We leveraged extracted data from the Web of Science Core Collection as of August 31, 2023. We used Web of Science for its comprehensive coverage across multiple databases, comprising a wide range of bibliometric indicators and literature from various disciplines. Using a predefined search strategy, we intended to include all relevant literature for bibliometric analysis. We used the following key words: artificial intelligence OR computational intelligence OR deep learning OR computer aided OR machine learning OR support vector machine OR data learning OR artificial neural network OR digital image OR convolutional neural network OR evolutionary algorithms OR feature learning OR reinforcement learning OR big data OR image segmentation OR hybrid intelligent system OR hybrid intelligent system OR recurrent neural network OR natural language processing OR Bayesian network OR Bayesian learning OR random forest OR evolutionary algorithms OR multiagent system AND sepsis. The collected records contained essential attributes, including publication date, authorship, institutional affiliation, geographic origin, and cited references. This data set served as the foundation for our subsequent analytical investigations.

The titles and abstracts underwent initial screening by 2 independent authors (MW and TNP). If there was uncertainty from one reviewer regarding whether the article met the inclusion criteria, it was included for a thorough full-text review. Following this, both authors independently assessed the full text, and any differences in opinion were resolved through consensus with the research team. We considered studies for inclusion if they met the following criteria: (1) they were written in English, and (2) they applied AI models in the context of sepsis. In this study’s screening process, we included research or review articles published in peer-reviewed journals, conference proceedings, reviews, and early access articles. We excluded studies if they were published as letters, editorials, book chapters, or books.

To ensure compatibility with Bibliometrix and VOSviewer [12], we saved the data in the “**.txt” format, a format recognized by both tools for conducting analyses. Our data set encompasses a comprehensive range of information, including titles, list of authors, name of countries, list of institutions, abstracts, keywords, name of journals, and publication dates.

Bibliometrix and VOSviewer tools were used to uncover the knowledge structure, most influential countries, research hot spots, and productive authors, along with various bibliometric insights. The processed data were uploaded into these bibliometric tools, and analysis was conducted based on the information included within the data documents [13]. Afterward, we generated network maps among journals, authors, countries, and institutions, where individual points symbolized authors, countries, or institutions. Moreover, connected lines in the network maps depicted the relationships between these entities. Larger points and more robust lines indicated a higher number of articles and more substantial collaborative relationships, respectively [14,15].

We computed the annual growth rate of publications. The annual publication count, annual growth, and average growth rate of publications were determined through the following methods:

Where N is the total number of articles in the current year, and N_k–1 is the total number of articles in the previous year.

Furthermore, we conducted an analysis of publication trends based on the following criteria: the top 10 most prolific countries, institutions, journals, authors, and studies in this area. The rankings of countries, institutions, journals, and authors were determined based on the number of articles.

The initial search yielded 327 articles focused on the application of AI in sepsis. After applying predefined inclusion criteria, 68 articles were deleted, leaving 259 articles for the final analysis (Figure 1).

Over time, there has been a substantial rise in the number of publications in this field. Notably, the yearly publication number increased from just 2 articles in 2014 to 72 articles in 2022. It is important to note that before 2018, the yearly publication count did not cross 10 articles. The calculated annual growth rate was found to be 44.81% (Figure 2).

A total of 122 journals published articles on the application of AI in sepsis. Among them, the Critical Care Medicine journal was the most productive, having published 6.6% (17/122) of articles in this domain (Table 1). Frontiers in Medicine, Scientific Reports, and the American Journal of Respiratory and Critical Care Medicine were in the second, third, and fourth positions, publishing 17, 11, and 7 articles, respectively, on this topic. However, the top 10 journals published 86 articles, accounting for 33.2% (86/259) of all publications in this area.

This study revealed that researchers from 73 countries and regions engaged in research on these subjects and published their work in various international peer-reviewed journals. Out of the total 259 articles, the United States made the most substantial contribution with 103 publications (39.8%), followed by China with 83 publications (32%), United Kingdom with 14 publications (5.4%), and Taiwan with 12 publications (4.6%) (Table 2).

Table 3 shows the top 10 most productive institutes that used AI applications in sepsis. The University of California system (22/259 articles, 8.5%) ranked first among all research institutions, followed by Emory University (10/259 articles, 3.9%), Harvard University (10/259 articles, 3.9%), and Central South University (8/259 articles, 3.1%).

Figure 3 shows the institution cooperation network of 117 institutions that published at least 1 article.

A total of 259 articles were authored by 1444 individuals, each of whom had at least 1 article to their name. In Table 4, we present the top 10 most prolific authors who conducted and published research in these fields. Ritankar Das secured the top position with 8 articles, closely followed by Chris Barton (6 articles), Rishikesan Kamaleswaran (6 articles), and Suchi Saria (6 articles).

Our analysis shows that 1444 authors have published at least 1 article. The largest set of associated authors consisted of 20 authors in 3 clusters (Figure 4).

Table 5 shows the top 10 most frequently cited publications. The publication that received the most citations was by Komorowski et al [16], titled “The Artificial Intelligence Clinician learns optimal treatment strategies for sepsis in intensive care,” published in Nature Medicine in 2018 and received a total of 408 citations as of August 31, 2023.

Keywords encapsulate the central themes within a publication and are ideal for examining interconnected areas of research. In this study, we performed co-occurrence analysis to pinpoint the prominent research focal points in the field of AI application in sepsis research, using the top 100 keywords. The extraction and clustering of these top 100 keywords were performed using VOSviewer.

Figure 5 illustrates our use of VOSviewer to create a visual network map, consisting of 6 clusters based on the co-occurrence of the top 100 keywords. The core of this visualization network map is occupied by the following keywords: sepsis (n=138), machine learning (n=122), artificial intelligence (n=35), and deep learning (n=20).

This study investigated the citation analyses of AI models in sepsis research, scrutinizing the publication patterns related to their application. This study reports a noteworthy upswing in interest in this subject over the past decade, particularly increasing from 2018 to 2022.

The substantial increase in the use of AI in health care research is noteworthy. For instance, there was an 88.88% increase in the use of AI in health care research after 2012, with only 10 countries contributing to over 96% of the studies [25]. Our findings also indicate an annual growth rate of 44.81% in the use of AI in sepsis-related research. If this growth rate continues in the future, we can anticipate the publication volume in the domain of AI in sepsis-related research doubling approximately every 5 years. This increased number of publications indicates advancements, improved functionality, and the overall progress of AI in sepsis, especially when compared to other application areas.

The number of publications exhibits geographical disparities [26-28]. Recent research indicates a rise in the application of AI in health care research, particularly in developed countries [12,29,30]. These nations are investing increased funds into AI research and the development of AI tools to improve health outcomes. This study aligns with these findings, highlighting an increased number of publications originating from developed countries. However, it underscores the importance of researchers from developing nations stepping forward to contribute toward achieving ultimate health goals. Effective collaboration among clinical experts, AI model developers, and health care providers is essential to addressing the challenges at hand [31].

Selecting the appropriate journals for publication can be a complex decision [32]. Authors take various factors into account when submitting, such as the accessibility of open-access journals and the higher impact factor associated with certain subscription journals [33]. Some authors prefer open-access journals for their widespread availability after publication and the potential for a higher number of citations [34]. Conversely, subscription journals from reputable publishers attract attention due to their high impact factors [35,36]. New researchers often struggle with the dilemma of where to submit their work. This study shows that authors frequently weigh both open-access and higher-impact–factor journals when publishing research related to the application of AI in sepsis.

This study aimed primarily to highlight the most commonly used data sets and algorithms in the current literature. The majority of studies used the Multiparameter Intelligent Monitoring in Intensive Care III (MIMIC-III) data set and supervised machine learning models [37-39]. Notably, Komorowski et al [16] introduced a reinforcement learning model to predict sepsis in patients, demonstrating average reliability levels higher than those of human clinicians. While many studies used extensive training and testing data sets, the majority focused on single-centered data [39]. To apply these models in real-world clinical settings, external validation becomes necessary.

This study has several strengths. First, it is the first comprehensive bibliometric analysis that sheds light on the research trends of the application of AI in sepsis, illustrating how this field has evolved. Second, this study gauges productivity in terms of sources, authors, institutions, and countries, while also visualizing word trends. This provides novel and in-depth insights for both researchers and practitioners. This study also has some limitations to address. First, we only collected relevant publications from the Web of Science, a widely used academic resource, for bibliometric analyses [13,40-42]. Nevertheless, using other databases, such as PubMed or Scopus, might have provided slightly varied findings. Second, our inclusion criteria comprised articles published solely in English. However, inclusion of other languages, gray literature, and books might have influenced outcomes, particularly considering diverse cultural perspectives among scholars on the application of AI in sepsis. Finally, relying solely on article titles for the search may pose limitations. However, our aim was to focus on publications specifically addressing the application of AI in sepsis. Therefore, a title screening was deemed more suitable than a broader topic search.

This study aimed to present a comprehensive overview of the use of AI in sepsis through a systematic analysis of existing literature. The findings of this study reveal a noticeable increase in the number of publications over the last 10 years. Until now, developed countries have been the primary contributors in this field. Researchers from developing countries should step forward, leveraging population advantages and core technologies in different regions to foster collaboration.

Leading multidisciplinary science journals, including Frontiers in Medicine, Scientific Reports, and the American Journal of Respiratory and Critical Care Medicine, emerge as key contributors to this topic based on the volume of published articles. As the application of AI in sepsis research continues to rise, this study serves as a valuable resource for researchers seeking direction and opportunities for collaboration.