General information

Since the release of ChatGPT in November 2022, AIGC’s research in the medical field has shown exponential growth. In this study, a bibliometric analysis of the research on AIGC application in healthcare has been conducted. The main findings are represented in the following aspects.

The United States holds a prominent position in the research of AIGC’s application in healthcare, with 1421 papers accounting for 41% of the total publications. Compared with other countries, the United States has the highest node centrality (0.51) in the network, which is positively correlated with its highest share of global GDP. This indicates that countries with higher economic levels are more likely to receive funding support, thereby promoting the integration of artificial intelligence and medical research. In addition, the top 10 countries in terms of publications account for over 97% of the total publications, revealing the technological dependence of peripheral countries/regions on core regions, and the one-way nature of knowledge flow.

The leadership position of the United States in the application of AIGC in healthcare research was similarly demonstrated in institutional analysis. Among the top 10 institutions, 80% are located in the United States, while the other 20% are based in two different countries. The top 10 institutions account for 27% of the total literature, reflecting the relative concentration and certain monopoly of top institutions in academic output.

The top 10 authors in terms of publications each have contributed more than 10 papers, while over 100 other authors have each produced less than 10 papers. This phenomenon precisely conforms to the inverse square distribution characteristics described by Lotka’s law: most scientists only publish a few papers, while a very small number of scientists have a very large number of papers.

The number of papers published in the top 10 journals accounts for 19% of the total number of publications, which is in line with the core journal characteristics revealed by Bradford’s Law, that is, a few journals dominate the research dynamics in this field in terms of output. In addition, the Nature (IF = 50.5, Q1) has the highest centrality (0.32) with only 4 publications, reflecting the elite characteristics of ‘high impact, low output’, where a few core journals dominate in knowledge dissemination.

Upon analyzing highly cited literature, it was found that only 16 papers have been cited over 200 times, while nearly 2,500 papers have been cited no more than 5 times. This phenomenon aligns with the unequal distribution of citation counts as described by Price’s Law, which posits that a minority of papers receive far more citations than the majority.

Research hotspots

Our study utilized highly cited literature and keyword analysis to identify active and emerging areas of AIGC’s application in healthcare research. As mentioned earlier, frequently occurring keywords in specific fields usually point to research hotspots [31]. In addition, by analyzing highly cited literature, we can determine which publications play a central role in academic discourse, typically representing important research findings or foundational theories in the field. Therefore, based on the high frequency keyword statistics and clustering, combined with highly cited literature analysis and abstract reading of 3411 articles, the research topic of this study is summarized into the following four dimensions for review.

Impact analysis of aigc’s applications in healthcare

The emergence of AIGC has catalyzed revolutionary changes across various industries, including the healthcare field. Prior to the AIGC application in the healthcare field, it is crucial to delve into its impacts on the healthcare field and understand the functionalities and limitations of AIGC tools. This research aids in comprehensively understanding how AIGC fosters innovation and development in the healthcare industry, as shown in Table 7.

Table 7 The potential and risk of aigc’s application in healthcarePotentialsPatient consultation and management

For patients, especially those who do not want to seek help directly from a doctor due to privacy protection [32], AIGC can act as a virtual healthcare staff to provide round-the-clock, low-cost healthcare information and advice based on patients’ inquiries [33] (e.g., customized prevention strategies [34], treatment plans, etc.), thereby enhancing personalized patient management and having great potential to improve lower levels of health knowledge due to language barriers, cultural customs, information access obstacles [35], and so on.

Patient communication and education

Patients have differences in health literacy level and native languages, which can make it difficult for them to understand specific medical reports or patient education materials. AIGC can deconstruct complex medical terminology into easily understandable explanations [36] and provide quick and accurate translations in multiple languages [37] to match patients’ health literacy levels and native languages, bridging the gap in patient-doctor communication [38] and providing valuable assistance to patients. AIGC demonstrates good communication potential in terms of clinical accuracy [39] and clarity of expression [40].

Clinical decision support

Humans could hinder their decision-making process due to cognitive biases such as prejudice or fatigue, but AI can alleviate these biases and improve the accuracy of decision-making [41]. AIGC can extract valuable features and information and organize its utilization by analyzing electronic health records, medical literature, clinical guidelines, etc., assisting physicians in providing support and recommendations for the most likely differential diagnosis [42], and ultimately providing personalized treatment options for patients [5, 43].

Streamlining the clinical workflow

The writing and management of medical records is a tedious and time-consuming process for healthcare providers, and errors in records are also common [44]. Studies indicate that documentation and management requirements occupy approximately 25% of a clinician’s workday [37]. AIGC can assist in completing these administrative tasks, such as writing clinical trials or patient care records, scheduling appointments, streamlining patient admissions procedures, etc [33]., to reduce the burden on healthcare staff, improve the accuracy of records, thereby reducing healthcare costs and saving healthcare providers’ and patients’ time [42].

Teaching and research assistance

AIGC provides invaluable support throughout the scientific research process, including posing questions, formulating hypotheses, conducting literature review, conducting meta-analyses, interpreting datasets, suggesting headings, writing drafts, improving writing, correcting grammar, translating language, etc [30, 38, 42]., helping researchers focus their work on the most impactful parts. Meanwhile, AIGC’s ability to generate code provides technical skills for medical researchers who lack extensive programming expertise to test hypotheses and improve research efficiency [37]. Furthermore, AIGC can create patient interactions that simulate real-life scenarios, providing learners with a personalized interactive learning environment [45] to practice skills such as communication, assessment, and intervention [46].

Data processing and analysis

AIGC exhibits diverse functions in medical data processing and analysis, significantly improving the efficiency and quality of medical services. By integrating patient complaints, electronic health records, and imaging data, AIGC is able to generate differential diagnosis lists with diagnostic accuracy comparable to traditional clinical methods, particularly in handling complex cases, significantly reducing doctors’ decision-making time [47]. In addition, AIGC has improved the efficiency of semantic analysis and topic extraction of medical record texts by 3–5 times compared to manual methods, and the qualitative analysis results have a consistency of over 85% with human experts [48]. In terms of resource allocation, AIGC achieved a classification accuracy of 92.3% through structured symptom input and risk prediction algorithms. Compared to manual triage, the misjudgment rate was reduced by 17%, effectively optimizing the allocation and utilization of medical resources [49].

RisksQuality of information

The accuracy of AIGC is a crucial constraint. Due to the fact that training data may not contain the latest advances and updated guidelines in medical research [36], or the lack of diversity in training dataset, AIGC may generate controversial or misleading content [30]. These errors can have serious consequences for patient safety, especially in cases of excessive or inappropriate reliance on AIGC [1]. This is particularly concerning in the healthcare field, where accuracy and credibility [50] are critical [51].

Data bias

AIGC responses and outputs are likely to amplify frequent occurrences and suppress rarities, leaving a small group of patients at risk of receiving inaccurate, irrelevant, or biased answers if the data from patients with certain cultural backgrounds or socioeconomic statuses in the original training set are underrepresented or biased [52]. The bias of such outputs may harm marginalized groups, while the unconscious dissemination of biased information by other users while trusting AIGC decisions may further exacerbate social divisions [53]. In the healthcare field, cultural differences and health inequality considerations are particularly crucial [54].

Excessive reliance

Over-reliance on AIGC-generated insights and recommendations in healthcare may limit healthcare professionals’ critical thinking, decision-making and innovation abilities, leading to a decrease in independent thinking in critical decision-making processes [46]. It is important for healthcare professionals to review and validate AIGC-generated outputs based on their expertise and background [55] to determine the accuracy and completeness of the content.

Intellectual property

The content output of AIGC relies on input from a variety of training data, which may pose the risk of violating the intellectual property rights of the authors [56]. For example, using AIGC to obtain medical information for medical research writing raises issues such as plagiarism and lack of originality [57], and when the writing content is misused for academic publications [58], the ensuing debate over the attribution of AIGC authorship arises.

Data security and privacy

The use and capability enhancement of AIGC require the collection, storage, and learning of a large amount of data. If the training data contains sensitive patient information, the possibility of misuse or accidental leakage of this patient-sensitive data becomes a significant risk, with negative consequences including privacy invasion [59], identity theft, etc [33]. Clinical staff or any user interacting with AIGC should be cautious about submitting sensitive information into the system unless the generated material has been moved to a secure environment [60].

Responsible oversight

The internal workings of the AIGC are largely a “black box” [53]. The opacity of this work mechanism limits accountability when the data set produces biased content or misleading information that leads to poor decision-making that raises legal issues of medical error [57]. The question of who bears responsibility raises concerns about human supervision and control of the generated content [56].

Applicability evaluation of aigc’s application in healthcare

Although several studies have explored the potential and risks of AIGC application in healthcare, the lack of quantitative studies has left a degree of ambiguity about the ideal use of AIGC in healthcare. Therefore, researchers evaluate the performance of AIGC in different scenarios in the healthcare field to examine the reliability of the AI system and determine AIGC’s actual application effects in the healthcare field, providing guidance for the subsequent practice of AIGC application, as shown in Table 8.

Table 8 Applicability evaluation of aigc’s application in healthcareParticipation in medical examinations

Researchers from multiple countries have examined the accuracy of AIGC in different medical specialties examinations and the results show that its performance meets or approaches the passing threshold [61], and it demonstrates a certain level of understanding and explanation of answers [62]. Specifically, it exhibits higher accuracy in answering simple and general questions compared to difficult ones [63]. Besides, its correct response rate is higher in single-choice question sets compared to multiple-choice question sets, and higher in short question sets compared to long question sets [64]. It also performed better in descriptive and memorization tasks than in analytical and critical-thinking tasks [65], suggesting limitations of AIGC in handling highly specialized issues requiring advanced judgement and diverse. In addition, AIGC shows higher accuracy and response quality when answering medical exam questions in English compared to other languages [66], indicating its performance may be affected by language differences in conveying information [67].

Answering healthcare questions

Researchers assessed the ability of AIGC to answer common patient queries in clinical medicine and found that while there may be discrepancies with some official information [68], AIGC usually gave accurate, comprehensive, and clear responses [69,70,71], providing appropriate advice to patients. However, as a healthcare consultation tool, AIGC still faces limitations, including difficulties in providing relevant reference materials [72], uncertainty about the timely updates of knowledge database [73], possible differences in diagnostic tests and treatment methods across regions [69], and challenges in offering personalized advice [74]. These factors underscore the importance of users exercising caution when contemplating the healthcare guidance offered by AIGC. Furthermore, there are conflicting conclusions in different studies regarding the repeatability and empathy of AIGC-generated answers. Some studies suggest that AIGC provides repeatable responses to the same questions [25] and responses with empathy comparable to that of physicians [20], while others draw opposite conclusions [75, 76].

Clinical decision diagnosis

Research literature suggests that AIGC can provide accurate, comprehensive, and relevant diagnoses based on patient symptoms, medical history, and test results [77, 78], in line with the literature and current medical knowledge [79]. Furthermore, the ability of AIGC to provide accurate clinical decision-making is enhanced when provided with contextualized information [80] or more targeted questions [81]. In addition, by assessing the consistency between AIGC and expert responses, some literature has found that the AIGC-generated results are similar to manually generated recommendations [82], and there are also conclusions that AIGC often overlooks key factors, leading to inconsistencies with expert clinical decisions [83]. This emphasizes AIGC as a supplementary tool for diagnosis rather than a replacement for experienced healthcare professionals [84].

Generation of medical Documentation

Researchers have found that AIGC has demonstrated excellent capabilities when utilized to generate medical documents such as medical reports, discharge summaries, and informed consent forms [85]. By using commonly used terms instead of medical jargon and refining the original information [86], AIGC provides concise, clear, and comprehensive medical documents with high accuracy [85, 87, 88], enabling patients to better understand relevant information [89], while alleviating the documentation burden on healthcare professionals [90]. However, it is important to note that in the generation process that AIGC may oversimplify during the information integration, leading to information loss [86].

Perspective of healthcare related personnel on AIGC

In addition to analyzing the impact and suitability of AIGC applications in the healthcare field, it is also crucial to understand the perceptions of healthcare relevant personnel on AIGC. Collecting such data helps identify user needs, knowledge gaps, and potential barriers, thereby ensuring responsible and effective integration of AIGC into healthcare practice. Researchers investigated the perceptions of healthcare related personnel on AIGC application in healthcare from three aspects: knowledge, attitude, and practice, as shown in Table 9.

Table 9 Perspective of healthcare related personnel on AIGCKnowledge

The researcher believes that this may be due to the fact that most people are not aware of the advantages of applying AIGC in healthcare [91]. Despite the low level of knowledge about AIGC, survey participants expressed interest in learning more about AIGC [92]. Besides, it was found that the younger generation is more likely and curious to be exposed to AIGC through social media or other digital platforms compared to older adults, and thus the younger generation has a higher level of knowledge about AIGC than older adults [93].

Attitude

The survey results showed that although most healthcare staff and medical students have a positive attitude towards the use of AIGC in healthcare [94], they are also concerned about the dangers of AIGC. Respondents’ attitudes were influenced by multiple factors, such as perceived utility, perceived risk, ease of use, and price [95]. Related Personnel believe that AIGC will be a valuable tool in healthcare, automating repetitive tasks, reducing the burden on humans, and providing more opportunities for creativity [93]. At the same time, respondents’ concerns include inaccurate medical information generated by AIGC, privacy issues associated with data collection, bias, and discrimination in training data [96], the possibility of replacing human roles in medical practice [92], the creation of medical academic misconduct [97], etc. In addition, people with limited knowledge of AIGC are more likely to express anxiety about the application of AIGC to healthcare compared to people with higher levels of AIGC knowledge [