Introduction
Pressure injury (PI) is a skin and soft tissue injury caused by intense and/or persistent pressure or pressure combined with shear force acting on bony prominences or medical device contact sites, commonly found in the sacrum, ankle, and buttocks[
1–
3]. PI is prevalent worldwide, especially in healthcare services. Epidemiological data show that the global prevalence of PI in hospitalized adults is 12.8%, and the incidence of hospital-acquired PI is 8.5%[
4]. Approximately 3 million patients are treated for PI each year in the United States, and PI is listed as one of the most common measurable adverse medical events. Studies in China show that the prevalence rate of PI among hospitalized patients in tertiary hospitals is 1.67%, and among intensive care unit (ICU) patients it is as high as 10.58%[
5]. The prevalence of PI in high-risk populations such as quadriplegic patients, spinal cord injury patients, long-term surgical patients, and elderly patients with hip or femur fractures can reach 50%–70%[
6–
8]. PI stages are mainly stage Ⅰ (43.5%) and stage Ⅱ (28.0%). Severe PI (including stage III, IV, deep tissue injury, and unstageable PI) accounts for approximately 30% of all PI and approximately 18% of hospital-acquired injuries[
4,
9]. Severe PI has 4 times the impact of superficial PI, often leading to refractory wounds and tissue necrosis, significantly increasing the risk of pain, disability, and death, and seriously affecting the patient’s psychological state, daily life, and social function[
10–
13]. The prolonged hospital stay and infection risk associated with PI further exacerbate the consumption of medical resources.
For a long time, PI has been prevented and managed through Braden Risk Scale screening, regular turning decompression, functional dressing protection and individualized nutrition intervention, and debridement dressing and negative pressure therapy have been adopted for the formed wounds[
14–
16]. However, traditional assessment tools are highly homogeneous, highly dependent on the experience and execution frequency of nursing staff, unable to realize dynamic and accurate prediction in combination with massive individualized physiological data of patients, difficult to reduce the new incidence of PI in severe patients from the source by conventional prevention and management mode, and limited in early recognition of deep tissue injury[
17,
18].
In recent years, with the application of artificial intelligence (AI) technology in PI research, the limitations of traditional nursing‑care optimization have been gradually overcome. Research has expanded to include intelligent technologies such as algorithm modeling, intelligent wound recognition, and dynamic risk alerting[
19–
21]. For example, Dweekat et al. constructed a hybrid predictive model of the Braden scale combined with machine learning to optimize risk identification by integrating multiple clinical indicators in electronic health records; its high-risk patient identification sensitivity reached 74.29%, which was about 11 percentage points higher than that of 66.9% evaluated by the traditional single Braden scale. Such intelligent exploration does not require additional clinical workload and provides innovative solutions for fine prevention and management of PI[
22]. Song et al. improved the PI risk assessment system based on a machine learning integration algorithm, integrating nursing assessment phenotype and electronic health record multi-dimensional clinical indicators, and the Area Under the Curve (AUC) of the high-risk patient prediction model reached 0.94. These intelligent explorations provide innovative solutions for the fine prevention and management of PI[
23].
In order to clarify the cutting-edge research direction of AI-enabled PI, it is very important to fully understand the research progress and hot spot evolution trend of this cross-cutting field. Therefore, scholars at home and abroad have carried out a large number of review studies in the fields of nursing, wound medicine, computer AI, geriatrics, evidence-based nursing, etc.[
24]. A systematic review by Barghouthi et al. (2023) showed that machine learning predictive models were able to identify the risk of PI in hospitalized adult patients earlier than traditional assessment tools. This review provides evidence support for identification and evaluation of the application of machine learning algorithms in PI risk prediction[
25]. Anisuzzaman et al. (2022) systematically evaluated the application of image-based AI techniques in wound assessment and found that AI image analysis methods exhibited reliable performance in wound size measurement and tissue classification, effectively reducing observer bias in subjective assessments[
26]. However, most of the narrative review studies rely on the author’s literature sorting and subjective induction, which makes the review conclusions heterogeneous and biased, and it is difficult to objectively present the overall research knowledge in this field from a quantitative perspective.
In this study, we used a convenient and reliable statistical analysis method—bibliometric analysis—to quantitatively and qualitatively evaluate the research in the intersection of AI and PI care[
27–
31]. Compared with traditional descriptive reviews, bibliometrics can quantitatively analyze the characteristics of discipline publications, author-institution collaboration, research hotspot evolution, and generate visual maps[
32]. At the same time, with the rise of the Internet and academic social media, the way scientific research results are disseminated has undergone profound changes. The limitations of traditional evaluation indicators (such as citation frequency and journal impact factor [IF]) are increasingly prominent. Altmetrics are attracting attention as a new tool to measure the impact of research among the public, clinical practitioners, and policymakers. There are few studies that combine bibliometrics and altmetrics in this field[
33,
34]. Here, this study combined bibliometrics and altmetrics indicators to analyze the core literature related to AI and PI nursing, explore the research hotspots, knowledge structure, and technical trends in this field, and explore the internal relationship between the two types of evaluation indicators, providing reference for the subsequent in-depth research of AI in PI prevention, assessment, and wound diagnosis.
Materials and methods
Search strategy
A search of the Web of Science Core Collection (WoSCC) database on June 1, 2026 provided a record of all references used in this study. We designated “artificial intelligence in pressure injury care” as the research field, and searched with “artificial intelligence” OR “machine learning” OR “deep learning” OR “neural network” OR “pressure ulcer” OR “pressure injury” OR “bedsore” OR “decubitus ulcer” OR “nursing” OR “wound care” OR “prevention” OR “assessment” OR “risk assessment” as the keywords. Publication dates were set for 2022 to May 31st 2026, and types include only articles. Publications were limited to English and exclude articles not directly related to the focus of the study.
Data extraction
The screening and selection process followed the PRISMA 2020 guidelines[
35], as illustrated in Figure 1. A systematic search was conducted in the WoSCC, yielding 227 records. Before screening, 91 records were removed by automation tools, leaving 136 records for title and abstract screening. During the initial screening phase, two independent reviewers assessed the titles and abstracts of the 136 records against the following criteria: (a) English-language publications; (b) original research articles (excluding reviews, editorials, conference abstracts, and letters); and (c) studies with a direct thematic focus on the application of AI in PI care. Articles were considered “directly relevant” if they explicitly addressed AI-assisted risk prediction, detection, classification, staging, or management of pressure injuries. Studies focusing exclusively on general wound care without AI components, or on PI topics unrelated to AI, were excluded. Disagreements between reviewers were resolved through discussion, with a third reviewer consulted when consensus could not be reached. Following this phase, 22 records were excluded, resulting in 114 reports proceeding to the next stage of assessment.
Based on WoSCC citation counts ranked in descending order, we retained all cited articles (those with at least one citation, n = 82) and supplemented 18 uncited articles—primarily recent publications from 2025 to 2026, with some having available Altmetric attention score (AAS) data. This was done to meet the minimum requirement for stable co-occurrence network analysis in VOSviewer, ensuring that the number of nodes in the network reached at least 100. The remaining 14 uncited articles were excluded because they did not have significant co-occurrence relationships with the core literature cluster. Ultimately, we obtained 100 studies that met the criteria for analysis. These samples encompass all articles indexed in the Web of Science (WoS) database with substantial citation counts, while also including articles that have received considerable societal attention.
Microsoft Excel 2019 software was used to summarize the author, country, institution, publication year, publication journal, total number of citations, research field, keywords, literature type, and journal IF of the 100 articles retrieved. The immediacy index (II), article influence score (AIS), journal IF, publication year, total citations (TCs), and journal normalized eigenfactor (Nor-EF) were all from the Journal Citation Report (JCR). As time passes, older manuscripts accumulate more citations. To enable a more equitable comparison of citation impact across articles with different publication ages, we calculated the number of years since publication (NYsP) for each article and the average citations per year (ACpY), defined as TC divided by NYsP.
Establishment of bibliometric network
VOSviewer version 1.6.20 was used to establish a bibliometric network to analyze countries, institutions, authors, journals, co-citations, and keywords. The visual network consists of circular nodes and connecting lines, the size of the nodes reflecting the frequency of co-occurrence, and the size of the lines between the nodes representing the strength between the two nodes. The colors of the nodes represent different clusters. The larger the node, the more frequently it appears. The wider the two-point line, the higher the frequency of co-occurrence.
Social attention analysis
AASs for the 100 most cited articles were obtained by searching the DOI of each article via the Altmetric Bookmarklet and comparing their social impact to the scientific attention expressed in the citation index.
Data analysis
SPSS 24.0 software was used for data analysis. The following results were analyzed: citations, years, countries, authors, institutions, journals, WoS categories, keywords, co-citations, and research categories. Shapiro-Wilk test was used to assess the normality of the data. Spearman correlation test was used to find the correlation between indicators. The correlation is classified as strong;
r ≥ 0.60 indicates strong correlation, 0.30 <
r < 0.60 indicates moderate correlation, and
r ≤ 0.30 indicates weak correlation[
36].
P < 0.05 means the result is statistically significant.
Results
The WoS search results contained 136 records. Articles were ranked from highest to lowest according to total citations, and the 100 articles were selected for analysis. Table 1 lists 100 articles sorted by citations after applying exclusion criteria. The number of citations ranged from 0 to 42 (median 5). Altmetric scores range from 0 to 76. Of the 100 articles counted, 87 were cited, and 35 had Altmetric scores. Lau et al., “An AI-enabled smartphone app for real-time pressure injury assessment,” Frontiers in Medical Technology, 2022, has been cited 42 times. Padula et al., “Predicting pressure injury risk in hospitalised patients using machine learning with electronic health records: a US multilevel cohort study,” published in BMJ OPEN in 2024, had an AAS score of 76 and has been mentioned by 10 news outlets, 1 blog, 2 X users, 1 Reddit user, and 1 Bluesky user. Meanwhile, the article ranks ninth in citations. “Effects of a transformer-based AI-based application to support incontinence-associated dermatitis and pressure injury assessment, nursing care and documentation: Controlled pilot intervention study,” International Journal of Nursing Studies Advances, June 2026. Qian et al., “Development and Internal Validation of a Gradient Boosting Model for Pressure Injury Risk in the ICU” and Lupianez-Perez et al., “Machine learning methods to evaluate fluctuation in oxygenation and peripheral tissue temperature, in the sacral and trochanteric area under pressure in healthy subjects and institutionalized patients. A prospective non-randomized trial” and Deng et al., “Braden score for early mortality risk assessment in critically ill intensive care patients with multiple organ dysfunction syndrome” were not cited, but all AAS scores were 1 and were mentioned by 1 X user, respectively.
Annual publications and citations
Among the 100 articles selected by ranking all retrieved records in descending order of total citation count, publications from 2025 accounted for the largest proportion (n = 27), representing more than one quarter of the total, followed by those from 2024 (n = 26). Sixteen articles were from 2022, 10 from 2023, and 21 from 2026. The publication trend indicates that 2024–2025 was the peak research period in this field.
Although the publication output was similar between the two years, notable differences were found in the mean total citations (MeanTC): 8.5 in 2024 versus 19.85 in 2025, suggesting that studies published in 2025 gained stronger academic attention within a shorter timeframe. Despite incomplete annual data collection for 2026, the mean annual citation count had already reached 34.24, reflecting a continuous rise in the academic influence of research in this field.
Notably, 2023 contributed the smallest number of included articles (n = 10) but achieved a mean annual citation count of 8.7, which was higher than the figures for 2022 (0.88) and 2024 (8.5). This indicates that the limited studies published in 2023 exhibited relatively high per-article academic impact (Figure 2).
Countries and institutions
Articles included in this study were affiliated with 256 institutions across 30 countries. China contributed the largest number of publications (n = 38), followed by the United States (n = 15) (Figure 3). The co‑authorship analysis by country and region reveals the collaboration networks and the relative impact of different countries in this research area (Figure 4). In Figure 4, each node denotes a country or region with at least one co‑authored publication, and the node size is proportional to the number of co‑authored papers. The spatial distance between two nodes is indicative of the strength of collaborative ties between them. China was the most influential country, followed by Australia.
Sciendis GmbH (n = 4), University Hospital Essen (n = 4), and Koc University (n = 4) were the most productive institutions, followed by the University of Toronto (n = 3). As presented in Table 2, Sciendis GmbH and University Hospital Essen exhibited the strongest total link strength among all institutions with at least one co-authored article. Figure 5 demonstrates the institutional collaboration network, in which Ghent University maintained collaborative ties with multiple institutions.
Journal analysis
The 100 included articles were published across 67 distinct journals. Diagnostics (n = 5, total citations = 46), International Wound Journal (n = 5, total citations = 60) and Journal of Clinical Medicine (n = 5, total citations = 26) were the most productive journals, followed by Scientific Reports (n = 3, total citations = 12), Sensors (n = 3, total citations = 11), Journal of Clinical Nursing (n = 3, total citations = 6), Journal of Advanced Nursing (n = 3, total citations = 24), CIN: Computers, Informatics, Nursing (n = 3, total citations = 15) and Advances in Skin and Wound Care (n = 3, total citations = 5) (Figure 6).
The IF of the journals hosting the 100 articles ranged from 0 to 21.3 (median = 2.9), with Advanced Fiber Materials having the highest IF. Two of the 100 included articles were published in journals with an IF above 10. Statistical analysis revealed a significant correlation between the total citation count of the articles and journal IF (P < 0.05).
Author and co-author analysis
A total of 632 authors were involved across the 100 included articles. Seven authors—Majjouti, Khalid; Brehmer, Alexander; Pinnekamp, Hannah; Aleithe, Michael; Fischer, Uli; Kleesiek, Jens; and Hosters, Bernadette—each published 4 articles, making them the most productive authors in this analysis. Nine authors, including Pandey, Bhaskar; Gefen, Amit; Ho, Joyce C.; Dutta, Tilak; Morriss, Joshua; Barriga-Gallegos, Fredy; Kirkland-Kyhn, Holly; Pinnekamp, Hannah; and Fischer, Uli, published the highest number of articles as first or corresponding author (n = 2 each). The article “In-Advance Prediction of Pressure Ulcers via Deep-Learning-Based Robust Missing Value Imputation on Real-Time Intensive Care Variables” had the largest number of co-authors, with 17 contributors.
As shown in Figure 7, the seven authors with 4 or more co-authored articles all belong to the same research group. They jointly contributed to four studies (Table 3): “Nursing-centered development of an AI-based decision support system in pressure ulcer and incontinence-associated dermatitis management—a mixed methods study” (n = 1 citation), “Controlled Pilot Intervention Study on the Effects of an AI-based Application to Support Incontinence-Associated Dermatitis and Pressure Injury Assessment, Nursing Care and Documentation: Study Protocol” (n = 1 citation), “Fine-Grained Classification of Pressure Ulcers and Incontinence-Associated Dermatitis Using Multimodal Deep Learning: Algorithm Development and Validation Study” (n = 1 citation), and “Effects of a transformer-based AI-based application to support incontinence-associated dermatitis and pressure injury assessment, nursing care and documentation: Controlled pilot intervention study” (n = 0 citations).
Keyword co-occurrence
Keyword co-occurrence analysis was performed in this study. Across the 100 included articles, a total of 476 unique keywords were identified. Nine keywords occurred at least 10 times: “pressure injury” (n = 35), “machine learning” (n = 32), “artificial intelligence” (n = 27), “pressure ulcer” (n = 26), “prevention” (n = 21), “deep learning” (n = 21), “ulcers” (n = 17), “pressure ulcers” (n = 12), and “care” (n = 11). Among all keywords, “machine learning” had the highest number of links (248). A total of 23 keywords occurred 5 or more times, which were grouped into 4 clusters with 155 interconnections in total (Figure 8).
As illustrated in Figure 9, keywords closely associated with AI technologies—including extended reality, gradient boosting, and healthcare technology—emerged in 2025. VOSviewer cluster analysis revealed a gradual increase in the occurrence frequency of such terms in recent years.
Co-citation references
Co-citation occurs when two publications are both cited by a third publication. We performed co-citation analysis of the references using VOSviewer to identify the most influential publications for the 100 included articles on PI. A total of 3,101 references were cited across the 100 articles, 12 of which were cited 10 or more times. The most frequently co-cited work was the “Revised National Pressure Ulcer Advisory Panel Pressure Injury Staging System” by Edsberg et al.[
37], published in
Journal of Wound, Ostomy and Continence Nursing in 2016, which was referenced by 20 of the 100 included articles. The second most co-cited publication was “Using Machine Learning Technologies in Pressure Injury Management: Systematic Review” by Jiang et al., published in
JMIR Medical Informatics in 2021, which was cited by 16 of the 100 included articles (Figure 10).
Analysis of research topics
The 100 included articles covered a broad range of WoS subject categories (Table 4). Nursing was the most prevalent category (n = 22), followed by Engineering (n = 19) and Medical Informatics (n = 16). Four articles were classified under both Health Care Sciences & Services and Medical Informatics, and five articles fell under both Dermatology and Surgery.
Correlation
The results of Spearman correlation analysis are summarized in Table 5. The AAS was significantly positively correlated with the NYsP (r = 0.333, P < 0.01), but showed no significant correlation with IF, immediacy index II, AIS, TCs, ACpY, or journal normalized eigenfactor (Nor-EF) (all P > 0.05).
TC was positively correlated with IF (r = 0.566, P < 0.01), II (r = 0.427, P < 0.01), AIS (r = 0.411, P < 0.01), and Nor-EF (r = 0.965, P < 0.01). At the journal level, IF was positively correlated with II (r = 0.778), AIS (r = 0.829), Nor-EF (r = 0.611), and ACpY (r = 0.268) (all P < 0.01). II was also positively correlated with AIS (r = 0.587) and Nor-EF (r = 0.460) (both P < 0.01), as well as with ACpY (r = 0.246, P < 0.05). AIS was positively correlated with both Nor-EF (r = 0.464, P < 0.01) and ACpY (r = 0.233, P < 0.05). Additionally, Nor-EF was positively correlated with ACpY (r = 0.198, P < 0.05), and NYsP was positively correlated with ACpY (r = 0.535, P < 0.01). Meanwhile, we found that the AAS of the articles exhibited a significant positive correlation with their citation counts (r = 0.236, P < 0.05).
Discussion
In this study, we conducted a quantitative analysis of the 100 original research articles retrieved from the WoSCC, covering the period from 2022 to May 31st 2026, using combined bibliometric and altmetric approaches. We systematically mapped the temporal publication trends, regional and institutional collaboration patterns, journal distribution, evolution of research hotspots, and characteristics of key research contributors in this field, and elucidated the intrinsic associations between journal metrics, academic influence, and societal influence based on Spearman correlation analysis. Our findings clearly demonstrate the development trajectory, disciplinary features, technological directions, and value differentiation of the global AI-enabled PI care field. We further discuss the implications of core findings and their research significance from multiple dimensions.
The 100 included articles were published across 67 journals, affiliated with 256 institutions in 30 countries, and involved a total of 632 authors. Publication output peaked in 2024 (n = 26) and 2025 (n = 27). As of May 2026, 21 articles from 2026 had been included, indicating growing research interest in this interdisciplinary field of AI and PI nursing. In terms of the MeanTC, the mean value in 2022 was only 0.88, suggesting the field was still in its infancy with limited technological and clinical recognition. In 2023, the number of included articles fell to the lowest point of the five-year period (n = 10), but the MeanTC rose to 8.70. Most publications from this year were seminal studies with outstanding per-article academic value. In 2024, as research scaled up and more applied studies were published, the MeanTC declined slightly to 8.50. MeanTC increased markedly in 2025 (19.85) and 2026 (34.24), driven both by the expanding citation network of earlier studies and by the clinical translation of emerging technologies including deep learning, large language models, and intelligent sensing. Although full-year data for 2026 are not yet available, its MeanTC already ranks highest among the five years, confirming that the field remains in a phase of rapid development.
The most highly cited article in our dataset was “An artificial intelligence-enabled smartphone app for real-time pressure injury assessment” by Lau et al., with a total of 42 citations. This study focuses on PI assessment based on wound image recognition, and adopts the You Only Look once (YOLO)v4 deep learning object detection algorithm as its core technical framework. Using smartphone-captured wound images as input, the algorithm automatically locates the wound area, extracts morphological features, and performs staging classification via computer vision technology, and is ultimately packaged as a lightweight mobile application to enable real-time bedside assessment. This approach effectively addresses the limitations of traditional manual assessment, such as high subjectivity and a lack of portable bedside tools, and establishes a general technical framework for mobile-based intelligent wound assessment. The YOLO algorithm series is widely applied in wound recognition. The updated YOLOv8 model delivers comprehensive improvements in detection accuracy, robustness to interference, and inference speed, and has been deployed for PI detection and staging in complex settings such as ICUs. Such lightweight algorithms are well suited for mobile clinical use and have gained widespread recognition in the field. This finding also suggests that applied studies addressing unmet clinical needs and prioritizing practical utility are more likely to generate sustained academic impact.
The 100 included articles were affiliated with 256 institutions across 30 countries. Most countries conducting research on AI in PI care rank among the world’s highest gross domestic product (GDP) economies, indicating that research and development in this field requires sufficient financial support. Based on overall publication output and collaboration intensity, China, the United States, South Korea, and Canada were identified as the four leading countries. Among them, China, the United States, and Canada are also pioneering nations in the field of medical AI[
38], which may limit the geographical generalizability of the study findings[
39]. At the international level, China and Australia showed the most active collaboration, with complementary research strengths across countries. Highly productive institutions such as Sciendis GmbH and University Hospital Essen maintained close collaborative ties, forming stable research clusters. A total of 632 authors were involved across all included articles, reflecting the formation of established research teams in this field. The maximum number of authors on a single article was 17, which highlights the multidisciplinary nature of collaborative research spanning nursing, computer science, and bioengineering.
Diagnostics, International Wound Journal, and Journal of Clinical Medicine are the core publishing platforms in this field. Relying on their own positioning, the three journals form a differentiated division of labor, focusing on diagnostic technology, wound care specialty research, and clinical medicine transformation, respectively, which conform to multiple research directions such as intelligent assessment, risk prediction, and clinical application in the field. The characteristics of different journals also form complementary advantages. The two journals from multidisciplinary digital publishing institute (MDPI) have common advantages in publishing efficiency and open dissemination, becoming an important platform for rapid publication of cutting-edge research achievements in this field; while International Wound Journal has a relatively stable publishing efficiency, but it has deep academic accumulation and high recognition in the field of wound care.
Lau, Chun Hon (n = 42), Xu, Jie (n = 31), and Sin, Petr (n = 27) were the first authors most frequently cited in papers published during this study period. Xu, Jie and Sin, Petr were from International Wound Journal and Diagnostics, respectively, which published the most articles in this study. Further analysis shows that the authors with the highest citations all rely on institutional platforms with research advantages. Relying on their respective institutional platforms, the research teams of the three regions respectively produce benchmark achievements in the directions of mobile AI evaluation tools, clinical prediction models, algorithm optimization, etc. This also suggests that selecting corresponding journals in combination with research directions and relying on superior scientific research platforms to carry out research are important ways to enhance the influence of the industry.
In this study, the most cited articles have considerable authority in academic circles, covering the staging standards, definition revision, and clinical identification basis of PI, providing a unified and internationally recognized staging system for clinical medical staff to accurately assess and record PI[
37,
39]. The second most cited article covers the technical path and evidence summary of machine learning algorithms in PI prediction, diagnosis, and management, providing methodological reference and cutting-edge technical direction for intelligent PI management[
21].
Keyword co-occurrence analysis indicates that PI prevention, intelligent assessment, and wound recognition are the core research directions in this field, with machine learning and deep learning as the core technical approaches. Various intelligent algorithms effectively address the limitations of traditional Braden Scale assessment and manual evaluation, enable continuous dynamic monitoring, and deliver favorable performance in improving assessment accuracy and reducing nursing workload. Since 2025, emerging technologies such as extended reality, large language models, and gradient boosting algorithms have appeared continuously. This trend marks a shift from single-algorithm development to integrated hardware-software and comprehensive application paradigms, with application scenarios gradually expanding into home care, nursing education, and other fields.
Research in this field has also gradually expanded beyond the scope of pure technological development. Kirkland-Kyhn et al. explored nurses’ acceptance of intelligent systems in their article “Exploring Nurses’ Acceptability and Readiness for Patient-Centered Artificial Intelligence Systems in Pressure Injury Prevention.” They identified implementation barriers including insufficient digital literacy among nursing staff, poor alignment between system workflows and clinical practice, and data privacy risks, providing a reference for technology optimization and clinical promotion. Liu et al. developed flexible multifunctional textile sensors woven from hierarchical yarns for integrated sleep activity monitoring and thermotherapeutic care for pressure injuries, which expands the design and development landscape of wearable intelligent care devices. These findings reflect that the field has begun to balance technological innovation with clinical translation, adopting more comprehensiveresearch perspectives.
Spearman correlation analysis was performed to examine the associations between traditional bibliometric indicators (IF, II, AIS, Nor-EF, TC, NYsP, ACpY) and AAS. The JCR journal metrics showed high internal consistency. TC had positive correlation with Nor-EF (r = 0.965, P < 0.01), IF (r = 0.566, P < 0.01), II (r = 0.427, P < 0.01), and AIS (r = 0.411, P < 0.01). Nor-EF reflects the overall dissemination reach and disciplinary influence of a journal. Our results suggest that the overall academic scale and dissemination capacity of a journal are associated with the ACpY of individual articles (r = 0.198, P < 0.05), and likely represent an important factor influencing article citation counts. Meanwhile, IF, II, AIS, and Nor-EF were all significantly positively correlated with one another. This indicates that within the sample scope of this study, mainstream journal evaluation metrics, including IF, II, AIS, and Nor-EF, demonstrate logical consistency and mutual corroboration in this field.
NYsP exhibits a moderate statistically significant positive correlation with ACpY (r = 0.535, P < 0.01), which aligns with the time accumulation effect of academic citations: the longer a paper has been published, the more opportunities it typically has to attract attention, be drawn upon, and receive citations from peers. Furthermore, journal metrics including II (r = 0.246, P < 0.05) and AIS (r = 0.233, P < 0.05) only demonstrate a weak to moderate positive correlation with ACpY. This finding suggests that the influence of journal platforms has limited explanatory power for the long-term citation performance of individual papers. The innovativeness, practicability, and intrinsic value of the research itself are likely to be more critical determinants of long-term citation efficiency. Some high-quality studies focused on specialized fields may also attain sustained academic recognition even when published in non-top-tier journals.
AAS measures the social dissemination popularity of research outputs across channels including news media, social platforms, blogs, and academic social networking sites, reflecting social influence at the industry and public levels. The correlation analysis in this study yields distinctly differentiated results. AAS only exhibits a moderate statistically significant positive correlation with NYsP (r = 0.333, P < 0.01), and shows no statistically significant correlation with all journal academic metrics such as IF, II, AIS, and Nor-EF. This finding indicates that within the sample scope of this study, no obvious statistically significant association is observed between the traditional academic influence of journals and the social dissemination capacity of research outputs. Papers published in high-impact academic journals do not necessarily attract higher attention from the media and frontline practitioners, and peer academic evaluation and public dissemination evaluation present relatively independent evaluation characteristics. Meanwhile, a time accumulation effect exists in the process of social dissemination. Earlier published research outputs have a longer dissemination window to be discovered and reposted by external audiences. For literature newly published in the past two years, their overall AAS is generally lower due to the shorter dissemination cycle.
Correlation analysis showed that AAS was not statistically significantly correlated with TC (r = –0.014, P > 0.05). This result suggests that public attention captured by altmetric indicators and peer recognition reflected by traditional citation metrics may correspond to distinct dimensions of research impact. TC primarily reflects the degree of peer recognition within the academic community, with its dissemination and evaluation confined to scholarly research contexts. In contrast, AAS largely captures attention from non-academic stakeholders such as media outlets, the general public, and policy agencies, with notable differences in the driving factors and dissemination mechanisms underlying the two metrics. This finding provides empirical reference for incorporating altmetric indicators as a complementary dimension to traditional citation-based evaluation, indicating that AAS can capture impact information not covered by conventional citation metrics and contribute to a more comprehensive assessment of the multifaceted value of research outputs. The literature with the highest social attention (AAS = 76) is a machine learning-based risk prediction cohort study published by Padula et al. in 2024. Focusing on risk prediction of pressure injuries in hospitalized patients, this study fits well with scenarios of hospital quality management and healthcare policy formulation. It has been mentioned by 10 news media outlets, 1 blog, and 2 users on platform X, and has been read by 87 readers on Mendeley, with prominent social influence. The literature with the highest academic citations (Citations = 42) is a study on mobile AI assessment technology published by Lau et al. in 2022. Focusing on algorithm development and hardware adaptation, it mainly targets researchers in medical engineering and specialized nursing. This literature received no attention from mainstream media, with a corresponding attention score of AAS = 5 on social platform X.
As an illustrative case, the work published by Jiang et al. (2022) concerning early recognition integrating infrared thermography and AI achieved both moderate academic citations and social attention (AAS = 13, TC = 20), owing to its balance between non-invasive technological innovation and clinical practicability. Traditional citation metrics tend to assess the academic value generated by technological innovation and methodological advances, whereas the Altmetric score highlights cross-sector social value stemming from clinical applicability and health management implications. Reliance on a single indicator therefore risks incomplete evaluation of the overall value of research outputs. This case lends support to the rationale of combining bibliometrics and altmetrics adopted in the present study. The dual-perspective evaluation framework can simultaneously account for academic rigor and practical clinical value, which may enable a more objective portrayal of the comprehensive impact of research in this field.
Limitation
This study has several limitations. First, the data source was restricted to the WoSCC alone, with no literature retrieved from mainstream databases including Scopus and Google Scholar. Consequently, some region-specific studies and high-impact niche articles were omitted, which means the findings may not fully reflect the global research landscape. Second, only original research articles were enrolled, while reviews and systematic reviews were excluded. This neglects the vital role of review papers in consolidating domain knowledge and guiding disciplinary development. Third, the temporal dataset was incomplete. Data collection ended in May 2026, so full-year trends in publication volume, citations, and emerging research hotspots remain to be supplemented and validated. Fourth, citation-based ranking inherently favors publications with higher citation counts. Although our sample includes 21 articles published in 2026, this approach may still undervalue high-quality papers that have been published more recently and have not yet had sufficient time to accumulate citations.
Conclusion
Combined bibliometric and altmetric analyses revealed rapid development and promising prospects for AI applications in PI nursing. China produced the largest number of publications globally, followed by the United States, South Korea and Canada, which are also key contributors to research breakthroughs in this field.
Diagnostics, International Wound Journal and Journal of Clinical Medicine serve as three core journals publishing high-quality relevant studies. Machine learning and deep learning constitute the dominant technical approaches, and PI risk prediction, intelligent wound assessment and patient repositioning monitoring are the central research directions. Emerging AI technologies such as YOLO-series algorithms, large language models and extended reality have gained growing popularity in recent years.
This study mapped the core research hotspots and technological trends within the field, offering practical references for advancing the clinical translation of intelligent PI care technologies.
The Author(s) 2026. This article is published by Higher Education Press at journal.hep.com.cn.
PDF
(6875KB)
