Abstract
Background ChatGPT(Chat Generative Pre-trained Transformer) is an artificial intelligence (AI) based on a natural language processing tool developed by OpenAI (California, USA). This systematic review examines the potential of Chat GPT in diagnosing and treating patients and its contributions to medical research.
Methods In order to locate articles on ChatGPT’s use in clinical practise and medical research, this systematic review used PRISMA standards and conducted database searches across several sources. Selected records were analysed using ChatGPT, which also produced a summary for each article. The resultant word document was transformed to a PDF and handled using ChatPDF. The review looked at topics pertaining to scholarly publishing, clinical practise, and medical research.
Results We reviewed 118 publications. There are difficulties and moral conundrums associated with using ChatGPT in therapeutic settings and medical research. Patient inquiries, note writing, decision-making, trial enrolment, data management, decision support, research support, and patient education are all things that ChatGPT can help with. However, the solutions it provides are frequently inadequate and inconsistent, presenting issues with its originality, privacy, accuracy, bias, and legality. When utilising ChatGPT for academic writings, there are issues with prejudice and plagiarism, and because it lacks human-like characteristics, its authority as an author is called into question.
Conclusions ChatGPT has limitations when used in research and healthcare. Even while it aids in patient treatment, concerns regarding accuracy, authorship, and bias arise. Currently, ChatGPT can serve as a “clinical assistant” and be a huge assistance with research and scholarly writing.
Introduction
ChatGPT(Chat Generative Pre-trained Transformer) is an artificial intelligence (AI) based on a natural language processing tool developed by OpenAI (California, USA). ChatGPT is chat boat based technology. A chatbot is in fact a type of software creates text akin to human-like conversation. ChatGPT has the capacity to respond to follow-up questions, recognise errors, debunk unfounded theories, and turn down inappropriate requests. Large Language Models (LLMs), which are frequently abbreviated as LLMs, are extremely complex deep-learning programmes that are capable of comprehending and producing text in a manner that is strikingly comparable to that of humans. LLMs can recognise, summarise, translate, predict, and create text as well as other sorts of information by using the large knowledge base they have amassed from massive datasets.1, 2
The possible uses of ChatGPT in medicine is currently under intense investigation. ChatGPT is considered to have enormous capability in helping experts with clinical and laboratory diagnosis to planning and execution of medical research. Another significant use of ChatGPT in medical researchers is the creation of virtual assistants to physicians helping them in writing manuscripts in more efficient way. Usage of ChatGPT in medical writing is considered to have associated with several ethical and legal issues. Possible copyright violations, medical-legal issues, and the demand for openness in AI-generated content are a few of these.3–5
In this systematic review we aimed to review published article and explore the potential of ChatGPT in facilitating patient care, medical research and medical writing. We will also focus on ethical issues associated with usage of ChatGPT.
Methods
We performed a systematic review of published articles on ChatGPT. The protocol of the systematic review was registered with PROSPERO (PROSPERO 2023 CRD42023415845).6 Our systematic review was conducted following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines.
Search strategy
We searched four databases, PubMed, Scopus, Embase, and Google Scholar. Our search was aimed at identifying all kinds of articles on ChatGPT and its application in medical research, scholarly and clinical practice, published till 24 May 2023. Articles related to medical education was not considered. The search item that we used was “ChatGPT”. We reviewed all kinds of publications including original articles, reviews, editorial/ commentaries and even letter to the editor describing ChatGPT.
Data extraction
The selection of the papers that were published was done in two steps. Two reviewers (RK and VKP) reviewed the titles and abstracts in the initial phase. Two reviewers (VU and SKC) then examined the entire texts of the chosen papers to determine their eligibility. A third author (SK) settled any differences that arose between the two authors. Two reviewers (RK and VKP) assessed the information available in the included publication for the suitability of the article to be included in the review. Any disagreement between them was resolved by mutual agreement. If a dispute persisted, it was resolved via consultation with a third reviewer (SK).
EndNote 20 web tool (Clarivate Analytics) was used to handle duplicate records. This process was carried out by two reviewers independently (RK and VKP). Any issue that arose was resolved with a discussion with another reviewer. The number of retrieved and assessed records at each stage was provided in the form of a PRISMA flow chart. EndNote 20 (Clarivate Analytics) was used to make a PRISMA flow chart.
Quality assessment
Quality assessment was not done.
Data analysis
ChatGPT was extensively used for analysing the selected records and writing this manuscript. A table was made with six columns (First author/sole author, country of origin, status of peer review (peer-reviewed or preprint), title of the paper and short point wise summary of full text. Short point wise summary of full text of each and every article was created with the help of ChatGPT. The voluminous word file was then converted to a pdf file and was processed with the sister software “ ChatPDF” (OpenAI, California, USA available at https://www.chatpdf.com/). Following questions were asked from ChatPDF.
What are potential role of ChatGPT in medical writing and research?
What could be the role of ChatGPT in clinical practice?
What are ethical issues associated with paper writing?
Can ChatGPT be an author?
Can ChatGPT write text in good English and free of plagiarism?
Role of ChatGPT so far in neurological disorders related clinical practice and research.
Effectiveness and efficiency of ChatGPT in medical research and clinical settings
Potential benefits and limitations of ChatGPT in medical research and clinical applications
The ethical implications of using ChatGPT in medical research and clinical practice
Identify the gaps in the current research on ChatGPT and suggest areas for further investigation.
Provide insights into the potential future applications of ChatGPT in medical research and clinical practice
Recommendations for researchers, clinicians, and policymakers on the use of ChatGPT in medical research and clinical practice
All the responses were compiled in a word file.
Results
Our data collection followed PRISMA guidelines. (Additional file 1: PRISMA checklist) The PRISMA flowchart for our systematic review is shown in Figure 1. We reviewed 118 publications. ChatGPT related publications are available from across the globe. There were 33 original articles and rest were commentary/ editorial, review articles, research letters or letter to the editors. Out of 118 articles, 18 articles were available as preprint only. Summaries of 118 articles and answers to 12 questions have been provided in form of tables. (Table-1 and Table-2)
The study’s PRISMA flow diagram shows how articles are selected for this systematic review.
Summaries of full text of 118 selected article. Summaries were created with the help of ChatGPT. PROMPT, that given was “Provide point wise summary” .
Summaries of full text of 118 selected article were uploaded in pdf format on ChatPDF, subsequently 12 questions were asked. Questions and respective answers given by ChatPDF are listed in this table.
Discussion
We looked into two main uses of ChatGPT: in healthcare settings and for medical writing and research. We studied 118 articles - most were opinion pieces, commentaries, and reviews. Another group, Ruksakulpiwat et al, also did a similar study. They analyzed six articles out of 114 that met their criteria. These articles covered a variety of ways to use ChatGPT, such as finding new drugs, writing literature reviews, improving medical reports, providing medical info, bettering research methods, analyzing data, and personalizing medicine.7
Levin et al, on the other hand, conducted an analysis of the first batch of publications about ChatGPT. They found 42 articles published in 26 journals in the 69 days after ChatGPT was launched. Only one was a research article. The rest were mostly editorials and news pieces. Five publications focused on studies on ChatGPT. There were no articles on its use in Obstetrics and Gynecology.In terms of where these articles were published, Nature was the top journal. Radiology and Lancet Digital Health came next. The articles mostly discussed the quality of ChatGPT’s scientific writing, its features, and its performance. Some also talked about who should get credit for the work and ethical concerns. Interestingly, when comparing the articles that described a study to the others, the average impact factor (a measure of the influence of a journal) was significantly lower for the study articles.8
In our review, we identified several potential advantages of using ChatGPT in the medical field. It appears to enhance productivity and expedite research workflows by aiding in data organization, assisting in the selection of trial candidates, and supporting overall research activities. Furthermore, ChatGPT’s capacity to review manuscripts and contribute to editing may potentiate the efficiency of academic publishing. Beyond the scope of research, it could also prove beneficial for patient education, fostering scientific exploration, and shaping clinical decision-making. However, we also need to consider certain limitations and ethical concerns associated with the use of ChatGPT. The model, as sophisticated as it is, lacks the capability to offer comprehensive diagnoses and cannot replace the human qualities inherent to medical practice. Ethical issues also arise, specifically in relation to potential biases in the machine learning model and potential breaches of privacy. Moreover, while ChatGPT can process and generate information, it might not exhibit the level of originality, creativity, and critical thinking that are often required in the medical field. However, the use of ChatGPT in producing scholarly articles is raising questions in the academic publishing. While these tools can greatly enhance the clarity and fluency of written material, it is crucial that human oversight is maintained throughout the process. This is because AI can potentially produce content that is authoritative-sounding, yet it might be inaccurate, incomplete, or biased. Incorrect GPT-4 responses, known as “hallucinations,” can be harmful, particularly in the field of medicine. Therefore, it is essential to check or validate GPT-4’s output. ChatGPT can generate references to made-up research publications. Therefore, authors must thoroughly check and modify the output of these tools. Furthermore, it is not appropriate to recognize AI or AI-assisted tools as authors or co-authors in the by-line of publications. Instead, their use should be transparently acknowledged within the manuscript. For example, according to Elsevier’s policy on AI for authors, the responsibility and accountability for the work ultimately still lie with the human authors, despite any technological assistance they may have received.9-11
In conclusion, ChatGPT has a great potential. Its full potentials are still evolving. ChatGPT as a source of information can not be trusted, many ethical issues are associated with it. Certainly, ChatGPT can be credited with authorship. However, ChatGPT is certainly a good clinical assistant. ChatGPT is nowhere near to replace human brain.
Data Availability
All data produced in the present work are contained in the manuscript.
Conflict of Interest
All authors have no conflict of interest to report.
Human/Animal Studies informed consent statement
No human or animal subjects were involved.
Financial support
None
Acknowledgment
The concept, data collection analysis, writing, and reporting of this article were solely done by authors. ChatGPT was extensively utilized as mentioned in the methods section.
References
References
- 1.
- 2.
- 3.
- 4.
- 5.
- 6.
- 7.
- 8.
- 9.
- 10.
- 11.
- 12.
- 13.
- 14.
- 15.
- 16.
- 17.
- 18.
- 19.
- 20.
- 21.
- 22.
- 23.
- 24.
- 25.
- 26.
- 27.
- 28.
- 29.
- 30.
- 31.
- 32.
- 33.
- 34.
- 35.
- 36.
- 37.
- 38.
- 39.
- 40.
- 41.
- 42.
- 43.
- 44.
- 45.
- 46.
- 47.
- 48.
- 49.
- 50.
- 51.
- 52.
- 53.
- 54.
- 55.
- 56.
- 57.
- 58.
- 59.
- 60.
- 61.
- 62.
- 63.
- 64.
- 65.
- 66.
- 67.
- 68.
- 69.
- 70.
- 71.
- 72.
- 73.
- 74.
- 75.
- 76.
- 77.
- 78.
- 79.
- 80.
- 81.
- 82.
- 83.
- 84.
- 85.
- 86.
- 87.
- 88.
- 89.
- 90.
- 91.
- 92.
- 93.
- 94.
- 95.
- 96.
- 97.
- 98.
- 99.
- 100.
- 101.
- 102.
- 103.
- 104.
- 105.
- 106.
- 107.
- 108.
- 109.
- 110.
- 111.
- 112.
- 113.
- 114.
- 115.
- 116.
- 117.
- 118.
