classics, prolific scholars, leading countries, institutions, and research characteristics). Unsupervised learning grouped all research characteristics into several clusters, demonstrating their temporal and spatial distributions and key nodes. In addition, the study revealed the burst status and temporal evolution of the top 13 research characteristics with the highest burst intensity, providing significant statistical confirmation through regression analysis. More interestingly, community maps generated by the Walktrap algorithm identified several research directions closely related to the field but requiring further exploration.

The field of conversion therapy for HCC entered a rapid growth stage post-2020 after a stable development phase from 2014 to 2020. During this growth phase, scientists increasingly recognized and applied immunotherapy in conversion therapy.¹² The most effective strategy in conversion therapy for tumors is to maximize the objective response rate (ORR) (RECIST 1.1 criteria) while ensuring liver function. Previous studies have shown that single local treatments or monotherapies with targeted or immunotherapy provide limited clinical benefits, with ORRs generally not exceeding 20%.¹³ Therefore, combination therapies are needed to improve ORR further. Notably, breakthrough progress has been made in studies combining antiangiogenic drugs with immune checkpoint inhibitors, which enhance host immune activity and tumor immunogenicity, synergistically exerting antitumor effects and significantly improving ORR compared to monotherapy.¹⁴ Since 2020, several clinical studies have reported impressive results: the IMbrave150 study showed an ORR of 27.3% with the “atezolizumab plus bevacizumab” regimen,¹⁵ the Keynote524 study reported an ORR of 36% with the “lenvatinib plus pembrolizumab” regimen,¹⁶ and China’s “sintilimab plus a bevacizumab biosimilar”¹⁷ and “camrelizumab plus apatinib”¹⁸ combinations achieved ORRs of 20.5% and 34%, respectively. In addition, dual immunotherapy combinations have also shown high ORRs, with the Checkmate040 study’s “nivolumab plus ipilimumab” regimen¹⁹ and the Himalaya study’s “tremelimumab plus durvalumab” regimen²⁰ achieving ORRs of 32% and 24%, respectively. These data demonstrate the superiority of systemic combination therapy in HCC treatment, particularly in ORR, making it an indispensable approach in conversion therapy.

Although cluster analysis, temporal evolution, regression curves, and potential findings highlight the researchers’ intense exploration of immune checkpoint inhibitors in conversion therapy, 3 key points warrant attention. First, current clinical practice shows that these inhibitors do not produce absolute efficacy in all patients. The tumor microenvironment or cancer ecosystem is highly complex and diverse, involving interactions between cancer cells and the host immune system, among cancer cells, and between cancer cells and other tissues.²¹ This complexity leads to varied responses among patients to current immune checkpoint inhibitors. There is an urgent need to explore potential mechanisms of immune checkpoints, identify new targets, and investigate combinations of different targets to enhance efficacy and overcome resistance.²² Second, safety is a major concern. Immune activation can cause immune-inflammatory reactions, autoimmune responses, and other adverse events. Balancing efficacy and safety is crucial for patients and a significant focus of current clinical research. Third, one of the main challenges of immunotherapy is the lack of biomarkers to predict the best responders. Traditional biomarkers are insufficient to identify the optimal treatment population. Recent studies suggest that combinations of multiple biomarkers are more effective in predicting patient prognosis than single biomarker. Utilizing various omics data in clinical practice (eg, radiomics²³ and genomics²⁴) to identify biomarkers or construct predictive models to forecast patient response to immunotherapy or toxicity is an important direction scientists are pursuing for stratified treatment.

In addition to immune checkpoint inhibitors, local treatments have been extensively explored in conversion therapy. TACE was initially used in clinical conversion therapy but has limitations, such as low conversion success rates and the risk of liver function deterioration and postoperative liver failure with repeated TACE.²⁵ The Walktrap algorithm and research frontier discovery indicate that HAIC holds significant importance but remains underdeveloped. HAIC partially compensates for TACE’s limitations in conversion therapy, with several prospective studies showing high ORR, especially for high tumor burden cases where TACE is less effective and causes significant liver damage.^26,27 HAIC can rapidly reduce tumor volume and significantly retract cancer thrombus, improving conversion success rates.²⁸ However, initiating and maintaining HAIC programs in Western countries faces logistical and feasibility challenges, including the lack of multidisciplinary care, access to equipment/chemotherapy drugs, and other logistical issues.²⁹ Developing a high-quality HAIC framework is crucial for optimizing patient safety and minimizing toxicity.³⁰ Currently, various local therapies combined with systemic treatments are available, but there is no consensus on the best combination strategy. Prospective clinical studies are needed to verify the optimal approaches.

In addition, increasing evidence indicates that patients with HCC who initially do not meet transplant eligibility criteria but undergo successful downstaging to qualify for transplantation can achieve survival outcomes comparable with those of patients who were initially within the transplant criteria. For instance, patients with HCC beyond the Milan criteria, who are downstaged using therapies such as TACE, radiofrequency ablation, radiotherapy, or radioembolization to meet Milan criteria, can exhibit survival rates similar to those of patients initially meeting the Milan criteria. Specifically, 1 study reported 1-, 3-, and 5-year overall survival rates of 91.4%, 82.8%, and 70.7%, respectively, for patients downstaged to the Milan criteria, which closely mirrored the survival rates for patients who were initially within the Milan criteria (92.0%, 85.7%, and 74.1%, respectively, P = 0.540).³¹ Furthermore, patients who were downstaged to meet the University of California, San Francisco criteria showed similar recurrence rates at 5 years (10.9% vs 10.8%, P = 0.84) when compared with those initially meeting University of California, San Francisco criteria for liver transplantation.³²

This study has some limitations. First, it only included data from the Web of Science Core Collection, so key data from other databases may be missing. This dataset was chosen for its comprehensiveness and authority, with its data reflecting the progress in conversion therapy research. Future studies could integrate other databases, though one should consider duplicate data and confounding factors. Second, earlier studies have more time to be cited compared with recently published ones, so the pioneering advantage should be considered when evaluating research and author factors. Third, due to space limitations, this paper may omit some potentially important information, warranting further exploration in future research.

CONCLUSION

In conclusion, this study is the first to analyze the global scientific landscape of conversion therapy for HCC. Over the past decade, the development and global collaboration in this field have been promising. To achieve effective conversion therapy for HCC, researchers may prioritize immunotherapy and locoregional treatments.

ACKNOWLEDGMENTS

The authors thank Sipeng Shen, a professor from the Department of Biostatistics of Nanjing Medical University, for his review concerning the statistical methods of this study.