PDF
Citation
Print
INTRODUCTION
Healthcare 1.0 refers to the traditional model of healthcare delivery that has existed for much of the 20th century and continues to be prevalent in many regions. In the late 1990s, Healthcare 1.0 focused on improving service efficiency and reducing bureaucracy. Healthcare 2.0 was aimed at enhancing efficiency and data exchange among healthcare organizations. Healthcare 3.0 introduced proactive care using electronic medical records (EMRs), big data analytics, and Internet of Things (IoT) wearables, emphasizing preventive treatment [1]. Finally, Healthcare 4.0 marks a shift toward predictive care and patient-centric approaches, using advanced technologies such as the IoT, blockchain, artificial intelligence (AI), and big data analytics to deliver high-quality, cost-effective healthcare [1]. In the era of Healthcare 4.0, the integration of AI, big data analysis, and the IoT is revolutionizing medical diagnostics and laboratory medicine (LM). The use of AI in diagnostic procedures enhances precision and efficiency, reduces human error, and improves patient outcomes [2-4].Similarly, machine learning (ML) offers a means to analyze complex biochemical data rapidly and accurately, and big data analysis allows for more personalized and predictive medicine [5]. The relationship among big data, ML, and AI is foundational to these advancements. Big data provides the extensive datasets necessary for analysis, ML processes the data to identify patterns and provide insights, and AI applies the findings to improve diagnostic and treatment decisions, collectively driving the evolution toward more informed and tailored healthcare solutions [2-5].
The influx of extensive data poses significant challenges in data management, emphasizing the need for robust systems to maintain data integrity and security. The management of sensitive patient information under these circumstances becomes a matter of stringent compliance with data privacy and ethical standards, introducing novel challenges in LM [2]. Consequently, there is a growing need for laboratory professionals skilled in digital literacy, data analysis, and AI-based tools [3, 4]. Educational programs and professional training are being adapted to include these skills to equip professionals for Healthcare 4.0 [2, 6, 7].
The Evidence-Based Laboratory Medicine (EBLM) Committee of the Korean Society for Laboratory Medicine (KSLM), recognizing the importance of understanding the impact of Healthcare 4.0 technologies in LM, aimed to evaluate the awareness and implementation of these technologies among KSLM members to identify knowledge gaps and prepare for future challenges. The objectives were to assess perceptions and readiness regarding big data, AI, and terminology standards among laboratory professionals in the Healthcare 4.0 era.
METHODS
Study design and participants
A web-based survey was conducted between June 12 and 30, 2023, using an anonymous questionnaire distributed via the Moaform online survey platform (www.moaform.co.kr). To ensure the validity and reliability of the survey results, 12 EBLM Committee members rigorously evaluated the questionnaire for comprehensibility and consistency before it was distributed. Then, KSLM members were invited to partake in the survey through a hyperlink embedded in an e-mail, facilitating direct access to the questionnaire. The study protocol and design were approved by the Institutional Review Board of the Hallym University Dongtan Sacred Heart Hospital, Gyeonggi, Korea (approval No.: 2023-03-005).
Questionnaire content
The survey comprised 36 questions covering demographic information (seven questions), big data (10 questions), and AI (19 questions).
The seven demographic questions included sex, age, employment type, years of practice in LM, workplace type and location, and the most important technologies introduced in LM with the advent of Healthcare 4.0.
Ten questions were addressed to clinical pathologists to assess their understanding of, interest in, and research experience with big data analysis and the statistical tools and programming languages used in big data analysis. The questionnaire included questions pertaining to the awareness of the common data model (CDM) and clinical data warehouse (CDW) and the availability thereof in the institution, and questions regarding the recognition of the Federated E-Health Big Data for Evidence Renovation Network (FeederNet), which is an environment for federated learning constructed based on the Observational Medical Outcomes Partnership (OMOP) CDM used in the Observational Health Data Sciences and Informatics [8].
Nineteen questions focused on the participants’ knowledge of, attitude regarding, and experience with AI, as well as their interest in using AI tools in their work, aiming to assess current AI usage and views on the use of AI in the next five years. The questions on AI usage assessed whether the participant’s laboratory workplace is currently integrating AI technologies, how the adoption of advanced technologies affects the workplace, and in which divisions AI is considered particularly useful. We asked for opinions on whether AI would improve the quality of the laboratory and on the legal liability for AI test results and decision-making.
RESULTS
Participants
In total, 182 participants (17.9%) of 1,017 KSLM members completed the survey over a three-week period. Demographic data are provided in Supplemental Data Table S1.
According to the responses to single-selection questions, AI (31.9%) and big data (31.3%) were regarded as the most important technologies in the era of Healthcare 4.0 in LM, followed by personalized medicine (24%). Three-dimensional (3D) printing, augmented reality (AR), virtual reality (VR), IoT, and surgical robots were less emphasized (Fig. 1).
Knowledge of the fundamental concept of big data
Among the 182 respondents, 134 (73.6%) acknowledged being somewhat familiar with the concept of big data, and 11 (6.0%) professed a comprehensive understanding. Thirty-six respondents (19.8%) reported having a limited grasp of the concept, and one (0.5%) reported a complete lack of knowledge on the topic.
Use of big data analytics in LM
A survey on big data research experience among laboratory professionals revealed that the majority (80.8%, N=147) were interested in but had not yet conducted big data analysis research. Only 10.4% (N=19) had conducted research in this field, 36.8% of whom had published their work. Additionally, 7.7% were involved in ongoing big data analysis research, whereas a small portion (3.3%, N=6) showed no interest in the subject. Regarding the use of statistical methods for big data analytics, 22.0% of respondents reported being able to use data analytics tools. According to the responses to the multiple-choice question regarding preferences for statistical software, the majority of participants (75.0%) preferred using R programming (Fig. 2A). Among programming languages, R was the most commonly used by 77.5% of respondents, followed by Python (67.5%), Visual Basic (25.0%), structured query language (SQL) (17.5%), and C (17.5%) (Fig. 2B).
State of the healthcare information technology
In 2020, the Korean Ministry of Health and Welfare launched a project to enhance medical data collection, processing, and interoperability [9]. Forty-eight respondents (26.4%) reported that their institution had participated in the project, whereas 73 (40.1%) reported non-participation. Sixty-one respondents (33.5%) were uncertain whether their institution had participated.
To facilitate the reuse of health data and ensure interoperability, medical information systems, such as CDWs and CDMs, have been developed [9, 10]. Approximately 60% of respondents were unaware of CDMs, whereas 40% reported varying degrees of awareness. When asked about the establishment of a CDW or CDM at their institution, 14.8% reported the presence of a CDW, and 7.7% reported CDM implementation. Notably, 11.5% indicated that their institution has both a CDW and a CDM. Conversely, 27.5% stated the absence of both systems in their institution, and 38.5% were unsure of their availability.
Nationwide big data platform and center
The Health Insurance Review and Assessment Service (HIRA) releases data via the Healthcare Bigdata Hub platform, which provides public datasets, medical big data analysis tools, and statistics tools for the analysis of diseases and medicines [11]. Only nine respondents (4.9%) reported having used Healthcare Bigdata Hub data for research. Forty-one (22.5%) had only visited the portal, 79 (43.4%) were aware of it but had not visited it, and 53 (29.1%) were unfamiliar with it. Regarding the 2021 memorandum of understanding between the KSLM and the big data research department of the National Health Insurance Service, 36.8% of respondents were informed about its existence, whereas 63.2% were not. FeederNet, a clinical data analysis platform in Korea established in 2019 [8], was familiar to 43 (23.6%) respondents. Among them, 9.3% had engaged in analysis using real-world big datasets from the OMOP CDM, whereas 90.7% had not actively participated in such research. A significant portion of respondents (139, 76%) lacked awareness of this nationwide network. One hundred thirty respondents (71.4%) expressed willingness to participate in future big data research by KSLM.
Awareness and perceptions of AI
Most described their knowledge of AI as somewhat (73.6%) or very well-informed (4.9%), whereas 39 (21.4%) respondents reported not being well-informed.
Research and publications on AI
One hundred forty-eight of 182 respondents (81.3%) were interested in but had not yet published AI research. Twenty-three respondents (12.6%) had published AI research papers, and 16 (8.8%) were currently engaged in AI research. Eleven respondents (6.0%) reported no interest in AI research.
Use of AI in LM
Table 1 outlines the impact of AI on current and future LM work. Fig. 3 illustrates the respondents’ perceptions of the potential utility of AI within various LM divisions. AI was considered most useful in molecular genetics (28.7%) and diagnostic hematology (27.5%), followed by laboratory management (24.2%). AI was perceived to be less applicable in clinical chemistry and other areas. Two respondents highlighted the role of AI in test result verification and point-of-care testing. Concerning AI adoption, the vast majority (95.6%) were open to using AI applications in their work, whereas a small fraction (4.4%) expressed no such intention. Regarding AI training, 79.7% recognized the need for and were willing to pursue specialized AI training, 17.6% preferred learning from AI experts, and 2.7% expected only basic training.
Impact of AI on clinical pathologists or LM residents
Table 2 summarizes the views on the impact of AI on clinical pathologists and LM residents. More than half of the respondents (56.0%) anticipated an expansion of their roles because of AI, 27.5% expected no change, and 16.5% expected their roles to become redundant. Regarding the impact of AI on the current numbers of clinical pathologists and LM residents, 52.2% predicted no change, whereas 26.4% expected an increase and 21.4% a decrease. Opinions on the impact of AI on residency training duration also varied: 81.9% expected no effect, 12.1% anticipated a need for extended training, and 6.0% suggested a potential reduction of the conventional 4-yr training duration. Regarding medical education, 46.8% of clinical pathologists believed AI would increase their teaching responsibilities, 45.6% predicted no significant change, and 7.1% expected a decrease in educational duties.
Impact of AI on laboratory test reporting
Table 3 presents the views on the impact of AI on laboratory test reporting and liability. Most respondents (80.2%) were optimistic about AI’s potential to improve laboratory test quality and accuracy, whereas 20% remained neutral or cautious. With AI poised to contribute to laboratory reporting and EMR integration, 45.6% of professionals expressed concerns about potential unforeseen issues with AI-generated reports, whereas 40.7% would not be concerned if AI would achieve expert-level interpretation. Regarding liability for AI diagnostic errors, 54.4% advocated shared legal responsibility between AI developers and physicians, 33.5% found that physicians should be fully accountable for final decisions, and 12.1% believed that manufacturers should compensate for any patient harm resulting from defective AI products or designs.
DISCUSSION
This survey was the first of its kind, targeting KSLM members in Korea, specifically addressing Healthcare 4.0. The respondents considered AI the most vital in LM, followed by big data and personalized medicine. In contrast, a previous report highlighted IoT as the most promising healthcare technology because of 5G and data-sharing capabilities, followed closely by AI in the second position and robotics and digital manufacturing, including 3D printing, in the third and fourth positions, respectively [12]. ML, a subset of AI, is already implemented or is likely to be used to enhance laboratory workflows, provide decision support, facilitate diagnostic and predictive applications, and integrate laboratory data [13]. Our findings revealed some differences in the ranking of the importance of Healthcare 4.0 technologies.
Most survey respondents were somewhat familiar with big data, with varying levels of understanding of and interest in big data analysis. Only a small percentage were proficient in using statistical software tools, indicating a gap in skills related to statistical analysis and programming language fluency. R programming was the most preferred language for statistical analysis, which is attributed to its open-source nature and diverse packages. In 2020, the most popular programming language for big data projects was Python, followed by C++ and JavaScript [14]. In a 2021 survey regarding tools for data scientists and analytics professionals, the order of preference was Python>SAS>R [15]. This suggests that preferences differ based on the industry, study area, and intended use, as well as according to the advantages and disadvantages of each tool.
In the era of big data, the integration of laboratory data into healthcare research presents both unprecedented opportunities and significant challenges. With the rapid expansion of AI applications and the utilization of big data in healthcare, laboratory results have become a crucial component of the data landscape. However, the assurance of data quality of laboratory results is far from straightforward [16], particularly given that identical samples can yield significantly biased results when tested across different laboratories despite ongoing standardization and harmonization efforts [17-19]. Correcting biased results is essential to minimize laboratory errors, enhance patient safety, and reduce costs. Significant biases, both statistical and medical, must be identified and rectified [20]. Therefore, clinical pathologists bear the responsibility to guarantee the accuracy and reliability of the data that form the basis of research and clinical decisions [21]. Research efforts focusing on the development of models for evaluating real-world laboratory results for big data research are ongoing [22].
With the increasing availability of digital health records and data analytics techniques, data-driven approaches to healthcare have offered the potential to improve clinical decision support and the healthcare industry as a whole [23]. In line with current trends, in 2020, the Medical Data-Driven Hospital Support Project was initiated in Korea [9]. A significant portion of respondents appeared to have limited knowledge of their hospital’s involvement in the project or the availability of a CDW or CDM within their institution. Familiarity with platforms such as the Healthcare Bigdata Hub and FeederNet was also limited.
The role of AI in healthcare and medical education, data interpretation, and reporting in LM is growing. The use of AI, particularly deep learning in radiology, has significantly increased, with studies growing from 14 in 2015 to 237 in 2019 [24, 25]. However, its application in routine LM practice remains limited [6, 26]. Clinical pathologists are essential in the development, training, and evaluation of AI algorithms.
Most respondents had a general understanding of AI, which was more advanced than that of medical students [27, 28], and anticipated its adoption in LM, particularly for image analysis and data interpretation. AI has shown effectiveness in diagnostic imaging [29, 30] and is used in data analysis. AI is expected to significantly transform healthcare in the future [29, 31-33]. The respondents were keen on AI adoption, emphasizing the need for education and skill development. Data analysis was viewed as the most beneficial AI application in LM, with molecular genetics and diagnostic hematology regarded as the most suitable areas for AI adoption. While AI is rapidly being adopted in various sectors, its integration in medicine poses technical and ethical challenges.
More than half of the respondents were open to AI supporting rather than replacing clinical pathologists, whereas 16.5% expressed concerns about potential job loss. Studies by Syed and Al-Rawi [34] and Ansari, et al. [35] revealed varying beliefs about AI replacing human roles in healthcare, with 17.8% of healthcare students and 85% of radiologists anticipating an impact of AI on jobs. Opinions on AI replacing humans differ according to education, training, and medical specialties. While some specialists view AI as influencing medical education, others do not foresee major changes in educational responsibilities. Adequate education is needed to address the healthcare risks associated with AI [36].
A significant portion of the respondents had a positive view of AI’s potential to improve quality in LM, expecting it to enhance interpretation accuracy and reduce diagnostic errors. However, concerns remained about technical challenges and unpredictability. Most respondents agreed that AI could improve quality across different phases of laboratory testing but were divided on its impact on patient diagnosis and treatment. Legal liability associated with AI-based test results is a significant issue debated by programmers, manufacturers, hospitals, governments, and users. Most respondents perceived that as users, they bear legal responsibility for AI-related medical errors, with two-thirds believing that manufacturers and users share responsibility. A minority of respondents argued for manufacturer liability in cases of defective AI products. The responsibility for AI-related errors remains a complex topic without a definitive answer. Opinions vary among physicians, manufacturers, and patients. Khullar, et al. [37] reported that physicians tend to believe that manufacturers and healthcare organizations are responsible, whereas recent studies suggested manufacturer liability if AI tools are appropriately validated and meet labeling requirements [38, 39].
The adoption of new technologies in the medical field is a complex process that involves not only technical aspects but also acceptance by healthcare professionals and patients [40]. Similarly, the integration of AI and ML into LM is a multifaceted and complex process. There is a critical need to understand human factors that influence technology adoption in LM, such as perceived utility, usability, and the readiness of the LM community to adapt to changes. The successful incorporation of advanced technologies in LM necessitates a balanced approach that considers both the technical prowess of these technologies and the human aspects crucial in this transformational phase. Paranjape, et al. [7] identified a significant AI knowledge gap in the medical community, underscoring the importance of continuous information updates and structured educational programs in LM, enabling laboratory professionals to stay informed and skilled in using AI and ML effectively in their work.
One limitation of our study was the low level of survey participation (17.9%), which may have led to a bias toward those with at least some awareness of AI and LM, potentially presenting an overly optimistic view. Second, the majority of survey respondents were under the age of 50 yrs and had less than 15 yrs of professional experience (Supplemental Data Table S1) and, therefore, may not fully represent the opinions within KSLM, suggesting that the findings may not include the perspectives of the entire membership.
In conclusion, this study was the first to examine how KSLM members view big data, AI, and terminology standards in LM. Our observations revealed that AI is considered a crucial tool in LM, particularly in areas such as molecular genetics and diagnostic hematology. Further, our findings showed that integrating AI into healthcare is complex because of technical and ethical issues, which led to varied opinions on its effect on jobs and training. This emphasizes the need for a well-rounded approach to adopting new technologies, balancing technical aspects with human factors. Finally, our findings highlight the importance of ongoing education and information updates to ensure that laboratory professionals can effectively use AI and ML in their work.
XML Download