Blog

Introduction

Falls represent one of the most significant health risks for the elderly population worldwide, often resulting in severe injuries, hospitalization, and even fatalities.¹ According to global health statistics, approximately 30% of people aged 65 and older experience at least one fall annually, with this percentage increasing to 50% for those over 80 years of age.² The consequences of falls extend beyond physical injuries to include psychological impacts such as fear of falling, reduced mobility, social isolation, and diminished quality of life.¹ With the global aging population projected to reach 2.1 billion by 2050, addressing fall-related health challenges has become an urgent public health priority requiring innovative technological interventions.^2,3

The emergence of the Internet of Health Things (IoHT) has created unprecedented opportunities for real-time monitoring and early intervention in elderly healthcare.⁴ While the Internet of Things (IoT) broadly encompasses interconnected devices across various domains, IoHT represents a specialized branch focused specifically on healthcare applications, encompassing medical sensors, wearable devices, and systems that collect, analyze, and transmit health data.⁵ Within this domain, embedded systems—which integrate computing capabilities within physical devices using microcontrollers, sensors, and actuators—have shown remarkable potential for developing effective fall detection solutions. These embedded IoHT systems enable continuous monitoring without human intervention, allowing for immediate alert generation when falls are detected.³

Fall detection research using embedded IoHT has evolved substantially over the past decade, progressing from simple threshold-based approaches to sophisticated machine learning and deep learning methodologies.¹ Early systems primarily relied on wearable accelerometers with predefined threshold values to identify falls but suffered from high false alarm rates and limited sensitivity.² Contemporary research increasingly leverages the fusion of multiple sensor types (motion, physiological, and environmental) with advanced algorithms to achieve more accurate detection while addressing challenges related to privacy, power consumption, and user acceptance.^3,4,6

Despite the growing body of literature in this field, there has been no comprehensive bibliometric analysis examining the research landscape of embedded IoHT fall detection systems for elderly care. Bibliometric analysis represents a systematic and quantitative method for exploring large volumes of scientific data, offering insights into research trends, influential works, collaboration networks, and emerging topics.⁷ Such analysis is invaluable for researchers seeking to understand the intellectual structure and evolution of this interdisciplinary field that spans healthcare, computer science, engineering, and gerontology.

The purpose of this study is to conduct a thorough bibliometric analysis of research articles focusing on embedded IoHT fall detection systems for the elderly. By examining publication patterns, citation structures, author collaborations, and thematic developments, we aim to map the intellectual landscape of this field, identify research hotspots, and highlight emerging trends.⁷ This study employs established bibliometric techniques including performance analysis and science mapping using specialized tools such as VOSviewer.

Our analysis specifically addresses four key research questions:¹ How has research output in embedded IoHT fall detection evolved over time in terms of volume and impact?² Which countries, institutions, and authors have been most productive and influential in this field?³ What are the dominant research themes and how have they shifted over time?⁴ What methodological approaches are most prevalent and how are they evolving?

Methods

Data Source and Search Strategy

We conducted a comprehensive literature search using the Scopus database (Elsevier). The Scopus database was selected for this study due to its comprehensive coverage of peer-reviewed scientific literature across multiple disciplines and its compatibility with bibliometric analysis tools such as VOSviewer. The analysis covers the period from January 1, 2006, to April 24, 2025; therefore, publication data for the year 2025 should be considered provisional, as it reflects only partial-year results and may not represent the full research output for that year. The search strategy with keyword (“embedded” OR “integrated” OR “inbuilt”) AND (“internet” OR “web” OR “network”) AND (“health” OR “wellness” OR “medical”) AND (“things” OR “devices” OR “objects”) AND (“fall” OR “collapse” OR “trip”) AND (“detection” OR “monitoring” OR “identification”) AND (“elderly” OR “aged” OR “senior” OR “geriatric”). These keywords were applied to titles, abstracts, and author keywords using Boolean operators. No language restrictions were initially applied.

Data Extraction and Coding

Bibliographic data were extracted for each eligible publication, including title, authors, year of publication, journal, and the first author’s country of affiliation. In addition, bibliometric indicators such as citation counts and author keywords were recorded. Data extraction was performed independently by two reviewers using a piloted extraction sheet to ensure accuracy and consistency. Any discrepancies were resolved through discussion.

Bibliometric and Network Analysis

Bibliometric analyses were conducted to examine annual publication trends, including the number of articles published per year and the overall growth trajectory. We identified the most productive authors, institutions, countries, and journals, evaluating both publication volume and citation impact. Collaboration patterns, including co-authorship and international collaborations, were analyzed through network maps generated using VOSviewer (version 1.6.18). Keyword co-occurrence analysis was also performed to detect thematic clusters and emerging areas of research interest.

Statistical Analysis

Descriptive statistics, including frequencies, percentages, means, and standard deviations (SD), were calculated using Microsoft Excel 2021. These metrics were used to quantitatively summarize publication characteristics and bibliometric indicators.

Visualization

Visual representations of the findings were generated to facilitate interpretation. Co-authorship networks and keyword co-occurrence maps were created with VOSviewer, using a minimum threshold of five occurrences per term. Temporal trends and geographic distributions were visualized through trend charts and heatmaps produced using Microsoft Excel 2021.

Results

A total of 79 documents were identified. Conference papers represented the majority of records (n = 44; 55.7%), followed by journal articles (n = 31; 39.2%), book chapters (n = 2; 2.5%), and review papers (n = 2; 2.5%). Nearly all publications were in English (n = 78; 98.7%), with one document in Chinese (n = 1; 1.3%). By source type, 36 documents (45.6%) appeared in conference proceedings, 32 (40.5%) in peer-reviewed journals, 9 (11.4%) within book series, and 2 (2.5%) as standalone books.

The annual trend of publications on embedded IoHT-based fall detection shows a rapid increase after 2018, peaking in 2024 (Figure 1).

Analysis of the Table 1 top 20 most-cited articles in embedded Internet of Healthcare Things (IoHT) fall detection reveals that, between 2006 and 2014, research was primarily concerned with integrating basic sensors (accelerometers, gyroscopes) into wearable or home-based prototypes and applying simple threshold-based algorithms for SMS or telereported fall alerts. After 2015, there was a marked acceleration in both publication volume and technological maturity: studies began leveraging IoT architectures and cloud computing for real-time monitoring and data storage, and adopted machine-learning models such as decision trees and big-data analytics to improve detection accuracy. The peak citation years of 2018–2019 led by Yacchirema et al (Escuela Politécnica Nacional, Ecuador) and Palau et al (Universitat Politècnica de València, Spain)⁸ highlighted advances in wearable, ML-driven systems that fuse multiple sensors and distribute processing between edge devices and the cloud. Geographically, Spain and Ecuador have produced the most impactful contributions, with notable work also emerging from the United States, South Korea, and China on telemedicine, embedded radar, and AI-enabled platforms. Methodologically in Table 2, the field transitioned from rule-based detection to ensemble machine learning and deep-learning approaches (eg, random forests, RNNs, DCNNs) around 2019–2022, achieving higher accuracy through sensor fusion and vision-based, privacy-preserving techniques. Since 2022, focus has shifted toward secure, AI-enabled cyber-physical systems that emphasize human-centric design, clinical workflow integration, usability studies, and scalability within smart-home environments. Overall, the evolution from simple, data-limited prototypes to sophisticated, adaptive, and secure IoHT solutions reflects broader trends in digital health and AI for aging populations.

Table 1 Top 20 Articles with the Highest Total Citation Scores

Table 2 Methodological Approaches and Their Evolution

The analysis of document types indicates that conference papers dominate the field, followed by journal articles (Figure 2).

Figure 2 The distribution of document types from 2006 to 2025. Notes: Figure 2 the distribution of document types retrieved from Scopus for our fall-detection IoHT corpus. Conference papers constitute the majority of outputs (55.7%), reflecting the community’s preference for rapid dissemination of preliminary results and emerging prototypes in highly iterative engineering and computing venues. Journal articles follow at 39.2%, indicating that a substantial portion of work has undergone more extensive peer review and formal presentation of mature systems or comprehensive evaluations. Book chapters and review articles each account for only 2.5% of the total, suggesting that integrative syntheses and theoretical expositions remain relatively scarce in this domain. Together, these proportions underscore a research landscape driven by fast-moving, implementation-oriented studies, with fewer dedicated efforts toward retrospective analysis or broad, conceptual overviews.

In terms of subject areas, Computer Science and Engineering account for more than half of publications (Figure 3).

Figure 3 The Document by subject from 2006 to 2025. Notes: Figure 3 presents the document according to Scopus subject categories. Computer Science leads with 28.8% of documents, closely followed by Engineering at 25.7%, together accounting for over half of all high-impact outputs. Medicine contributes 8.4% of the papers, reflecting clinical interest in fall-detection technologies, while Physics and Astronomy (6.3%), Decision Sciences (4.7%), and Mathematics (4.7%) each play a more modest role, indicative of analytical and modeling efforts underpinning algorithm development. Smaller slices in Biochemistry, Genetics & Molecular Biology (3.7%), Health Professions (3.1%), Materials Science (2.6%), and Chemical Engineering (2.1%) highlight the involvement of sensor materials, molecular‐level biosensing concepts, and applied healthcare expertise. The “Other” category (9.9%) captures remaining interdisciplinary areas, underscoring that while fall-detection research is grounded primarily in computing and engineering, it nevertheless draws on a broad array of fields to address sensor design, data analysis, and clinical integration.

India, China, and the United States are the top contributors, highlighting their leadership in this domain (Figure 4).

Figure 4 The Document by country or territory from 2006 to 2025. Notes: Figure 4 illustrates the distribution of documents by country or territory based on Scopus data. India leads with the highest number of publications, followed closely by China and the United States, each contributing a significant volume of research outputs. These three countries collectively dominate the publication landscape, indicating their strong research engagement in the relevant field. Italy, Spain, Germany, Taiwan, and the United Kingdom follow, each producing a moderate number of documents. Brazil and Ecuador contribute fewer publications compared to other countries, but their presence still reflects a growing global interest. Overall, the data suggests that research activities are concentrated in a few leading countries, particularly in Asia, North America, and Europe, emphasizing their pivotal role in advancing the scholarly discourse in this domain.

The distribution of top-cited authors shows a collaborative rather than single-author dominance (Figure 5).

Figure 5 The Document by author from 2006 to 2025. Notes: Figure 5 shows the distribution of top‐cited fall‐detection publications by author. Nine researchers—Ashwin T.S., Esteve M., Hsu K., Meghana N.P., Palau C., Rachakonda L., Rakhecha S., Yacchirema D., and Yousuff S.—each appear as co‐authors on two of the twenty most‐cited papers, making them the most prolific contributors in this elite group. In contrast, Abd Elmalek A.H. is represented by a single article. The fact that no individual author exceeds two high‐impact publications suggests that progress in embedded IoHT fall detection is driven by a distributed network of collaborators rather than by any single dominant figure.

High-impact outputs are distributed across diverse institutions worldwide (Figure 6).

Figure 6 The Document by affiliation from 2006 to 2025. Notes: Figure 6 shows that ten different institutions, with no single affiliation contributing more than this. These institutions span North America (Rochester Institute of Technology; University of North Carolina Wilmington; Kate Gleason College of Engineering), Europe (University Politehnica of Bucharest; Universitat Politècnica de València; Escuela Politécnica Nacional), and Asia (National Institute of Technology Karnataka; National University of Singapore; SSN College of Engineering, Kalavakkam; Chinese Academy of Sciences). The even distribution of high-impact outputs across a diverse set of technical universities underscores the field’s collaborative and multi-regional character, suggesting that advances in fall-detection IoHT are driven by a broad network of engineering and computer-science research centres rather than being concentrated within a single “hub.”.

Keyword co-occurrence mapping reveals four main clusters, including ambient intelligence and networked wearables (Figure 7).

Figure 7 VOSviewer Keyword Co-Occurrence Network. Notes: The VOSviewer‐derived keyword co‐occurrence network (Figure 7) positions “fall detection” at its core, reflected by its disproportionately large node and dense links to “accelerometers”, “elderly care”, and “prevention and control”, underscoring the field’s primary focus on sensor‐based geriatric monitoring and risk mitigation. Four distinct thematic clusters emerge: an ambient intelligence cluster (green) highlighting “smart homes”, “continuous monitoring”, and “device‐free” approaches for unobtrusive surveillance; an inertial sensor technology cluster (red) centered on hardware and signal‐processing terms such as “angular orientation” and “accelerometer design”; a demographic influences cluster (light blue) emphasizing “female”, “body mass”, and “elderly care”, which reflects growing attention to how individual characteristics affect fall risk and detection accuracy; and a networked wearables cluster (purple) featuring “sensor networks”, “mobile devices”, and “digital healthcare”, indicative of efforts to integrate wearable systems with telemedicine infrastructures. Notably, machine‐learning methods (eg, “k‐means++” and “data‐driven decision support”) and data privacy/security terms occupy peripheral positions, suggesting that advanced real‐time decision‐support frameworks and privacy‐preserving mechanisms remain underexplored. Moreover, the absence of standardized dataset and evaluation protocol nodes highlights an urgent need for community consensus on shared benchmarks. Future research should therefore prioritize the integration of robust ML‐driven decision support into real‐time monitoring architectures, the development of device‐free, privacy‐centric solutions, and the establishment of unified datasets and evaluation standards, while tailoring algorithms to demographic and health profiles to enhance both sensitivity and clinical utility.

The overlay visualization shows the chronological evolution from hardware prototyping to AI-driven systems (Figure 8).

Figure 8 VOSviewer The overlay map. Notes: The overlay map generated by VOSviewer (Figure 8) reveals a clear temporal progression in fall-detection research. Early themes (circa 2010–2012), shown in dark blue, focus on infrastructure-level topics such as “sensor networks”, “mobile devices”, and “hardware”, reflecting foundational work on connectivity and device design. Between 2014 and 2018, labeled in green to yellow-green hues, the focus shifts toward applied sensor-based monitoring—most notably “fall detection”, “accelerometers”, and “health care”—marking the field’s most prolific period of inertia-sensor innovations within clinical contexts. From approximately 2020 onward, terms appearing in bright yellow signal emerging frontiers: “device-free” and “smart homes” indicate a move toward ambient intelligence and unobtrusive environments, while “k-means++”, “data-driven decision support”, and “aspect ratio” point to the integration of advanced machine-learning algorithms and computer-vision techniques. This evolution underscores how the discipline has matured from hardware and network prototyping to sophisticated, privacy-conscious, real-time analytical systems.

The density visualization highlights fall detection, accelerometers, and health care as the most central themes (Figure 9).

Figure 9 Density Visualization of Keyword Co-Occurrence. Notes: The density visualization of keyword co-occurrence (Figure 9) underscores three core research foci in fall-detection literature: “fall detection”, “accelerometers”, and “health care”, each surrounded by bright yellow regions indicating the highest keyword frequency and co-occurrence strength, reflecting the central role of inertial sensors in clinical fall-monitoring studies. To the right, a secondary hotspot around “female”, “body mass”, and “elderly care” indicates substantial attention to demographic subgroups—particularly older women with varying anthropometric profiles. In contrast, topics such as “smart homes” and “continuous monitoring” appear in slightly cooler yellow-green hues, suggesting that ambient-intelligence applications and continuous surveillance systems are emerging but not yet as dominant as wearable-sensor research. Darker green areas around “mobile devices”, “sensor networks”, and “prevention and control” reveal significant yet peripheral thematic clusters. Machine-learning and decision-support terms like “k-means++” and “data-driven decision support” occupy blue zones of low density, highlighting their underrepresentation in current studies. Notably, privacy and data-security concepts are virtually absent, pointing to a critical gap and an opportunity for future work on privacy-preserving, intelligent fall-detection frameworks.

Discussion

Evolution of Research Output: Volume and Impact

The field of embedded Internet of Health Things (IoHT) fall detection for the elderly has experienced significant growth and transformation over the past two decades. Early research (2006–2012) was sporadic, with annual publication counts rarely exceeding three documents. This period was characterized by foundational work on integrating basic sensors (accelerometers, gyroscopes) into wearable or home-based prototypes, primarily utilizing simple threshold-based algorithms for fall alerts. These early systems, while innovative, suffered from high false alarm rates and limited sensitivity.

A marked acceleration began in 2018, with publication counts rising sharply and peaking at twelve documents in 2024. This surge reflects both increased research interest and technological maturation in the field. The rapid growth over the past five years underscores the urgency and relevance of fall detection as the global population ages and the prevalence of fall-related injuries rises.

Citation analysis of the top 20 most-cited articles reveals a parallel evolution in impact: highly cited works from 2018–2019, such as those by Yacchirema et al and Palau et al, introduced wearable, machine learning-driven systems that leveraged sensor fusion and distributed processing between edge devices and the cloud. These studies achieved high accuracy (often above 94%) and demonstrated the feasibility of real-time, IoHT-based fall detection with direct alerts to caregivers. The transition from simple, rule-based prototypes to sophisticated, adaptive, and secure IoHT solutions reflects broader trends in digital health and artificial intelligence for aging populations.

Geographic, Institutional, and Author Contributions

The research landscape is dominated by a few leading countries-India, China, and the United States-each making substantial contributions in terms of publication volume. These countries are followed by Italy, Spain, Germany, Taiwan, and the United Kingdom, reflecting a strong global interest, particularly in Asia, North America, and Europe. Notably, Spain and Ecuador have produced some of the most impactful (highly cited) contributions, especially in the development and clinical evaluation of wearable and IoT-based systems.

Institutional analysis shows that high-impact research is distributed across a diverse set of technical universities and research centers, with no single institution dominating the field. This suggests a collaborative, multi-regional character, with progress driven by networks of engineering and computer science centers rather than isolated “hubs”.

At the author level, the most prolific contributors to the top-cited literature-such as Ashwin T.S., Esteve M., Hsu K., Meghana N.P., Palau C., Rachakonda L., Rakhecha S., Yacchirema D., and Yousuff S.-each appear as co-authors on two of the twenty most-cited papers. This indicates a distributed network of collaboration rather than dominance by any single figure.

Dominant Research Themes and Thematic Shifts

Keyword co-occurrence analysis identified four major thematic clusters in the field: ambient intelligence, emphasizing smart homes, continuous monitoring, and device-free surveillance; inertial sensor technology, focusing on hardware and signal-processing aspects such as angular orientation and accelerometer design; demographic influences, highlighting personalized approaches to fall risk and detection based on factors like gender, body mass, and elderly care; and networked wearables, centering on the integration of sensor networks, mobile devices, and digital health infrastructures. The thematic evolution over time demonstrates a clear progression, with early research (2010–2012) concentrating on infrastructure-level topics such as sensor networks and mobile devices, shifting between 2014–2018 toward applied sensor-based monitoring and healthcare applications, and, from 2020 onward, expanding to ambient intelligence, advanced machine learning techniques (eg, k-means++), and privacy-preserving strategies. Nonetheless, terms related to machine learning and privacy/security remain relatively underrepresented, revealing significant gaps and opportunities for future research. Furthermore, the absence of standardized datasets and evaluation protocols underscores the urgent need for greater consensus on benchmarks and methodological frameworks within the research community.

Methodological Approaches and Their Evolution

The methodological landscape of embedded IoHT fall detection systems has evolved markedly over time. During the early period (2006–2014), research was largely dominated by threshold-based algorithms and simple sensor integrations, often paired with SMS or telemedicine alert systems. Between 2015 and 2019, the field saw the introduction of IoT architectures, cloud computing, and machine learning models such as decision trees, random forests, and ensemble learning to enhance detection accuracy and enable real-time monitoring. From 2019 to 2022, deep learning methods, including recurrent neural networks (RNNs) and deep convolutional neural networks (DCNNs), became widespread, accompanied by advancements in sensor fusion and the emergence of vision-based and privacy-preserving approaches. Most recently, the 2022–2025 period emphasizes the development of secure, AI-enabled cyber-physical systems, human-centric designs, clinical workflow integration, and usability studies, with growing attention to scalability, interoperability, and smart-home integration. Experimental studies consistently report high levels of accuracy, precision, sensitivity, and specificity, often exceeding 94%, particularly in systems utilizing ensemble machine learning and multimodal sensor fusion. Edge computing has also gained prominence for reducing latency and enhancing real-time performance, alongside a growing interest in integrating fall prediction and prevention features. The bibliometric analysis further highlights rapid technological advancement, driven by a globally collaborative research community, and an emerging focus on ambient intelligence, privacy-preserving monitoring, and context-aware machine learning. However, significant unmet needs persist, including the development of standardized datasets, robust evaluation protocols, and comprehensive privacy/security frameworks, as well as the need for algorithms better tailored to diverse demographic and health profiles to ensure greater clinical applicability and user acceptance.

Limitations

This study is limited to the Scopus database without applying any inclusion or exclusion criteria. All document types including journal articles, conference papers, and books were included regardless of research method or language. As a result, the findings may reflect broad trends but lack selective filtering for study quality or relevance.

Conclusion

This bibliometric analysis demonstrates the rapid development of research on embedded IoHT fall detection for the elderly. In line with our research objectives, we observed a significant increase in publication volume particularly after 2018 indicating rising global interest. India, China, and the US lead in output, while Spain and Ecuador contribute highly cited works, highlighting influential authors and institutions. Thematic analysis revealed four major clusters—ambient intelligence, inertial sensors, personalized elderly care, and telemedicine linked wearables showing a shift from hardware centric systems to AI-driven, real time monitoring solutions. Methodologically, the field has progressed from threshold-based models to machine learning, deep learning, and cloud/edge integration. Despite advancements, gaps remain in areas such as standardized datasets, privacy-preserving methods, and inclusive demographic modeling. These findings directly address the research questions posed, offering insight into the evolution, key contributors, dominant themes, and methodological trends in the field. Going forward, innovation must be coupled with clinical integration and user-centered design to realize the full potential of IoHT technologies. Bridging current gaps will be critical to developing scalable, effective fall detection systems for the aging population.

Ethical Approval

Ethical approval was not required as the study did not involve human participants.

Acknowledgments

All authors thank you to Universitas Padjadjaran who has facilitating us to make this study.

Author Contributions

All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Informed Consent

Informed consent was not required as the study did not involve.

Funding

This research and publication were supported by Universitas Padjadjaran.

Disclosure

The authors report no conflicts of interest in this work.

References

1. Wang X, Ellul J, Azzopardi G. Elderly fall detection systems: a literature survey. Front Robot AI. 2020;7. doi:10.3389/frobt.2020.00071

2. Thomas M, Dileep N, Sangeetha P, Bineesh M. Review of fall detection and alert systems for elderly people. International Journal of Engineering Research & Technology (IJERT). 2021; Available from: www.ijert.org. Accessed August 25, 2025.

3. Reyad O, Ibrahim Shehata H, Esmail Karar M. Developed fall detection of elderly patients in internet of healthcare things. Comput Mater Continua. 2023;76(2):1689–1700. doi:10.32604/cmc.2023.039084

4. Dharmansyah D. Literature review: design of internet of Health Things (IoHT) model for fall risk detection in elderly at home. J Nurs Culture Technol. 2024;1(1):30–36. doi:10.70049/jnctech.v1i1.8

5. Huang C, Wang J, Wang S, Zhang Y. Internet of medical things: a systematic review. Neurocomputing. 2023;557(126719):126719. doi:10.1016/j.neucom.2023.126719

6. Ma X, Wang X, Zhang K. Fall detection method for embedded devices. J Imaging Sci Technol. 2022;66(4). doi:10.2352/J.ImagingSci.Technol.2022.66.4.040407

7. Donthu N, Kumar S, Mukherjee D, Pandey N, Lim WM. How to conduct a bibliometric analysis: an overview and guidelines. J Bus Res. 2021;133:285–296. doi:10.1016/j.jbusres.2021.04.070

8. Yacchirema D, De Puga JS, Palau C, Esteve M. Fall detection system for elderly people using IoT and big data. procedia computer science. Elsevier BV. 2018;603–610.

9. Yacchirema D, de Puga JS, Palau C, Esteve M, de Puga JS. Fall detection system for elderly people using IoT and ensemble machine learning algorithm. Pers Ubiquitous Comput. 2019;23(5–6):801–817. doi:10.1007/s00779-018-01196-8

10. Kang JM, Yoo T, Kim HC. A wrist-worn integrated health monitoring instrument with a tele-reporting device for telemedicine and telecare. IEEE Trans Instrum Meas. 2006;55(5):1655–1661. doi:10.1109/TIM.2006.881035

11. Virginia Anikwe C, Friday Nweke H, Chukwu Ikegwu A, et al. Mobile and wearable sensors for data-driven health monitoring system: state-of-the-art and future prospect. Expert Syst Appl. 2022;202:117362. doi:10.1016/j.eswa.2022.117362

12. Garripoli C, Mercuri M, Karsmakers P, et al. Embedded DSP-based telehealth radar system for remote in-door fall detection. IEEE J Biomed Health Inform. 2015;19(1):92–101. doi:10.1109/JBHI.2014.2361252

13. Ramasamy LK, Khan F, Shah M, BVVS P, Iwendi C, Biamba C. Secure smart wearable computing through artificial intelligence-enabled internet of things and cyber-physical systems for health monitoring. Sensors. 2022;22(3):1076. doi:10.3390/s22031076

14. Luna-Perejón F, Domínguez-Morales MJ, Civit-Balcells A. Wearable fall detector using recurrent neural networks. Sensors. 2019;19(22):4885. doi:10.3390/s19224885

15. Kotronis C, Routis I, Politi E, et al. Evaluating Internet of Medical Things (IoMT)-based systems from a human-centric perspective. Int Things. 2019;8.

16. Issa ME, Helm AM, Al-Qaness MAA, Dahou A, Elaziz MA, Damaševičius R. Human activity recognition based on embedded sensor data fusion for the internet of healthcare things. Healthcare. 2022;10(6):1084. doi:10.3390/healthcare10061084

17. Gochoo M, Tan TH, Batjargal T, Seredin O, Huang SC. Device-free non-privacy invasive indoor human posture recognition using low-resolution infrared sensor-based wireless sensor networks and DCNN. In: Proceedings – 2018 IEEE International Conference on Systems, Man, and Cybernetics, SMC 2018. Institute of Electrical and Electronics Engineers Inc.; 2018. p. 2311–2316.

18. Wayan Wiprayoga Wisesa I, Mahardika G. Fall detection algorithm based on accelerometer and gyroscope sensor data using Recurrent Neural Networks. In: IOP Conference Series: Earth and Environmental Science. Institute of Physics Publishing; 2019.

19. Shahiduzzaman KM, Hei X, Guo C, Cheng W. Enhancing fall detection for elderly with smart helmet in a cloud-network-edge architecture. In: 2019 IEEE International Conference on Consumer Electronics – Taiwan, ICCE-TW 2019. Institute of Electrical and Electronics Engineers Inc.; 2019.

20. Chandel V, Sinharay A, Ahmed N, Ghose A. Exploiting IMU sensors for IoT enabled health monitoring. In: IoTofHealth 2016 – Proceedings of the 1st Workshop on IoT-Enabled Healthcare and Wellness Technologies and Systems, co-located with MobiSys 2016 [Internet]. Association for Computing Machinery, Inc; 2016. p. 21–22.

21. Balaguera HU, Wise D, Ng CY, et al. Using a medical intranet of things system to prevent bed falls in an acute care hospital:a pilot study. J Med Internet Res. 2017;19(5):e150. doi:10.2196/jmir.7131

22. Torres GG, Bayan Henriques RV, Pereira CE, Müller I. An EnOcean wearable device with fall detection algorithm integrated with a smart home system. Elsevier BV. 2018;9–14.

23. Zhang Y, Zheng X, Liang W, Zhang S, Yuan X. Visual surveillance for human fall detection in healthcare IoT. IEEE Multimedia. 2022;29(1):36–46. doi:10.1109/MMUL.2022.3155768

24. Kolakowski J, Djaja-Josko V, Kolakowski M. UWB monitoring system for AAL applications. Sensors. 2017;17(9):2092. doi:10.3390/s17092092

25. Cai WY, Guo JH, Zhang MY, Ruan ZX, Zheng XC, Lv SS. GBDT-based fall detection with comprehensive data from posture sensor and human skeleton extraction. J Healthc Eng. 2020;2020:1–15. doi:10.1155/2020/8887340

26. Joshi NB, Nalbalwar SL. A fall detection and alert system for an elderly using computer vision and Internet of Things. In: RTEICT 2017 – 2nd IEEE International Conference on Recent Trends in Electronics, Information and Communication Technology, Proceedings [Internet]. Institute of Electrical and Electronics Engineers Inc.; 2017. p. 1276–1281.

27. Losardo A, Bianchi V, Grossi F, Matrella G, De Munari I, Ciampolini P. Web-enabled home assistive tools. Assist Technol Res Ser. 2011;29:448–455.

Introduction

According to 2022 statistics, liver cancer ranks as the sixth most common malignancy and the third leading cause of cancer-related death globally.¹ Hepatocellular carcinoma (HCC) constitutes the majority of liver cancer cases. Due to the absence of distinct early-stage symptoms and inadequate surveillance in high-risk populations, most HCC cases are diagnosed at an advanced stage, resulting in a poor prognosis—particularly in the presence of extrahepatic metastases. According to the Barcelona Clinic Liver Cancer (BCLC) staging system, advanced HCC (BCLC stage C) is defined by vascular invasion and/or extrahepatic spread.² The lungs are the most frequent site of extrahepatic metastasis and are associated with especially poor survival outcomes. While treatment of advanced HCC with vascular invasion has seen steady progress, optimal strategies for managing extrahepatic metastases, especially lung involvement, remain uncertain.

The presence of lung metastasis signifies advanced-stage HCC, for which systemic therapy is the standard of care. Current systemic options include tyrosine kinase inhibitors (TKIs), anti-angiogenic agents, and immune checkpoint inhibitors (ICIs). Sorafenib, the first approved TKI for advanced HCC, demonstrated a median overall survival (OS) of 10.7 months in the SHARP trial. However, subgroup analysis revealed that sorafenib conferred no significant benefit over placebo in patients with extrahepatic metastases.³ In 2018, lenvatinib was shown to be non-inferior to sorafenib in OS and superior in terms of objective response rate (ORR) among untreated advanced HCC patients, although data specific to those with extrahepatic spread were limited.⁴ In 2020, the IMbrave150 study established the superiority of atezolizumab (a PD-L1 inhibitor) combined with bevacizumab (an anti-angiogenic agent) over sorafenib, including in patients with extrahepatic disease.⁵ Tislelizumab (a PD-1 inhibitor) has also demonstrated improved OS, progression-free survival (PFS), and ORR compared to sorafenib as a first-line treatment for BCLC stage B–C HCC in the RATIONALE-301 study.⁶ Additionally, various clinical studies have indicated that TKIs combined with PD-1 inhibitors may further enhance outcomes in advanced HCC. However, the LEAP-002 trial showed that lenvatinib plus pembrolizumab did not significantly improve OS or PFS compared with lenvatinib alone in advanced HCC patients, including those with extrahepatic metastases,⁷ suggesting that further optimization of combination regimens is warranted. Therefore, the latest guidelines emphasize a multidisciplinary approach for patients with advanced HCC, especially those with pulmonary metastases, where systemic therapies such as TKIs, ICIs, and combination regimens are increasingly being recommended. Specifically, as recommended in the 2025 EASL guideline, for patients with advanced HCC, including those with pulmonary metastasis, systemic combination therapy including at least one PD-1 or PD-L1 inhibitor should be offered,⁸ consistent with the recommendation in the 2023 ASALD guideline.⁹

Although systemic therapy is the mainstay for advanced HCC with extrahepatic spread, intrahepatic tumor burden is the leading cause of death, as most patients succumb to progression of intrahepatic lesions rather than metastatic complications. Locoregional therapies such as transarterial chemoembolization (TACE) and hepatic arterial infusion chemotherapy (HAIC) are commonly employed to control intrahepatic disease. TACE is widely used and recommended as the standard treatment for intermediate-stage HCC.² HAIC, particularly the FOLFOX regimen (oxaliplatin, fluorouracil, and leucovorin) developed in China, has shown efficacy in large, unresectable tumors.¹⁰ Compared to TACE, HAIC has demonstrated improved OS and fewer severe adverse events in patients with unresectable HCC. A study showed that combining HAIC with sorafenib prolonged median OS by approximately 11 months compared to sorafenib alone in patients with lung metastases.¹¹ More recently, a Phase II clinical trial revealed that apatinib plus HAIC achieved a median OS of 11.3 months as a second-line treatment in advanced HCC with extrahepatic metastasis.¹² Another study suggested that HAIC might potentiate the efficacy of lenvatinib combined with PD-1 inhibitor in HCC patients with extrahepatic metastases; however, the metastatic sites varied.¹³ Given the heterogeneity of metastatic patterns, it remains unclear whether adding HAIC to lenvatinib and PD-1 inhibitor offers survival benefits over dual therapy in patients specifically with lung metastases.

Therefore, this study aimed to evaluate the efficacy and safety of HAIC in combination with lenvatinib and PD-1 inhibitor versus lenvatinib plus PD-1 inhibitor alone in patients with advanced HCC and lung metastasis.

Methods

Participants

In this retrospective study, patients with HCC and lung metastasis who received lenvatinib and PD-1 inhibitor between January 2019 and January 2024 at three centers in China – the First Affiliated Hospital of Sun Yat-sen University, Guangdong Provincial People’s Hospital, and the Memorial Hospital of Sun Yat-sen University – were reviewed. Based on their treatment regimen, patients were divided into two groups: the HLP group (receiving a triple combination of HAIC, lenvatinib, and PD-1 inhibitor) and the LP group (receiving the dual combination of lenvatinib and PD-1 inhibitor). The number of patients enrolled at each center is presented in Table S1.

The inclusion criteria were as follows: (1) age between 18 and 75 years; (2) diagnosis of HCC confirmed by contrast-enhanced imaging in high-risk individuals or histopathology, according to the American Association for the Study of Liver Diseases (AASLD) guidelines; (3) lung metastasis diagnosed by computed tomography (CT) or biopsy, beyond the indications for ablation; (4) Eastern Cooperative Oncology Group Performance Status score (ECOG PS) ≤ 2 and Child-Pugh class A or B; (5) no prior treatment for HCC; (6) no history of other malignancies within the past five years; (7) receipt of at least six cycles of PD-1 inhibitor and two months of lenvatinib in both groups, and at least two cycles of HAIC in the HLP group; (8) a minimum follow-up period of 12 months from enrollment to the study cut-off; and (9) complete clinical and follow-up data. Key exclusion criteria included: (1) pathological diagnosis of fibrolamellar HCC, sarcomatous HCC, or combined hepatocellular-cholangiocarcinoma (HCC-CC); (2) signs of hepatic decompensation, such as hepatic encephalopathy or gastrointestinal variceal bleeding; and (3) discontinuation or change of therapy without a valid medical justification. Laboratory tests and imaging evaluations – including contrast-enhanced CT or magnetic resonance imaging (MRI) – were performed within one week prior to the initiation of treatment.

The study was approved by the Ethics Committee of the First Affiliated Hospital of Sun Yat-sen University.

Treatment Procedures

HAIC was performed by experienced interventional radiologists at each participating center. The procedure followed established protocols for administering the FOLFOX regimen, as described in previous studies.¹⁴ The catheter tip was super-selectively placed into the tumor-feeding branch of the hepatic artery, based on the size, location, and vascular supply of the intrahepatic tumor. The chemotherapy regimen included oxaliplatin (130 mg/m², administered over 2 hours on day 1), leucovorin (200 mg/m², administered from hour 2 to 4 on day 1), and fluorouracil (400 mg/m² as a bolus over 15 minutes, followed by 2400 mg/m² via continuous infusion over 46 hours on days 1 and 2). This cycle was repeated every 3 weeks for a maximum of 8 cycles, depending on the physician’s assessment.

Both groups received oral lenvatinib at a dose of 8 mg (for body weight ≤60 kg) or 12 mg (for body weight >60 kg), along with intravenous PD-1 inhibitor (200 mg every 3 weeks). The types of PD-1 inhibitors used are listed in Table S2. In the HLP group, lenvatinib was administered on the first day of HAIC, and PD-1 inhibitor was infused on the second day immediately following completion of HAIC. In the LP group, lenvatinib and PD-1 inhibitor were administered on the same day. If patients experienced intolerable adverse events (AEs), the dose of lenvatinib or PD-1 inhibitor was either reduced or temporarily discontinued until symptoms resolved.

Efficacy and Safety Assessment

Chest CT and abdominal contrast-enhanced CT or MRI were performed every two treatment cycles. Imaging assessments were independently conducted by two radiologists with expertise in liver diseases. In cases of disagreement, a senior radiologist reviewed the images, and a consensus was reached.

Treatment efficacy was evaluated according to the Response Evaluation Criteria in Solid Tumors (RECIST) version 1.1. The primary endpoint was overall survival (OS), defined as the time from initial admission to death from any cause. Secondary endpoints included progression-free survival (PFS), objective response rate (ORR), and disease control rate (DCR). PFS was defined as the time from admission to either radiological disease progression or death, whichever occurred first. ORR was defined as the proportion of patients who achieved a complete response (CR) or partial response (PR), while DCR was defined as the proportion of patients who achieved CR, PR, or stable disease (SD). Adverse events (AEs) occurring during treatment were documented and graded according to the Common Terminology Criteria for Adverse Events (CTCAE), version 5.0.

Propensity Score Matching Analysis

Propensity score matching (PSM) was employed to minimize selection bias and balance baseline characteristics between the two treatment groups. Stepwise logistic regression was used to identify variables associated with treatment allocation, including age, sex, ECOG-PS, etiology, number of intrahepatic lesions, tumor size, number of lung metastases, presence of portal vein tumor thrombosis (PVTT), alpha-fetoprotein (AFP) level, albumin-bilirubin (ALBI) grade, and Child-Pugh class. A 1:1 nearest-neighbor matching algorithm with a caliper width of 0.2 was applied.

Statistical Analysis

Statistical analyses were performed using R software (version 4.0.3; R Foundation for Statistical Computing, Vienna, Austria) and SPSS version 27.0 (IBM Corp., Armonk, NY, USA). The normality of continuous variables was assessed using the Shapiro–Wilk test. Continuous variables were expressed as mean ± standard deviation or median with interquartile range (IQR), depending on their distribution, and were compared between groups using either the Student’s t-test or the Mann–Whitney U-test. Categorical variables were presented as counts and percentages, and compared using the χ²-test or Fisher’s exact test, as appropriate. The Kaplan–Meier method was used to estimate OS and PFS, and differences between groups were assessed using the Log rank test. Univariable and multivariable Cox proportional hazards regression analyses were conducted to identify factors associated with survival outcomes. Variables with a P-value < 0.1 in univariate analysis were included in the multivariate model. A two-sided P-value < 0.05 was considered statistically significant.

Results

Participants

The patient enrollment flowchart is presented in Figure 1. A total of 422 patients with HCC and lung metastases were included in the study. Among them, 169 patients received the triple combination therapy of HAIC, lenvatinib, and PD-1 inhibitor (HLP group), while the remaining 253 patients were treated with lenvatinib and PD-1 inhibitor without HAIC (LP group). The follow-up period concluded on April 30, 2025, with median follow-up durations of 30.8 months (95% confidence interval [CI]: 16.8–51.9) in the HLP group and 32.6 months (95% CI: 17.4–50.8) in the LP group, respectively. To minimize selection bias, 1:1 PSM was performed, yielding 151 patients in each group. After PSM, baseline characteristics were well balanced between the two groups (Table 1).

Figure 1 Patient selection flowchart. A patient might meet several exclusion criteria, but they were excluded only once from the uppermost criteria. Abbreviations: PSM, propensity score matching; HLP, HAIC combined with lenvatinib and PD-1 inhibitor; LP, lenvatinib combined with PD-1 inhibitor; HAIC, hepatic arterial infusion chemotherapy; BCLC, Barcelona Clinic Liver Cancer; ECOG, Eastern Cooperative Oncology Group; PS, performance score.

Before PSM, patients in the HLP group received a median of 4 cycles of HAIC (range: 2–8), 6.8 months of lenvatinib (range: 2.5–12.2) and 12 cycles of PD-1 inhibitor (range: 6–20). In comparison, patients in the LP group received a median of 5.0 months of lenvatinib (range: 2.0–11.7) and 10 cycles of PD-1 inhibitor (range: 6–17).

Tumor Response

Treatment responses were assessed according to RECIST 1.1 criteria before and after PSM (Table 2). Prior to PSM, the HLP group demonstrated significantly higher ORR (46.7% versus 22.9%, P < 0.001) and DCR (82.8% versus 69.1%, P = 0.002) compared to LP group. After PSM, the superiority of the HLP group remained evident, with significantly higher ORR (47.7% versus 20.5%, P < 0.001) and DCR (86.1% versus 72.2%, P = 0.003).

Table 2 Treatment Efficacy Evaluated by RECIST 1.1 Criteria Before and After PSM

Survival Outcomes

Before PSM, 194 patients (76.7%) in the LP group and 68 (40.2%) patients in the HLP group had died by the end of follow-up. The median OS was significant longer in the HLP group (26.0 months, 95% CI: 18.4–NA) compared to the LP group (8.6 months, 95% CI: 7.9–10.0; HR: 0.36, 95% CI: 0.28–0.46, P < 0.001). Similarly, the HLP group exhibited a significantly longer median PFS than the LP group (7.7 months, 95% CI: 6.7–10.8 versus 5.4 months, 95% CI: 4.5–6.1; HR: 0.76, 95% CI: 0.61–0.95, P = 0.017). Among the 151 matched pairs after PSM, the HLP group continued to show a significantly longer median OS than the LP group (26.0 months, 95% CI: 18.6–NA versus 8.4 months, 95% CI: 6.7–11.8; HR: 0.36, 95% CI: 0.27–0.49, P < 0.001). Median PFS was also favored the HLP group (7.6 months, 95% CI: 6.3–9.0 versus 5.5 months, 95% CI: 3.9–6.3; HR: 0.77, 95% CI: 0.59–1.00, P = 0.048) (Figure 2).

Figure 2 Kaplan-Meier curves comparing OS and PFS among patients who underwent HAIC plus lenvatinib and PD-1 inhibitor (HLP) or lenvatinib plus PD-1 inhibitor (LP) before (a–b) and after (c–d) PSM. P values were calculated using Log rank test. Abbreviations: PSM, propensity score matching; HLP, HAIC combined with lenvatinib and PD-1 inhibitor; LP, lenvatinib combined with PD-1 inhibitor; HAIC, hepatic arterial infusion chemotherapy; OS, overall survival; PFS, progression-free survival; HR, hazard ratio; CI, confidence interval.

Univariate and Multivariate Analysis

Univariate and multivariate analyses were conducted to identify risk factors associated with OS and PFS (Table 3). Multivariate Cox regression analysis revealed that treatment option and PVTT were independent risk factors for both OS and PFS. Additionally, the presence of multiple lung metastases was an independent risk factor for OS, but not for PFS.

Table 3 Univariate and Multivariate Analyses of Predictors of Survival After Treatment

Subgroup Analysis

Forest plots were generated to compare outcomes between subgroups after PSM (Figure 3). For both OS and PFS, the HLP group showed greater benefit across nearly all subgroups compared to the LP group, except in cases with small sample sizes or impaired liver function. These findings suggest that HAIC combined with lenvatinib and PD-1 inhibitor may be effective across various subgroups of HCC patients with lung metastases. However, for patients with compromised liver function – such as those classified as ALBI grade 3 or Child-Pugh class B – treatment with lenvatinib and PD-1 inhibitor may be more appropriate.

Figure 3 Forest plots based on OS (a) and PFS (b) of each subgroup. P values were calculated using Log rank test. Abbreviations: PSM, propensity score matching; HLP, HAIC combined with lenvatinib and PD-1 inhibitor; LP, lenvatinib combined with PD-1 inhibitor; HAIC, hepatic arterial infusion chemotherapy; OS, overall survival; PFS, progression-free survival; HR, hazard ratio; CI, confidence interval; ECOG, Eastern Cooperative Oncology Group; PS, performance score; ALBI, albumin-bilirubin; AFP, alpha-fetoprotein; PVTT, portal vein tumor thrombus.

Progression Reason Analysis

Regarding the analysis of progression patterns, five modes of tumor progression were identified: intrahepatic lesion progression, development of new intrahepatic lesion(s), extrahepatic lesion progression, development of new extrahepatic lesion(s), and death. At the data cutoff, 118 patients in the HLP group and 238 in the LP group had experienced disease progression. Notably, individual patients could exhibit more than one mode of progression simultaneously. The distribution of progression modes in the HLP and LP groups was as follows: intrahepatic lesion progression (12.0% versus 34.1%), new intrahepatic lesion(s) (9.7% versus 24.7%), extrahepatic lesion progression (33.1% versus 17.1%), new extrahepatic lesion(s) (39.4% versus 16.4%), and death (5.7% versus 7.7%). Clearly, the proportions of intrahepatic lesion progression and new intrahepatic lesions were significantly lower in the HLP group compared to the LP group. (Figure S1).

Safety

As shown in Table 4, the overall incidence of adverse events (AEs) was 75.1% in the HLP group and 57.7% in the LP group. In the HLP group, the most common AEs were abdominal pain (52.7%), nausea (48.5%), decreased appetite (43.2%), and fatigue (34.9%). The most frequent grade 3–4 AEs included abdominal pain (14.2%), nausea (10.1%), and diarrhea (9.5%), most of which were associated with HAIC. In the LP group, decreased appetite (36.0%), fatigue (34.0%), hypoproteinemia (21.7%), and elevated AST (20.9%) were the most common AEs, while the most frequent grade 3–4 AEs were fatigue (6.3%), immune-related AEs (4.3%), and proteinuria (3.6%). Although the incidence of both any-grade and grade 3–4 AEs was higher in the HLP group, these events were generally manageable, and no treatment-related deaths occurred during the study period.

Table 4 Treatment-Related Adverse Events

Discussion

This study is the first multicenter clinical trial to compare the efficacy of triple combination therapy—HAIC plus lenvatinib and PD-1 inhibitor—with that of the dual regimen of lenvatinib and PD-1 inhibitor in patients with advanced HCC and lung metastases. To minimize confounding variables, we applied propensity score matching (PSM), enrolled a relatively large cohort from multiple centers, and conducted long-term follow-up. In both the full and PSM-adjusted cohorts, the triple therapy significantly improved overall survival (OS), progression-free survival (PFS), and objective response rate (ORR). Notably, although the triple combination therapy was associated with a higher incidence of adverse events (AEs), all AEs were effectively managed with appropriate interventions, indicating that the regimen is both safe and tolerable.

Systemic therapy is recommended as the first-line treatment for advanced HCC, including cases with extrahepatic metastasis; however, control of intrahepatic lesions remains the most critical factor influencing patient survival.¹⁵ This highlights the importance of combining systemic and locoregional therapies. Transarterial chemoembolization (TACE), the most commonly used locoregional approach, involves intra-arterial delivery of cytotoxic agents emulsified with lipiodol into the lesions, followed by embolization to induce both cytotoxic and ischemic effects. TACE can effectively control intrahepatic tumors, achieving an objective response rate (ORR) of approximately 24% in advanced HCC.¹⁶ Nevertheless, its use is limited in cases with extensive intrahepatic tumor burden or severely compromised portal vein flow.¹⁷ Furthermore, the acute hypoxia induced by embolization may stimulate vascular endothelial growth factor (VEGF) expression, potentially promoting angiogenesis, local recurrence, and distant metastasis, including to the lungs.¹⁸ Compared with TACE, hepatic arterial infusion chemotherapy (HAIC) using the FOLFOX regimen has shown superior outcomes, with reported ORR of 46% versus 18% and median overall survival of 23.1 versus 16.2 months in patients with unresectable HCC.¹⁰ The high-dose, continuous infusion of chemotherapeutic agents in HAIC maximizes cytotoxic effects on intrahepatic tumors. In addition, HAIC preserves hepatic arterial flow, thereby mitigating tumor hypoxia and avoiding the pro-metastatic effects associated with embolization. The gradual systemic release of chemotherapeutic agents from the liver also contributes to systemic anti-tumor effect. Notably, chemotherapy has been associated with survival benefits in HCC patients with lung metastases.¹⁹ Moreover, the combination of FOLFOX-based HAIC with lenvatinib and toripalimab (a PD-1 inhibitor) has demonstrated promising anti-tumor efficacy in patients with HCC and extrahepatic spread.²⁰

In our study, the triple therapy combining HAIC-FOLFOX with lenvatinib and PD-1 inhibitor significantly improved median OS (26.0 versus 8.4 months) and ORR (46.7% versus 22.9%) compared to dual systemic therapy in patients with advanced HCC and lung metastases. A prior study reported that patients with lung metastases receiving sorafenib monotherapy had a median OS of only 7.37 months, whereas those treated with sorafenib plus locoregional therapies (primarily TACE) achieved a median OS of 18.37 months.¹¹ Compared with this combination, our triple combination regimen demonstrated superior survival outcomes in this high-risk subgroup. In a phase II trial, patients with extrahepatic metastases received apatinib (a TKI) plus HAIC as second-line therapy, with tumor shrinkage observed in 87.2% of intrahepatic and 74.4% of extrahepatic lesions,¹² suggesting the potential of HAIC to exert systemic anti-tumor effects beyond the liver. Locoregional therapies, when combined with targeted and immunotherapeutic agents, may modulate the tumor microenvironment and inhibit the invasion and migration of cancer cells, thereby potentially limiting extrahepatic metastasis.¹⁸

We selected the combination of lenvatinib and PD-1 inhibitor as the control group, given the promising synergistic antitumor effects observed in unresectable or advanced HCC. However, limited studies have specifically examined its efficacy in the subset of patients with lung metastases. Lenvatinib is a multitargeted tyrosine kinase inhibitor (TKI) that suppresses VEGFR 1–3, which play key roles in pathological angiogenesis. PD-1 inhibitor is a monoclonal antibody that binds to and inhibits the PD-1 receptor expressed on activated immune cells, thereby enhancing anticancer immune responses. Lenvatinib has been shown to augment the efficacy of anti-PD-1 antibodies by normalizing tumor vasculature and promoting immune cell infiltration in HCC.²¹ The combination of TKI with PD-1 inhibitor has been demonstrated to improve conversion rates in unresectable HCC.²² In prior studies, lenvatinib plus PD-1 inhibitor therapy yielded superior outcomes compared to PD-1 inhibitor monotherapy, with higher ORR (32.7% versus 10.3%), longer median PFS (10.6 versus 4.4 months), and OS (18.4 versus 8.5 months).²³ Similar findings have been reported in other dual-regimen studies compared to systemic monotherapy in advanced HCC.²⁴ However, in our study, the results for the dual-agent control group were less favorable, with an ORR of 18% and a median OS of 8.4 months.

Compared to the dual-agent control group, the triple combination therapy demonstrated encouraging outcomes in terms of long-term survival and tumor regression in patients with lung metastases. Similarly, a triple regimen combining TACE, lenvatinib, and camrelizumab (a PD-1 inhibitor) showed promising ORR and conversion rates in patients with unresectable HCC.²⁵ Another study reported that the combination of TACE, lenvatinib, and PD-1 inhibitor provided superior OS compared to TACE plus lenvatinib in patients with extrahepatic metastases, although the sample size was small.²⁶ A recent meta-analysis also supported the enhanced efficacy of this triple combination approach, showing that TACE combined with lenvatinib and PD-1 inhibitor significantly improved OS, PFS, and ORR compared to monotherapy, dual-agent regimens, and even the triple combination of TACE, sorafenib, and PD-1 inhibitors.²⁷ In a single-arm trial involving 36 patients with advanced HCC (27.8% with extrahepatic spread), treated with HAIC plus lenvatinib and toripalimab yielded a median OS of 17.9 months and an ORR of 63.9%. Moreover, increased levels of peripheral CD8⁺ and CD4⁺ T cells were observed following the combination therapy, indicating an enhanced systemic immune response.²⁸ Preclinical studies have also shown that lenvatinib combined with HAIC-FOLFOX exerts a synergistic inhibitory effect on HCC proliferation and angiogenesis by suppressing phosphorylation of multiple targets.²⁹

According to clinical guidelines, for advanced HCC with lung metastases, local ablation therapy is generally recommended when the lung tumor burden is relatively low. Typically, this includes patients with no more than three metastatic lesions, each with a maximum diameter of 3 cm, and well-controlled intrahepatic disease. Studies have shown that ablation of lung metastatic lesions, especially in cases of oligometastasis, can significantly improve patient survival and may even lead to long-term outcomes comparable to those of patients without lung metastases.³⁰ In our study, the enrolled patients with lung metastases were initially ineligible for ablation therapy. However, following treatment, 17 (10.1%) and 9 (3.6%) patients in the HLP and LP group met the criteria for pulmonary ablation and subsequently underwent the procedure. Furthermore, both groups of patients who received pulmonary ablation showed significantly better survival outcomes compared to those who did not receive ablation. Notably, the proportion of patients who were successfully converted to ablation candidates was significantly higher in the HLP group (P = 0.012), indicating superior clinical efficacy. These findings further support the synergistic antitumor effect of HAIC combined with systemic therapy. This combination not only effectively controls intrahepatic disease but also contributes to the suppression of lung metastases. The possible underlying mechanisms include: (1) systemic circulation of chemotherapeutic agents used in HAIC, which exert antitumor effects on lung lesions; (2) immunomodulatory effects of chemotherapy, which may enhance both hepatic and pulmonary immune microenvironments, thereby synergizing with targeted and immunotherapeutic agents; and (3) effective intrahepatic disease control by HAIC may improve the overall immunosuppressive state, thus potentiating the efficacy of systemic treatments. Further investigations are warranted in future clinical and preclinical studies.

We also conducted subgroup analyses to evaluate factors influencing OS and PFS. The results showed that triple therapy demonstrated superior survival outcomes across the majority of subgroups. However, in subgroups with impaired liver function, including patients classified as Child-Pugh class B or ALBI grade 3, dual therapy showed a trend toward better prognosis. This suggests that patients with compromised liver function may benefit more from systemic therapy alone, and the addition of HAIC may not be appropriate. The primary reason is that HAIC, as a form of locoregional therapy, may inevitably impose additional stress on liver function. For patients with already impaired hepatic reserve, preserving liver function is often a higher priority than aggressive tumor control.³¹ Therefore, in such cases, adopting a milder treatment strategy may yield greater overall benefit.

In addition to its favorable clinical outcomes, the combination of HAIC with lenvatinib and PD-1 inhibitor was associated with an increased incidence of adverse events (AEs) to some extent. A higher frequency of HAIC-related AEs was observed, which is consistent with findings from previous HAIC clinical trials. However, these AEs were generally manageable with appropriate supportive medications and did not lead to disease progression or treatment discontinuation. The incidence of AEs in the HLP group was higher compared to patients treated with either locoregional or systemic therapy in earlier studies.^4,6 This may be attributed to the poorer baseline conditions of the enrolled patients, as well as the potential additive toxicity from systemic therapy. A common HAIC-specific AE was abdominal pain caused by arterial vasospasm during oxaliplatin infusion. Currently, there is no effective method to completely prevent this particular type of pain, other than administering analgesics and antispasmodics or reducing the infusion rate of oxaliplatin.³² Overall, the combination of HAIC, lenvatinib, and PD-1 inhibitor was found to be safe and tolerable.

There are several potential limitations to our study. First, as a retrospective analysis, the possibility of selection bias cannot be entirely excluded. Although propensity score matching (PSM) was employed to minimize baseline differences between the two groups, residual confounding may still exist. Therefore, prospective randomized controlled trials are necessary to further validate our findings. Second, the majority of patients in our cohort had hepatitis B virus (HBV)-related HCC, which may limit the generalizability of our results to patients with other etiologies. Third, this study exclusively included patients treated with lenvatinib and PD-1 inhibitor; thus, additional research is warranted to assess whether similar outcomes can be achieved using alternative systemic agents.

In conclusion, compared to lenvatinib plus PD-1 inhibitor, the addition of HAIC to lenvatinib and PD-1 inhibitor significantly improves OS, PFS, and ORR in patients with HCC and lung metastases, while maintaining an acceptable safety profile.

Ethic Approval

This study was conducted in accordance with the principles of the Declaration of Helsinki. Ethical approval was obtained from the Ethics Committee of the First Affiliated Hospital of Sun Yat-sen University. Written informed consent for the treatment was obtained from all patients.

Funding

This work was supported by grants from the National Natural Science Foundation of China (82202271), Guangzhou Basic and Applied Basic Research Foundation [no. 202201011304].

Disclosure

The authors report no relevant financial or non-financial interests in this work.

References

1. Bray F, Laversanne M, Sung H, et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2024;74:229–263. doi:10.3322/caac.21834

2. Reig M, Forner A, Rimola J, et al. BCLC strategy for prognosis prediction and treatment recommendation: the 2022 update. J Hepatol. 2022;76:681–693. doi:10.1016/j.jhep.2021.11.018

3. Llovet JM, Ricci S, Mazzaferro V, et al. Sorafenib in advanced hepatocellular carcinoma. N Engl J Med. 2008;359:378–390. doi:10.1056/NEJMoa0708857

4. Kudo M, Finn RS, Qin S, et al. Lenvatinib versus sorafenib in first-line treatment of patients with unresectable hepatocellular carcinoma: a randomised Phase 3 non-inferiority trial. Lancet. 2018;391:1163–1173. doi:10.1016/S0140-6736(18)30207-1

5. Cheng AL, Qin S, Ikeda M, et al. Updated efficacy and safety data from IMbrave150: atezolizumab plus bevacizumab vs. sorafenib for unresectable hepatocellular carcinoma. J Hepatol. 2022;76:862–873. doi:10.1016/j.jhep.2021.11.030

6. Qin S, Kudo M, Meyer T, et al. Tislelizumab vs sorafenib as first-line treatment for unresectable hepatocellular carcinoma: a phase 3 randomized clinical trial. JAMA Oncol. 2023;9:1651–1659. doi:10.1001/jamaoncol.2023.4003

7. Llovet JM, Kudo M, Merle P, et al. Lenvatinib plus pembrolizumab versus lenvatinib plus placebo for advanced hepatocellular carcinoma (LEAP-002): a randomised, double-blind, phase 3 trial. Lancet Oncol. 2023;24(12):1399–1410. doi:10.1016/S1470-2045(23)00469-2

8. EASL clinical practice guidelines on the management of hepatocellular carcinoma. J Hepatol. 2025;82:315–374. doi:10.1016/j.jhep.2024.08.028

9. Singal AG, Llovet JM, Yarchoan M, et al. AASLD Practice Guidance on prevention, diagnosis, and treatment of hepatocellular carcinoma. Hepatology. 2023;78:1922–1965. doi:10.1097/HEP.0000000000000466

10. Li QJ, He MK, Chen HW, et al. Hepatic arterial infusion of oxaliplatin, fluorouracil, and leucovorin versus transarterial chemoembolization for large hepatocellular carcinoma: a randomized phase III trial. J Clin Oncol. 2022;40:150–160. doi:10.1200/JCO.21.00608

11. Chen J, Lu S, Zhang Y, et al. Sorafenib monotherapy versus sorafenib combined with regional therapies for hepatocellular carcinoma patients with pulmonary oligometastases: a propensity score-matched analysis. J Cancer. 2018;9:1745–1753. doi:10.7150/jca.24568

12. Chen S, Wang X, Yuan B, et al. Apatinib plus hepatic arterial infusion of oxaliplatin and raltitrexed for hepatocellular carcinoma with extrahepatic metastasis: phase II trial. Nat Commun. 2024;15:8857. doi:10.1038/s41467-024-52700-z

13. Guan R, Zhang N, Deng M, et al. Patients with hepatocellular carcinoma extrahepatic metastases can benefit from hepatic arterial infusion chemotherapy combined with lenvatinib plus programmed Death-1 inhibitors. Int J Surg. 2024;110:4062–4073. doi:10.1097/JS9.0000000000001378

14. Lyu N, Wang X, Li JB, et al. Arterial chemotherapy of oxaliplatin plus fluorouracil versus sorafenib in advanced hepatocellular carcinoma: a biomolecular exploratory, randomized, phase III trial (FOHAIC-1). J Clin Oncol. 2022;40:468–480. doi:10.1200/JCO.21.01963

15. Lee JI, Kim JK, Kim DY, et al. Prognosis of hepatocellular carcinoma patients with extrahepatic metastasis and the controllability of intrahepatic lesions. Clin Exp Metastasis. 2014;31:475–482. doi:10.1007/s10585-014-9641-x

16. Pinter M, Hucke F, Graziadei I, et al. Advanced-stage hepatocellular carcinoma: transarterial chemoembolization versus sorafenib. Radiology. 2012;263:590–599. doi:10.1148/radiol.12111550

17. Sieghart W, Hucke F, Peck-Radosavljevic M. Transarterial chemoembolization: modalities, indication, and patient selection. J Hepatol. 2015;62:1187–1195. doi:10.1016/j.jhep.2015.02.010

18. Jia W, Liang S, Lin W, et al. Hypoxia-induced exosomes facilitate lung pre-metastatic niche formation in hepatocellular carcinoma through the miR-4508-RFX1-IL17A-p38 MAPK-NF-kappaB pathway. Int J Biol Sci. 2023;19:4744–4762. doi:10.7150/ijbs.86767

19. Wang H, Huang J, Zhang W, et al. Prognostic factors in patients with first diagnosis of hepatocellular carcinoma presenting with pulmonary metastasis and construction of a clinical prediction model. Updates Surg. 2024;76:71–85. doi:10.1007/s13304-023-01603-7

20. He M, Huang Y, Du Z, et al. Lenvatinib, Toripalimab plus folfox chemotherapy in hepatocellular carcinoma patients with extrahepatic metastasis: a biomolecular exploratory, phase II Trial (LTSC). Clin Cancer Res. 2023;29:5104–5115. doi:10.1158/1078-0432.CCR-23-0060

21. Deng H, Kan A, Lyu N, et al. Dual vascular endothelial growth factor receptor and fibroblast growth factor receptor inhibition elicits antitumor immunity and enhances programmed cell death-1 checkpoint blockade in hepatocellular carcinoma. Liver Cancer. 2020;9:338–357. doi:10.1159/000505695

22. Zhang W, Tong S, Hu B, et al. Lenvatinib plus anti-PD-1 antibodies as conversion therapy for patients with unresectable intermediate-advanced hepatocellular carcinoma: a single-arm, phase II trial. J Immunother Cancer. 2023;11.

23. Liu Q, Li R, Li L, et al. Efficacy and safety of anti-PD-1 monotherapy versus anti-PD-1 antibodies plus lenvatinib in patients with advanced hepatocellular carcinoma: a real-world experience. Ther Adv Med Oncol. 2023;15:17588359231206274. doi:10.1177/17588359231206274

24. Qin S, Chan SL, Gu S, et al. Camrelizumab plus rivoceranib versus sorafenib as first-line therapy for unresectable hepatocellular carcinoma (CARES-310): a randomised, open-label, international phase 3 study. Lancet. 2023;402:1133–1146. doi:10.1016/S0140-6736(23)00961-3

25. Wu X-K, Yang L-F, Chen Y-F, et al. Transcatheter arterial chemoembolisation combined with lenvatinib plus camrelizumab as conversion therapy for unresectable hepatocellular carcinoma: a single-arm, multicentre, prospective study. EClinicalMedicine. 2024;67:102367. doi:10.1016/j.eclinm.2023.102367

26. Cai M, Huang W, Huang J, et al. Transarterial chemoembolization combined with lenvatinib plus PD-1 inhibitor for advanced hepatocellular carcinoma: a retrospective cohort study. Front Immunol. 2022;13:848387. doi:10.3389/fimmu.2022.848387

27. Wang L, Lin L, Zhou W. Efficacy and safety of transarterial chemoembolization combined with lenvatinib and PD-1 inhibitor in the treatment of advanced hepatocellular carcinoma: a meta-analysis. Pharmacol Ther. 2024;257:108634. doi:10.1016/j.pharmthera.2024.108634

28. Lai Z, He M, Bu X, et al. Lenvatinib, toripalimab plus hepatic arterial infusion chemotherapy in patients with high-risk advanced hepatocellular carcinoma: a biomolecular exploratory, phase II trial. Eur J Cancer. 2022;174:68–77. doi:10.1016/j.ejca.2022.07.005

29. Wang M, Yao X, Bo Z, et al. Synergistic effect of lenvatinib and chemotherapy in hepatocellular carcinoma using preclinical models. J Hepatocell Carcinoma. 2023;10:483–495. doi:10.2147/JHC.S395474

30. Zhou Q, Li R, Wu S, et al. Metastasis-directed ablation of hepatocellular carcinoma with pulmonary oligometastases: a long-term multicenter study. Radiol Med. 2025;130:25–36. doi:10.1007/s11547-024-01907-7

31. Chen S, Shuangyan T, Shi F, et al. TACE plus lenvatinib and tislelizumab for intermediate-stage hepatocellular carcinoma beyond up-to-11 criteria: a multicenter cohort study. Front Immunol. 2024;15:1430571. doi:10.3389/fimmu.2024.1430571

32. Wu Z, Guo W, Chen S, Zhuang W. Determinants of pain in advanced HCC patients recieving hepatic artery infusion chemotherapy. Invest New Drugs. 2021;39:394–399. doi:10.1007/s10637-020-01009-x