A thorough examination of ChatGPT-3.5 potential applications in medical writing: A preliminary study

Abstract

Effective communication of scientific knowledge plays a crucial role in the advancement of medical research and health care. Technological advancements have introduced large language models such as Chat Generative Pre-Trained Transformer (ChatGPT), powered by artificial intelligence (AI), which has already shown promise in revolutionizing medical writing. This study aimed to conduct a detailed evaluation of ChatGPT-3.5’s role in enhancing various aspects of medical writing. From May 10 to 12, 2023, the authors engaged in a series of interactions with ChatGPT-3.5 to evaluate its effectiveness in various tasks, particularly its application to medical writing, including vocabulary enhancement, text rewriting for plagiarism prevention, hypothesis generation, keyword generation, title generation, article summarization, simplification of medical jargon, transforming text from informal to scientific and data interpretation. The exploration of ChatGPT’s functionalities in medical writing revealed its potential in enhancing various aspects of the writing process, demonstrating its efficiency in improving vocabulary usage, suggesting alternative phrasing, and providing grammar enhancements. While the results indicate the effectiveness of ChatGPT (version 3.5), the presence of certain imperfections highlights the current indispensability of human intervention to refine and validate outputs, ensuring accuracy and relevance in medical settings. The integration of AI into medical writing shows significant potential for improving clarity, efficiency, and reliability. This evaluation highlights both the benefits and limitations of using ChatGPT-3.5, emphasizing its ability to enhance vocabulary, prevent plagiarism, generate hypotheses, suggest keywords, summarize articles, simplify medical jargon, and transform informal text into an academic format. However, AI tools should not replace human expertise. It is crucial for medical professionals to ensure thorough human review and validation to maintain the accuracy and relevance of the content in case they eventually use AI as a supplementary resource in medical writing. Accepting this mutually symbiotic partnership holds the promise of improving medical research and patient outcomes, and it sets the stage for the fusion of AI and human knowledge to produce a novel approach to medical assessment. Thus, while AI can streamline certain tasks, experienced medical writers and researchers must perform final reviews to uphold high standards in medical communications.

1. Introduction

Effective communication is fundamental to medical research, acting as the primary mechanism for disseminating knowledge, facilitating interdisciplinary collaboration, and driving innovation in healthcare.^[1] In this dynamic and rapidly advancing field, the articulation of complex medical concepts and research outcomes with precision and clarity is crucial.^[2] Recent technological advancements have introduced transformative tools, including large language models such as the Chat Generative Pre-trained Transformer (ChatGPT), which hold significant promise for revolutionizing the domain of medical writing.^[2–4] Many artificial intelligence (AI) chatbots have emerged in the past 3 years, and among them, ChatGPT stands out with its exceptional capabilities in natural language processing and generation.^[5] This technology offers immense potential to revolutionize the way medical professionals communicate their work.^[5–7] Using ChatGPT in medical writing introduces a new way to present complex information and raises important questions about the role of AI in a specialized field like medicine.^[8–10] It promotes a reassessment of conventional writing methods and encourages discussion on how AI may enhance, rather than replace, the knowledge and skills of medical practitioners. Investigating the interaction between human expertise and AI capabilities is crucial for assessing the feasibility and efficacy of ChatGPT in improving medical communication.

Despite the increasing recognition of ChatGPT’s benefits in medical writing,^[5–7,11] there are still aspects that remain relatively unexplored, such as the full practical implications^[5] and potential limitations of this large language model.^[5,8,12,13] While its abilities in refining vocabulary usage, aiding in data interpretation, and simplifying complex medical concepts are promising, the question of whether ChatGPT can be considered a trustworthy and effective tool remains unanswered.^[14] Further investigation is needed to assess the reliability and accuracy of ChatGPT’s outputs, as well as its ability to consistently deliver high-quality results.^[5,15–17] Moreover, ChatGPT’s responses can vary significantly depending on the input and interaction, highlighting the need for ongoing research to boost its performance and reliability. The diversity in perspectives and methodologies among researchers further emphasize the necessity for more studies to better understand ChatGPT’s practical applications in medical writing.

Given these challenges, the need for reliable tools to convey complex medical concepts accurately is evident. The limited exploration of AI in medical documentation reveals a substantial gap in the field. To address these unmet needs, our study aimed to evaluate ChatGPT’s effectiveness in medical writing, with a focus on its potential to enhance clarity, efficiency, and reliability in presenting medical information.

2. Materials and methods

The authors adopted a structured methodology that involved a deep examination of the capabilities of ChatGPT-3.5 and its application in medical writing tasks. Our research team members who evaluated ChatGPT’s skills had to meet specific criteria, including extensive experience with ChatGPT,^[5–7,11] a publication record of over 30 articles in high quartile journals (Scopus Quartile), and a great level of English proficiency. This ensured a rigorous assessment and accurate evaluation of ChatGPT’s performance.

To identify relevant articles for our literature review, we used a combination of Boolean operators and conducted searches in PubMed and Scopus. The Boolean operators and keywords used included “ChatGPT” AND “medical writing,” “AI” AND “scientific communication,” and “generative AI” AND “healthcare.” Articles were selected based on their relevance to our study’s focus on AI chatbots in medical writing.

To conduct this study, we engaged in conversations with ChatGPT between May 10 and 12, 2023. We used the publicly free-access GPT-3.5 Language Model (available at https://chat.openai.com/) to assist us in a focused medical research project. We chose one of our published papers, “Assessment tools for pulmonary involvement in patients with ankylosing spondylitis (SpA): is diaphragmatic ultrasonography correlated to spirometry?” published in the Journal of Multidisciplinary Healthcare^[18] as a practical model. This project provided real-world medical writing scenarios and data, which we used to evaluate ChatGPT’s functionalities.

Through a meticulous literature review, we identified the most relevant and impactful themes/tasks that ChatGPT can perform in medical writing. We chose these tasks to help improve various medical writing skills, making it easier for physicians and healthcare staff to be more productive and reduce the time spent on demanding writing tasks. Accordingly, the methodology involved 9 distinct themes/tasks, each exploring a specific aspect of ChatGPT-3.5’s potential in medical writing.

2.1. Vocabulary enhancement

We utilized ChatGPT to refine and enrich medical vocabulary, with the aim of improving the precision and clarity of scientific language.

2.2. Text rewriting for plagiarism prevention

We assessed the usefulness of ChatGPT-3.5 in generating alternative formulations of text to aid in plagiarism prevention while maintaining the integrity of scientific writing.

2.3. Hypothesis generation and research design

We employed ChatGPT-3.5 to propose research hypotheses and assist in the formulation of structured frameworks for study planning, including identifying potential limitations and refining the research questions.

2.4. Keyword generation

We explored ChatGPT-3.5’s functionality in extracting relevant keywords from abstracts to facilitate efficient literature review and information retrieval.

2.5. Article summarization

We evaluated the capability of ChatGPT-3.5 to generate concise summaries of research articles.

2.6. Title generation

The task involves utilizing ChatGPT-3.5 to generate suitable and impactful titles for academic articles, research papers, or other scientific documents.

2.7. Simplification of complex medical jargon

We examined ChatGPT-3.5’s ability to simplify complex medical terms and concepts to improve patient education, interdisciplinary collaboration, and communication with various stakeholders.

2.8. Transforming text from informal to scientific

The task involves using ChatGPT-3.5 to convert informal text into a formal, scientific format suitable for academic publications. This capability is particularly useful for researchers who need to elevate their writing to meet the standards of scholarly communication. The process includes enhancing the precision of language, adopting a more formal tone, and ensuring the text adheres to the conventions of scientific writing.

2.9. Data interpretation

We assessed the efficacy of ChatGPT-3.5 in extracting key findings, elucidating statistical interpretation, and providing concise explanations to enhance the comprehension of research results.

The evaluation process includes 4 criteria for evaluation and 5 evaluation steps. First, the evaluation of ChatGPT-3.5’s outputs was based on the following 4 key criteria: Accuracy: The correctness of the information and the factual soundness of the generated content; Relevance: The pertinence of the generated content to the given medical writing task; Clarity: The clarity of the language and the ease with which the generated content can be understood, and Language fluency: The grammatical correctness and fluency of the generated content. Second, the evaluation steps include the following 5 phases: Initial interaction: Each task began with an initial interaction where specific prompts related to the task were provided to ChatGPT-3.5; Output generation: ChatGPT-3.5 generated responses based on the given prompts, Critical analysis: The generated responses were critically analyzed by the research team for accuracy, relevance, clarity, and language fluency, Secondary interactions: If necessary, secondary interactions were conducted with ChatGPT-3.5 to refine the outputs or address any ambiguities, and Documentation: The entire process, including all interactions and analyses, was documented for each task (see below).

3. Results

We present the results in an original and structured format comprising 9 boxes. Each box represents a specific task related to medical writing. Within each box, we provide the original prompt (the text input given to ChatGPT-3.5), the response generated by ChatGPT-3.5, and any subsequent interactions or secondary responses, if applicable. This unique presentation allows for a clear and organized examination of the capabilities and limitations of the ChatGPT-3.5 in each task.

Throughout the evaluation process, we critically analyzed the responses generated by ChatGPT, considering factors such as accuracy, relevance, and language fluency. This analysis provides valuable insights into the practical implications, strengths, and limitations of utilizing ChatGPT-3.5 for medical writing tasks.

3.1. Vocabulary enhancement

ChatGPT-3.5 provided numerous synonyms and alternative phrasings that were generally accurate and relevant. For example, it suggested “pulmonary function” instead of “lung function,” which is a more precise term in medical contexts. However, some suggestions lacked the necessary specificity, indicating that while ChatGPT can assist in vocabulary enhancement, human oversight is essential to ensure scientific accuracy. Box 1 presents the ChatGPT-3.5’s ability to refine and enrich medical vocabulary, aiming to improve the precision and clarity of the scientific language.

Box 1. ChatGPT’s ability to improve academic writing: Vocabulary enhancer.

Authors ‘Input:

Task: Rephrase the following text to sound as if it was written by a native English speaker.

Provided text: Spondyloarthritis (SpA) is a diverse collection of chronic rheumatic diseases characterized by inflammation in the spine and sacroiliac joints. This group includes a prevalent form known as radiographic SpA, which primarily affects the axial skeleton. It is characterized by inflammatory back pain and can lead to varying levels of structural and functional impairment. Notably, spinal ankylosis, the fusion of the spinal joints, can result in respiratory complications. These complications may include restrictive pulmonary disease, which is caused by limitations in chest wall movement, as well as interstitial lung disease. Both conditions can have significant impacts on an individual’s overall health and well-being.

ChatGPT’s output:

Spondyloarthritis (SpA) refers to a broad spectrum of long-term rheumatic diseases, marked by inflammation primarily occurring in the spine and sacroiliac joints. One prominent type within this group is radiographic SpA, which majorly affects the axial skeleton. This condition is typified by inflammatory back pain and can cause varying degrees of structural and functional damage. Particularly, a condition known as spinal ankylosis, or the fusion of spinal joints, may lead to respiratory issues. These problems might include restrictive pulmonary disease, arising from restricted movement of the chest wall, and interstitial lung disease. Each of these conditions can significantly affect a person’s overall health and quality of life.

Note from the authors: The generated response by ChatGPT effectively paraphrased the original excerpt, adhering to the crucial content and overarching intent. Its lexical selection and sentence structure are congruent with those typically employed by native English speakers. Adherence to grammatical standards is demonstrated throughout the response, and the use of pertinent medical terminology is evident. The model successfully communicates the complexity of the subject matter in an accessible manner, emphasizing potential implications on overall health and quality of life. However, despite its proficiency in rephrasing the text, the response could benefit from more precise terminology and the inclusion of additional contextual details that could further specify the nature of (SpA). This could help in creating a more nuanced understanding of the condition, specifically when addressing the radiographic SpA subtype and its associated complications. Moreover, the model could improve in highlighting the severity of potential respiratory complications, which would provide a more comprehensive view of the disease’s impacts. To summarise, while the ChatGPT model demonstrates aptitude in rephrasing tasks, there is room for improvement in ensuring precise language usage and nuanced elaboration for more complex scientific concepts.

3.2. Text rewriting for plagiarism prevention

ChatGPT-3.5 effectively conveyed the complexity of the subject while preserving its meaning, but better terminology, context, and emphasis on respiratory complications could enhance comprehension of radiographic SpA subtype. ChatGPT-3.5 proficiently paraphrased complex medical content while preserving information integrity, using equivalent terminologies. Overall, the model’s performance was commendable. Box 2 presents ChatGPT-3.5’s text rewriting functionality for plagiarism prevention and rephrasing.

Box 2. ChatGPT’s text rewriting functionality for plagiarism prevention and rephrasing.

Authors ‘Input:

Task: Rephrase the paragraph to prevent any instances of plagiarism.

Provided text:

Spondyloarthritis (SpA) is a diverse collection of chronic rheumatic diseases characterized by inflammation in the spine and sacroiliac joints. This group includes a prevalent form known as radiographic SpA, which primarily affects the axial skeleton. It is characterized by inflammatory back pain and can lead to varying levels of structural and functional impairment. Notably, spinal ankylosis, the fusion of the spinal joints, can result in respiratory complications. These complications may include restrictive pulmonary disease, which is caused by limitations in chest wall movement, as well as interstitial lung disease. Both conditions can have significant impacts on an individual’s overall health and well-being.

ChatGPT’s output:

Spondyloarthritis (SpA) represents a varied group of persistent rheumatic conditions that primarily cause inflammation in the spine and sacroiliac joints. A common variant of this group is radiographic SpA, which chiefly targets the axial skeleton. This variant is marked by inflammatory back pain and could lead to a spectrum of structural and functional damages. A prominent issue, spinal ankylosis, which involves the fusion of spinal joints, could give rise to respiratory problems. These issues could manifest as restrictive pulmonary disease, a condition hindered by limited chest wall mobility, and interstitial lung disease. Each of these conditions can profoundly affect a person’s overall health status and quality of life.

Note from the authors: In the assessment of the revised paragraph, we acknowledge the adept proficiency exhibited by the ChatGPT model. Its capacity to paraphrase complex medical content while maintaining the original information’s integrity was demonstrated effectively in the task. The revised text successfully encapsulated the essence of the original paragraph, delineating the core concepts and attributes related to (SpA). The usage of alternative but equivalent terminologies, such as ‘persistent’ for ‘chronic’ and ‘variant’ for ‘form’, effectively achieved the objective of text rephrasing without sacrificing content fidelity. Moreover, the artificial intelligence (AI)’s ability to provide appropriate synonyms while preserving the scientific nuances of the original text is noteworthy. However, some critical considerations exist. The phrase ‘profoundly affect a person’s overall health status and quality of life’ seems to generalize the effects of the described medical conditions. While it is true that these conditions can have a severe impact, the degree of impact can be variable and individual-specific. A more cautious phrasing might have more accurately conveyed this aspect. Despite these observations, the model’s overall performance in this task was commendable.

3.3. Hypothesis generation

ChatGPT-3.5’s response on the correlation between diaphragmatic ultrasonography and spirometry in evaluating pulmonary disorders in radiographic SpA patients has been well-informed and coherent but lacked specific objectives and failed to acknowledge limitations for a more comprehensive research approach. Box 3 presents the ChatGPT-3.5’s text rewriting functionality to provide a hypothesis and research design based on the introduction of a research paper.

Box 3. ChatGPT’s text writing functionality to provide hypothesis and research design.

Authors ‘Input:

Task: provide hypothesis and research design based on the introduction below.

Provided text: Introduction of the article “Dhahri, R., et al (2023). Assessment tools for pulmonary involvement in patients with ankylosing spondylitis: Is diaphragmatic ultrasonography correlated to spirometry? Journal of Multidisciplinary Healthcare, 51-61.”

ChatGPT’s output:

Hypothesis: We hypothesize that ultrasonography (US) can serve as an accurate screening tool for assessing diaphragmatic involvement and pulmonary disorders in patients with radiographic SpA. We further hypothesize that US findings will show a correlation with the disease’s activity, as well as spirometric measurements.

Research Design: To test our hypotheses, we propose a cross-sectional observational study involving patients diagnosed with radiographic SpA. The study design will include the following key elements:

1. Participant Recruitment: Patients with radiographic SpA will be recruited from rheumatology clinics or specialized SpA centers. Informed consent will be obtained from all participants.

2. Sample Size Calculation: A sample size calculation will be performed to ensure adequate statistical power. This calculation will be based on the desired level of significance, effect size, and expected dropout rate.

3. Data Collection: Relevant demographic and clinical data will be collected, including disease activity scores, duration of SpA, and medication use. Spirometry will be performed to measure lung volumes and identify restrictive patterns. Ultrasonography will be used to assess diaphragmatic motion and morphology.

4. Statistical Analysis: Statistical analysis will be conducted to compare the accuracy of diaphragm US and spirometry in detecting pulmonary disorders in radiographic SpA patients. Sensitivity, specificity, positive predictive value, and negative predictive value will be calculated. Correlation analyses will be performed to determine the relationship between US findings, disease activity, and spirometric measurements.

5. Ethical Considerations: The study will adhere to ethical guidelines, ensuring patient confidentiality, privacy, and informed consent. Approval from the appropriate ethics committee will be obtained before the commencement of data collection.

6. Limitations: Potential limitations, such as the cross-sectional nature of the study and the selection bias associated with recruiting from specialized clinics, will be acknowledged. The proposed research design aims to investigate the accuracy of diaphragm US as a screening tool for pulmonary disorders in radiographic SpA patients and explore its correlation with disease activity. The results of this study may provide valuable insights into the utility of US in assessing respiratory involvement in radiographic SpA and contribute to the development of improved diagnostic and monitoring strategies for these patients.

Note from the authors: The response generated by ChatGPT in formulating the hypothesis and research design based on the provided introduction exhibits a commendable level of understanding and coherence with the subject matter. The hypothesis successfully posits the potential relationship between diaphragmatic ultrasonography (US) and spirometry in the assessment of pulmonary disorders in radiographic SpA patients, while also proposing a correlation with disease activity. The proposed research design, structured as a cross-sectional observational study, appears suitable for investigating the accuracy of diaphragm US and its correlation with spirometry in this context. However, one critical aspect to consider is the absence of specific objectives that align with the stated hypothesis. A more precise formulation of research objectives, such as assessing the agreement between diaphragmatic US and spirometry findings, evaluating the diagnostic accuracy of diaphragm US in identifying pulmonary disorders, and exploring the association between disease activity and diaphragm involvement, would enhance the clarity of the research design. Furthermore, the limitations of the proposed study, such as potential selection bias and the cross-sectional nature of the investigation, could have been acknowledged to ensure a comprehensive evaluation of the research approach. Overall, while the response demonstrates a sound grasp of the research elements involved in hypothesis formulation and research design.

3.4. Keyword generation

ChatGPT’s generated keywords for the research paper on diaphragm ultrasound and spirometry in radiographic SpA patients have been precise, but some repetition with terms in the abstract or title could have been avoided. Including keywords related to “Inflammatory back pain” and “Structural and functional impairment” would have enhanced the representation of the research. Nonetheless, ChatGPT-3.5 accomplished its task of providing the requested input and follow-up interaction effectively. Box 4 presents ChatGPT-3.5’s functionality in extracting relevant keywords from abstracts.

Box 4. ChatGPT’s functionality in extracting relevant keywords from abstracts.

Authors ‘Input:

Provided text: Abstract of the article “Dhahri, R., et al (2023). Assessment tools for pulmonary involvement in patients with ankylosing spondylitis: Is diaphragmatic ultrasonography correlated to spirometry? Journal of Multidisciplinary Healthcare, 51-61.”

Task 1. Based on the provided abstract provide five keywords for this research article:

ChatGPT’s output 1:

Spondyloarthritis (SpA); Radiographic SpA; Pulmonary complications; Diaphragmatic involvement; Ultrasonography (US)

Interaction 1: Provide keywords that are not duplicated with the title.

ChatGPT’s output 2:

Inflammatory back pain; Structural and functional impairment; Restrictive pulmonary disease; Chest wall limitations; Interstitial lung disease

Note from the authors: The keywords generated by ChatGPT in response to the provided abstract accurately capture the main themes of the research paper on diaphragm ultrasound and spirometry in radiographic SpA patients. The keywords effectively represent the disease context, pulmonary complications, diaphragmatic involvement, and the imaging modality used. However, the response could have been improved by avoiding duplication with terms already present in the abstract or title. The addition of keywords like “Inflammatory back pain” and “Structural and functional impairment” would have further enriched the representation of the research. Overall, ChatGPT achieved its goal with the input it was asked to deliver on both the initial inout and follow-up interaction.

3.5. Title generation

ChatGPT-3.5’s generated title for the study on diaphragmatic involvement and pulmonary complications in radiographic SpA, comparing ultrasonography and spirometry, has been effective. However, including specific terms from the abstract would have improved relevance and specificity. Alternative titles lacked variation in structure and phrasing. Box 5 presents the functionality of ChatGPT-3.5 in proposing a potential title for a research paper based on abstract provision.

Box 5. ChatGPT’s functionality in proposing potential title for a research paper.

Authors ‘Input:

Task: provide a potential title for this research paper based on the provided abstract.

Provided text: Abstract of the article (please refer to “Dhahri, R., et al (2023). Assessment tools for pulmonary involvement in patients with ankylosing spondylitis: Is diaphragmatic ultrasonography correlated to spirometry? Journal of Multidisciplinary Healthcare, 51-61.”)

ChatGPT’s output:

“Assessment of diaphragmatic involvement and pulmonary complications in radiographic spondyloarthritis: A comparative study between ultrasonography and spirometry”

Interaction 1: Give five other alternatives

• “Exploring diaphragmatic involvement and pulmonary complications in radiographic spondyloarthritis: A comparative analysis of ultrasonography and spirometry”

• “Diaphragm evaluation and pulmonary impairment in radiographic spondyloarthritis: A comparative study of ultrasonography versus spirometry”

• “Comparing ultrasonography and spirometry in assessing diaphragmatic involvement and pulmonary dysfunction in radiographic spondyloarthritis”

• “Role of ultrasonography and spirometry in evaluating pulmonary complications and diaphragmatic involvement in radiographic spondyloarthritis”

• “Diagnostic utility of ultrasonography and spirometry for assessing pulmonary disorders and diaphragmatic involvement in radiographic spondyloarthritis”

Note from the authors: The title generated by ChatGPT for the research paper effectively captures the focus of the study on diaphragmatic involvement and pulmonary complications in radiographic SpA, with a comparative analysis between ultrasonography and spirometry. While the alternative titles provided lacked variation in structure and phrasing, the overall titles convey the key elements of the research. However, it would have been beneficial if the titles incorporated specific terms from the abstract to enhance their relevance and specificity. Overall, the titles adequately represent the study’s focus.

3.6. Article summarization

ChatGPT’s summary of Berdal et al’s^[19] study has been concise and accurate, emphasizing the higher prevalence of restrictive pulmonary function in ankylosing spondylitis patients linked to impaired spinal mobility. While suitable for scientific papers, additional details and limitations could have been added. ChatGPT-3.5 can save time in the early stages of literature review, but researchers should independently verify the information due to potential limitations and biases of the AI model. Box 6 presents the capability of ChatGPT-3.5 to generate concise summaries of research articles from their references.

Box 6. ChatGPT’s capability to generate concise summaries of research articles from their references.

Authors ‘Input:

Task number 1: No task, we just pasted the five first references of the article.

Provided text:

1. Berdal G, Halvorsen S, van der Heijde D, Mowe M, Dagfinrud H. Restrictive pulmonary function is more prevalent in patients with ankylosingspondylitis than in matched population controls and is associated with impaired spinal mobility: a comparative study. Arthritis Res Ther. 2012;14(1):R19. doi:10.1186/ar36992.

2. Alaee S, Jones Q. Case of drug-induced interstitial lung disease secondary to Adalimumab. BMJ Case Rep. 2018;2018. doi:10.1136/bcr-2018-2243753.

3. Dincer U, Cakar E, Kiralp MZ, Bozkanat E, Kilac H, Dursun H. The pulmonary involvement in rheumatic diseases: pulmonary effects ofankylosing spondylitis and its impact on functionality and quality of life. Tohoku J Exp Med. 2007;212(4):423–430. doi:10.1620/tjem.212.4234.

4. Fisher LR, Cawley MI, Holgate ST. Relation between chest expansion, pulmonary function, and exercise tolerance in patients with ankylosingspondylitis. Ann Rheum Dis. 1990;49(11):921–925. doi:10.1136/ard.49.11.9215.

5. Romagnoli I, Gigliotti F, Galarducci A, et al. Chest wall kinematics and respiratory muscle action in ankylosing spondylitis patients. Eur Respir J.2004;24(3):453–460. doi:10.1183/09031936.04.00123903

ChatGPT’s output:

The references you provided are a list of scientific studies related to ankylosing spondylitis and its impact on pulmonary function. Here is a brief summary of each study:

1. Berdal G, Halvorsen S, van der Heijde D, Mowe M, Dagfinrud H. “Restrictive pulmonary function is more prevalent in patients with ankylosing spondylitis than in matched population controls and is associated with impaired spinal mobility: a comparative study.” Arthritis Res Ther. 2012;14(1):R19.

2. This study compared the prevalence of restrictive pulmonary function in patients with ankylosing spondylitis to that of matched population controls. It found that restrictive pulmonary function was more common in ankylosing spondylitis patients and was associated with impaired spinal mobility.

3. Alaee S, Jones Q. “Case of drug-induced interstitial lung disease secondary to Adalimumab.” BMJ Case Rep. 2018;2018.

4. This is a case report that describes a specific case of interstitial lung disease (ILD) that occurred as a side effect of Adalimumab, a medication commonly used to treat ankylosing spondylitis. The report highlights the importance of monitoring patients for potential pulmonary complications associated with this medication.

5. Dincer U, Cakar E, Kiralp MZ, Bozkanat E, Kilac H, Dursun H. “The pulmonary involvement in rheumatic diseases: pulmonary effects of ankylosing spondylitis and its impact on functionality and quality of life.” Tohoku J Exp Med. 2007;212(4):423–430.

6. This study investigated the pulmonary effects of ankylosing spondylitis and their impact on functionality and quality of life. It provides insights into how ankylosing spondylitis can affect the lungs and respiratory function, which can lead to functional limitations and decreased quality of life.

7. Fisher LR, Cawley MI, Holgate ST. “Relation between chest expansion, pulmonary function, and exercise tolerance in patients with ankylosing spondylitis.” Ann Rheum Dis. 1990;49(11):921–925.

8. This study examined the relationship between chest expansion, pulmonary function, and exercise tolerance in patients with ankylosing spondylitis. It explored how limitations in chest expansion, which are common in ankylosing spondylitis, can affect pulmonary function and exercise capacity.

9. Romagnoli I, Gigliotti F, Galarducci A, et al “Chest wall kinematics and respiratory muscle action in ankylosing spondylitis patients.” Eur Respir J. 2004;24(3):453–460.

10. This study investigated the chest wall kinematics (movement) and respiratory muscle action in patients with ankylosing spondylitis. It aimed to understand the impact of the disease on the mechanics of breathing and how respiratory muscles are affected in individuals with ankylosing spondylitis.

11. These studies collectively contribute to our understanding of the pulmonary involvement in ankylosing spondylitis and shed light on the respiratory impairments and challenges faced by individuals with this condition.

Task number 2: Summarise the findings of the study 1.

ChatGPT Response for Task number 2:

The findings of study 1, conducted by Berdal et al (2012), indicate that restrictive pulmonary function is more prevalent in patients with ankylosing spondylitis compared to matched population controls. The study also found that this restrictive pulmonary function is associated with impaired spinal mobility in ankylosing spondylitis patients. In other words, individuals with ankylosing spondylitis are more likely to have reduced lung function and difficulty expanding their chest due to the limitations in their spinal mobility. This study highlights the importance of assessing and managing pulmonary function in patients with ankylosing spondylitis to optimize their respiratory health.

Task number 3: expand it.

ChatGPT Response for Task number 3:

The study conducted by Berdal et al (2012) aimed to investigate the prevalence of restrictive pulmonary function in patients with ankylosing spondylitis and its association with impaired spinal mobility. Restrictive pulmonary function refers to a decrease in lung volume or reduced ability to expand the chest during breathing. The researchers compared a group of patients with ankylosing spondylitis to a matched control group from the general population. They assessed pulmonary function using spirometry, which measures lung volume and airflow. Spinal mobility was evaluated using various clinical measures. The results of the study revealed that patients with ankylosing spondylitis had a higher prevalence of restrictive pulmonary function compared to the control group. This suggests that ankylosing spondylitis may have a direct impact on lung function, leading to reduced lung volume or impaired chest expansion. Furthermore, the study found a significant association between restrictive pulmonary function and impaired spinal mobility in ankylosing spondylitis patients. The limitations in spinal mobility, which are characteristic of the disease, can restrict the movement of the rib cage and chest, making it difficult for individuals to fully expand their lungs during breathing. These findings underscore the importance of considering pulmonary function in the management of ankylosing spondylitis. Assessing lung function and monitoring for restrictive pulmonary patterns can help identify individuals who may require additional respiratory support or interventions. Additionally, optimizing spinal mobility through appropriate exercises and therapies may help improve lung function and overall respiratory health in patients with ankylosing spondylitis. Overall, the study provides valuable insights into the relationship between ankylosing spondylitis, restrictive pulmonary function, and impaired spinal mobility. It emphasizes the need for a comprehensive approach to the management of ankylosing spondylitis that considers both musculoskeletal and respiratory aspects of the disease.

Note from the authors: ChatGPT has provided a concise and accurate summary of the findings of study 1 conducted by Berdal et al (2012). The summary effectively highlights that patients with ankylosing spondylitis have a higher prevalence of restrictive pulmonary function compared to the control group, and this impairment is associated with impaired spinal mobility. The response demonstrates a solid understanding of the study’s methodology and clinical implications. However, it could have included additional details such as sample size and demographic characteristics of the participants, as well as any limitations of the study. Overall, the response effectively captures the key findings and would be suitable for inclusion in a scientific paper. In addition to the summary of the study findings, it is important to acknowledge that ChatGPT can serve as a valuable tool in facilitating the process of exploring research papers. By generating concise summaries, ChatGPT can assist researchers in quickly obtaining an overview of a paper’s key points, thereby saving time during the literature review process. With the ever-increasing volume of scientific literature, researchers often face the challenge of efficiently sifting through numerous papers. ChatGPT’s ability to provide summarized insights can aid researchers in rapidly identifying relevant articles and gaining a preliminary understanding of their content. By using ChatGPT as a tool for initial exploration, researchers can streamline their literature review and allocate more time for in-depth analysis and critical evaluation of the selected papers. It is important to note, however, that while ChatGPT can be a helpful resource, it should not replace the thorough reading and critical assessment of scientific papers. Researchers should independently verify the information provided by ChatGPT, as it is an artificial intelligence model and may not capture all the nuances or potential biases present in the original texts. Nonetheless, as a time-saving tool in the early stages of literature review, ChatGPT can be beneficial in quickly gaining an understanding of research papers.

3.7. Simplification of medical jargon

ChatGPT simplifies complex medical text, enabling easier comprehension of complex concepts and providing a clear and concise translation; however, a balance between simplification and retention of key terms is needed, along with stronger emphasis on the severity of potential complications. Despite this, ChatGPT-3.5 has been a valuable tool for time-constrained situations and diverse audiences. Box 7 presents ChatGPT’s ability to simplify medical jargon.

Box 7. ChatGPT’s ability to simplify medical jargon.

Authors ‘Input:

Task: Simplify this text so that a 5-year-old can understand it easily.

Provided text:

Spondyloarthritis (SpA) is a heterogeneous group of chronic rheumatic diseases characterized by inflammation of the spine and sacroiliac joints. Radiographic SpA is one common form of SpA, affecting mainly the axial skeleton, causing inflammatory back pain and resulting in varying degrees of structural and functional impairment. Spinal ankylosis can lead to respiratory complications including restrictive pulmonary disease induced by chest wall excursion limitations and/or interstitial lung disease, both of which can cause serious impairment.

ChatGPT’s output 1:

Spondyloarthritis (SpA) is a group of diseases that cause inflammation in the spine and joints. One type of SpA affects the backbone and can cause pain and problems with how the body moves. Sometimes, this can make it hard to breathe and can hurt the lungs. This can happen because the chest doesn’t move well or because the lungs get damaged. These problems can make it difficult to do things and can be very serious.

Interaction 1: make it simpler.

ChatGPT’s output 2:

Spondyloarthritis, or SpA, is a sickness that makes your back and some parts of your body hurt a lot. It can also make it hard for you to move and breathe. It’s very important to take care of this because it can be really serious.

Note from the authors: ChatGPT has demonstrated its remarkable capability in simplifying complex medical text, allowing for easier comprehension of intricate concepts. By providing a clear and concise translation of the original excerpt on spondyloarthritis, it successfully maintains the overall meaning and essence of the information. This practical application becomes evident in situations where time is limited, such as when preparing for meetings or needing to explain medical conditions to geriatric or paediatric populations in a simplified manner. The model effectively distils medical terminology into language that is easily understandable, even to a 5-year-old audience. While the simplified version effectively conveys the basics of spondyloarthritis, there is room for improvement in striking a balance between simplification and the retention of key disease-specific terms. Preserving these terms is crucial for fostering familiarity with the concepts and facilitating further learning. Furthermore, the simplified explanation may benefit from a stronger emphasis on the severity of potential respiratory complications, ensuring that their impact is adequately conveyed. By highlighting the challenges and implications of such complications, a more accurate understanding of the condition can be achieved. Indeed, ChatGPT’s ability to simplify complex medical text proves to be highly beneficial in time-constrained situations. While it effectively conveys the core aspects of the subject matter, refining the balance between comprehension and information retention would enhance its overall usefulness. Leveraging the model as a tool for efficient comprehension of medical literature and simplification of explanations for diverse audiences holds significant potential for healthcare professionals in various contexts.

3.8. Transforming text from informal to scientific

ChatGPT-3.5 demonstrated linguistic acuity and adeptness in transforming informal exposition into academic vernacular, maintaining the essence of the original text; however, incorporating pertinent medical terminology, optimizing sentence structure, and adhering to academic standards would have enhance its precision and professionalism as a tool in medical writing. Box 8 presents the ChatGPT-3.5’s ability to transform text from an informal to an academic writing style.

Box 8. ChatGPT’s ability to transform a text from informal to academic.

Authors ‘Input:

Task: transform this text below from informal to academic.

Provided text: Asthma is a common lung condition that causes occasional breathing difficulties. It affects people of all ages and often starts in childhood. Symptoms include wheezing, breathlessness, a tight chest, and coughing.

ChatGPT’s output:

Asthma is a prevalent pulmonary disorder, leading to sporadic episodes of impaired respiration. It transcends age boundaries, frequently initiating during the early stages of life. Clinically, it is characterized by a constellation of symptoms such as wheezing, dyspnoea, a sensation of chest constriction, and persistent coughing.

Note from the authors: In our critical assessment of the provided response by ChatGPT, we observe a satisfactory demonstration of linguistic acuity and adeptness at transforming informal exposition into an academic vernacular. The artificial intelligence model exhibited capability in accurately paraphrasing the original text, maintaining the essence while refining the language to align with scholarly discourse, thus underscoring its potential as an ancillary tool in academic writing. However, a few constructive critiques merit attention. While ChatGPT adeptly translated the text into an academic format, the resulting content could be further enhanced with the incorporation of pertinent medical terminology. For instance, using the term ‘bronchial hyperreactivity’ could offer a more precise description of asthma’s pathophysiological characteristic in place of ‘impaired respiration.’ Additionally, more specific terms such as ‘paediatric onset’ could provide a more accurate depiction than ‘initiating during the early stages of life.’ Furthermore, the structure of the sentence could be optimized to streamline the flow of information. Performing such tasks aids in enhancing the precision and professionalism of medical writing, tailoring the language to align with scholarly discourse. It can assist in effectively conveying complex medical concepts while adhering to academic standards. Additionally, the ability to transform informal text into an academic format can significantly streamline the writing process, saving valuable time and effort for medical professionals.

3.9. Data interpretation

ChatGPT-3.5 translated complex research results into a clear, accessible format, aiding those struggling with scientific writing and improving understanding of correlations and findings. However, it could have provided potential clinical implications for a more holistic perspective. Overall, ChatGPT’s utility in decoding complex scientific data and increasing accessibility has undoubtedly been beneficial. Box 9 presents ChatGPT-3.5’s ability to interpret data.

Box 9. ChatGPT’s ability to interpret data.

Authors ‘Input:

Task: Interpret this result section:

Provided text: Correlation between US and PFT findings (Table 2). A significant positive moderate correlation was found between the right tdi-insp and the FVC (r = 0.36; p = 0.02) and the FVC in the supine position (r = 0.29; p = 0.04). A significant positive moderate correlation was found between the left tdi-insp and the percentage of decrease in FVC in supine (Δ FVC) (r = 0.35; p = 0.01). A significant positive moderate correlation was found between the right tdi-exp and the FVC (r = 0.32; p = 0.02). There was no correlation between the DE nor the TF measures and the spirometry findings.

ChatGPT’s output:

This result section is presenting the correlation between some measurements from an ultrasound (US) and a Pulmonary Function Test (PFT). Correlation coefficients (denoted by ‘r’) range from -1 to 1, where 1 is a perfect positive correlation, -1 is a perfect negative correlation, and 0 is no correlation. The p-value (denoted by ‘p’) is used to determine the significance of results; a p-value of less than 0.05 is generally considered statistically significant. The right transdiaphragmatic inspiration (tdi-insp) shows a moderate, statistically significant positive correlation with Forced Vital Capacity (FVC). The correlation coefficient is 0.36 (p = 0.02), indicating that as the right tdi-insp increases; the FVC also tends to increase. The right tdi-insp also shows a moderate, statistically significant positive correlation with the FVC in the supine position. The correlation coefficient is 0.29 (p = 0.04), suggesting that as the right tdi-insp increases, the FVC in the supine position also tends to increase. There’s a moderate, statistically significant positive correlation between the left tdi-insp and the percentage decrease in FVC in the supine position (Δ FVC). The correlation coefficient is 0.35 (p = 0.01). This suggests that as the left tdi-insp increases, the percentage decrease in FVC in the supine position also tends to increase. The right transdiaphragmatic expiration (tdi-exp) shows a moderate, statistically significant positive correlation with FVC. The correlation coefficient is 0.32 (p = 0.02), suggesting that as the right tdi-exp increases, the FVC also tends to increase. There was no correlation between the diaphragmatic excursion (DE) nor the thickening fraction (TF) measures and the spirometry findings, which means that variations in DE and TF do not seem to have a significant impact on the spirometry results.

Note from the authors: ChatGPT interpretation excelled in translating the complex research results into a clear, accessible format, making it a valuable tool for those who might struggle with the nuances of scientific writing, such as physicians with limited experience in this field. By breaking down intricate correlations and findings from the ultrasound and pulmonary function tests, it enhanced understanding of the study’s results. Although the artificial intelligence performed well in this area, it could have provided a more holistic perspective by discussing the potential clinical implications of the results. Despite these shortcomings, the utility of ChatGPT in decoding complex scientific data and increasing its accessibility remains unquestionably beneficial.

4. Discussion

This study looked at how ChatGPT-3.5 can help improve medical writing and communication of medical information. We tested its ability to enhance vocabulary, interpret data, summarize articles, and simplify complex medical jargons.

It is important to note that ChatGPT provides a variety of functionalities beyond medical writing. It can be utilized for tasks such as composing emails, generating inquiries,^[5,9] and even performing sample size calculations.^[11] However, in our study, we specifically concentrated on the most practical and widely employed features of ChatGPT-3.5 in medical writing. We identified certain limitations and areas requiring improvement to ensure a more balanced assessment of the competencies of ChatGPT-3.5 in medical writing. While we found that ChatGPT-3.5 demonstrated remarkable capabilities in improving medical writing productivity and enhancing the clarity and precision of scientific language, it is important to note that the model is not without its shortcomings. In some cases, it provided suggestions for alternative phrasing, improved grammar, and relevant synonyms. However, there were instances where the provided suggestions were inaccurate or contextually inappropriate. This highlights the need for careful review and critical evaluation of the model’s suggestions before incorporating them into any manuscript. Furthermore, while ChatGPT-3.5 can be a valuable tool for researchers to present their work with greater impact and engage readers more effectively, it should not replace the expertise and judgment of human medical experts.

In addition, ChatGPT-3.5’s writing style can be repetitive, with similar syntax and sentence structures. This uniformity can be noticed by reviewers and editors who are familiar with AI-generated text, which poses ethical risks if the text is used without changes. In this context, universities and editorial boards now use AI detection tools to screen documents. Therefore, students and researchers must revise AI-generated content carefully to avoid ethical issues.^[5]

It is important to view AI chatbots like ChatGPT-3.5 as tools to enhance productivity rather than as replacements for the writing styles of students, researchers, and medical experts. These tools should support the writing process, helping to improve clarity and efficiency, but the final content should reflect the unique voice and expertise of the human author(s).

Our investigation of ChatGPT-3.5’s data interpretation features showed its ability to highlight key findings, explain statistical correlations, and provide clear summaries. This could help researchers and clinicians understand their results more easily and effectively apply their findings. ChatGPT-3.5 could also help make complex data more understandable, improving communication among medical professionals and potentially support evidence-based decisions. This would be particularly useful for medical experts who may not have advanced statistical skills. Physicians often focus on patient care and may not be well-versed in research or statistics. ChatGPT could help them understand the latest scientific findings without needing deep statistical knowledge. Additionally, medical students and researchers who are less experienced with statistics can use AI to check if they are interpreting their results correctly. This support can be invaluable in ensuring accurate understanding and application of research data.

However, ChatGPT-3.5’s reliance on continuous feedback and interactions with domain specialists to ensure accuracy, relevance, and ethical considerations highlights its limitations. Human expertise is crucial for validating and interpreting AI outputs. ChatGPT-3.5 should be viewed as a support tool rather than a replacement for medical professionals, emphasizing the importance of collaboration between AI systems and domain experts.^[20]

One notable limitation is ChatGPT-3.5’s inconsistency, as it rarely provides the same response to identical inputs. This variability can challenge maintaining consistency and reliability. Additionally, the contextual limitations of ChatGPT-3.5 should be considered. While proficient in general medical writing, it may struggle with more complex or specialized topics requiring specific domain knowledge. Users must exercise caution and critical thinking, ensuring that AI-generated content aligns with established medical guidelines and standards. Another issue is related to the time period over which it bases its response. Free access versions of chatbots (as ChatGPT-3.5) are using data which is relatively outdated (usually missing the last year(s) of information). This has to be taken into account if ever the investigated topic is time sensitive. The practical implications of integrating ChatGPT-3.5 into medical writing are significant. By addressing its limitations, researchers can enhance productivity, streamline the writing process, and devote more time to critical thinking and analysis. ChatGPT-3.5’s ability to simplify complex medical jargon can improve patient education, facilitate interdisciplinary collaboration, and enhance communication among healthcare professionals.

Concerns about AI replacing medical doctors are prevalent, but our findings suggest otherwise. ChatGPT-3.5 and similar models offer opportunities for transformation and growth, not obsolescence. Medical professionals who embrace AI integration are better positioned to revolutionize the field. The synergy between human doctors and AI can surpass traditional methods, improving efficiency and patient care. True pioneers in medicine will be those who effectively integrate AI tools into their practice, navigating the evolving healthcare landscape to provide accurate diagnoses and personalized treatments.

The rising influence of AI chatbots like ChatGPT-3.5 in medicine necessitates careful implementation. Their use may become indispensable, but healthcare practitioners and academics must understand their abilities and limitations. Future research should prioritize developing detailed ethical guidelines for AI use; ensuring integration into healthcare upholds high standards of patient care and data integrity. The uncritical use of AI-generated text, especially verbatim copying, can raise ethical concerns of plagiarism or misconduct. AI tools should be used judiciously, with thorough review and fine-tuning of generated content, as the responsibility for published material lies with the researchers and medical experts.^[8,20–22]

In conclusion, integrating AI, including models like ChatGPT-3.5, into medical practice marks a significant shift. Medical professionals who use these technologies could influence the future of healthcare. The symbiosis of human expertise and AI has the potential to transform medical writing and the broader field of medicine. This relationship could improve patient care and advance medical research.^[5,23,24] However, it is crucial to maintain high ethical standards and transparency when using ChatGPT and similar technologies. Clear ethical guidelines and transparency in AI model development and use are needed to help build trust, protect patient privacy, and ensure responsible medical writing practices.^[25,26] Moreover, recommendations for researchers and writers to ensure responsible and effective use of ChatGPT-like Chatbots in academic writing are urgently needed.^[27]

Author contributions

Conceptualization: Ismail Dergaa, Helmi Ben Saad.

Formal analysis: Ismail Dergaa, Helmi Ben Saad, Jordan M. Glenn, Karim Chamari.

Visualization: Ismail Dergaa, Helmi Ben Saad.

Writing—original draft: Ismail Dergaa, Helmi Ben Saad, Jordan M. Glenn, Mohamed Ben Aissa, Morteza Taheri, Sarya Swed, Noomen Guelmami, Karim Chamari.

Writing—review & editing: Ismail Dergaa, Helmi Ben Saad, Jordan M. Glenn, Karim Chamari.