ChatGPT in Medicine: Ark of the Covenant or Pandora’s Box? Present status and future perspectives: how, what, who and where?

Background:
ChatGPT, an artificial intelligence chatbot developed by OpenAI has ushered in a new era of possibilities, offering exciting opportunities for natural language processing in healthcare; it marks a clear paradigm shift in academic research, patient care, and clinical communication. However, a hurried and uninformed implementation of ChatGPT in healthcare settings can have unforeseen results.
Objective:
The objective of this study is to discuss and provide a comprehensive summary of the existing literature on ChatGPT based on insights derived from different research teams across diverse disciplines, focusing on the main applications in medicine as conceived by authors who have published on the subject. Furthermore, we seek to address potential concerns associated with the integration of ChatGPT in the medical field, while simultaneously opening up unexplored paths.
Methods:
The corresponding author selected a group of authors through an initial screening of papers published on ChatGPT from various databases. Based on the replies and their level of cooperation, a dozen authors were chosen for the study and included based on their cooperation and geographic representation. We conducted a comprehensive search across databases, including Web of Science, PubMed, and Scopus. The study followed the PRISMA guidelines for screening and selection. Analysis was performed on the selected articles (n=223), categorizing them based on article type, specialty, and specific role discussed. Descriptive statistics were calculated, and bibliometric analyses were conducted using VOSviewer.
Results:
We identified 223 peer-reviewed articles. Writing assistance emerged as the most widely discussed role of ChatGPT. The rest of the roles included: general use (16%, n=35), education tool (17%, n=37), solution and discussion of clinical vignettes (15%, n=34), valuing and predicting future impact (7%, n=15), information source (5%, n=12), and as a discussion forum about ethical issues (6%, n=14). To detect any difference in usage across continents, a chi-square test for distribution of categorical variables was carried out and yielded a p-value of P=0.56, in article type. Regarding specialty and the role of ChatGPT, the p-values were P=0.86 and P=0.87, respectively. We found no significant association between a specific geographic location and the distribution of any of the variables.
Conclusions:
This study provides a comprehensive overview of the literature on ChatGPT, revealing its diverse range of applications as observed across different disciplines. By examining the concerns surrounding the integration of ChatGPT in medicine, we underscore the importance of responsible implementation and highlight the need for ongoing research and development in this area. Ultimately, this work opens new perspectives for the detection of unexplored paths and the realization of ChatGPT's potential in enhancing various domains, including medicine.