A comparison of the responses between ChatGPT and doctors in the field of cholelithiasis based on clinical practice guidelines: a cross-sectional study

Abstract

Background

With the development of the information age, an increasing number of patients are seeking information about related diseases on the Internet. In the medical field, several studies have confirmed that ChatGPT has great potential for use in medical education, generating imaging reports, and even providing clinical diagnosis and treatment decisions, but its ability to answer questions related to gallstones has not yet been reported in the literature.

Objective

The aim of this study was to evaluate the consistency and accuracy of ChatGPT-generated answers to clinical questions in cholelithiasis, compared to answers provided by clinical expert.

Methods

This study designs an answer task based on the clinical practice guidelines for cholelithiasis. The answers are presented in the form of keywords. The questions are categorized into general questions and professional questions. To evaluate the performance of ChatGPT and clinical expert answers, the study employs a modified matching scoring system, a keyword proportion evaluation system, and the DISCERN tool.

Results

ChatGPT often provides more keywords in its responses, but its accuracy is significantly lower than that of doctors (P < .001). In the evaluation of 33 general questions, ChatGPT and doctors demonstrated similar performance in both the modified matching score system and keyword proportion evaluation (P = .856 and P = .829, respectively). However, in the evaluation of 32 professional questions, doctors consistently outperformed ChatGPT (P = .004 and P = .016). Additionally, while the DISCERN tool showed differences between general and professional questions (P = .001), both types of questions were evaluated at a high level overall.

Conclusions

Currently, ChatGPT performs similarly to clinical experts in answering general questions related to cholelithiasis. However, it cannot replace clinical experts in professional clinical decision-making. As ChatGPT's performance improves through deep learning, it is expected to become more useful and effective in the field of cholelithiasis. Nevertheless, in more specialized areas, careful attention and continuous evaluation will be necessary to ensure its accuracy, reliability, and safety in the medical field.

Keywords

Cholelithiasis ChatGPT guideline artificial intelligence

Introduction

Background

Cholelithiasis (Gallstone disease) is one of the most common digestive diseases worldwide, with an incidence of 6%–25% and an increasing trend.^1,2 The main types of cholelithiasis are gallstones, common bile duct stones, and intrahepatic bile duct stones. The principal clinical manifestations include abdominal pain, nausea, vomiting, and jaundice, among which laparoscopic cholecystectomy is considered the gold standard for the treatment of gallstones.³ For common bile duct stones and intrahepatic bile duct stones, individualized treatment is often tailored to the specific conditions of the patient.^4–8

With the development of the information age, an increasing number of patients are seeking information about related diseases on the Internet. According to surveys, there are at least 6.75 million searches for health-related information on the web every day.⁹ Recently, artificial intelligence (AI) has gained popularity in the medical field, and ChatGPT (Generative Pre-trained Transformer) is one of the most representative tools among them.¹⁰ ChatGPT is a large language model (LLM) with 175 billion parameters that can answer questions and provide results and is widely used in several fields,¹¹ It may even replace internet searches. In the medical field, several studies have confirmed that ChatGPT has great potential for use in medical education, generating imaging reports, and even providing clinical diagnosis and treatment decisions.^12,13T The results of Deng et al.'s study show that ChatGPT has excellent applicability in clinical applications of breast cancer.¹⁴ Yeo et al.¹⁵ have also demonstrated that ChatGPT can be an effective auxiliary tool for patients and doctors in cirrhosis and hepatocellular carcinoma. A growing body of research has proven that ChatGPT is capable of taking on medical tasks, but its ability to answer questions related to gallstones has not yet been reported in the literature. Moreover, we still question whether AI can replace clinical experts when it comes to answering more professional questions. This is essential to ensure patient safety.

The main objective of this study is to evaluate the potential of ChatGPT and clinical experts to differentiate between general and professional questions, using the recommendations from the clinical guidelines for cholelithiasis as standard answers. This evaluation aims to provide clear insights for patients with cholelithiasis and clinicians when utilizing ChatGPT, to understand the value and limitations of ChatGPT in clinical decision support. The secondary objective involves using the DISCERN tool to assess the quality of ChatGPT's responses to questions related to gallstones.

Methods

Ethics

This cross-sectional study was conducted at the Department of Hepato-Pancreato-biliary Surgery, People's Hospital of Leshan from April 27, 2024 to May 21, 2024. As no patients are involved in this study, approval from the ethics committee is not required.

Question generation and standard answer development

This study referenced the clinical practice guidelines for cholelithiasis issued by the European Association for the Study of the Liver (EASL) and the Japanese Society of Gastroenterology (JSGE). A series of questions mainly covering the etiology, diagnosis, treatment decision, and prognosis of cholelithiasis were formulated. Standard answers, developed according to these guidelines, are used to evaluate the correct response rate of ChatGPT and clinical experts to questions related to cholelithiasis. To ensure the feasibility and accuracy of the assessment, and to avoid subjective judgment by the evaluation researcher in identifying ChatGPT's answers, we only compare the keywords in the answers, not the whole sentence. In addition to words such as Yes, No, Unknown, and Similar, we define keywords as: (1) only involving core nouns related to the problem, retaining only one keyword with the same meaning, and if it is not related to the problem, it is not recorded, such as asking whether gallstones can be prevented but involving specific preventive measures; when asked about clinical manifestations and the answer is related to treatment, such keywords are not recorded; (2) medical professional terms, involving examinations, treatments, drugs, etc., such as cholesterol, Ultrasound, Laparoscopy, Antibiotics, Nonsteroidal Anti-Inflammatory Drug (NSAID), etc. Keywords congruence is defined as: (1) exact consistency; (2) different nouns but with the same meaning; (3) Keywords of affiliation (Figure 1).

Figure 1.

Flowchart of questions development and evaluation.

Question stratification

Two independent researchers, a professional doctor of hepatobiliary and pancreatic surgery who has worked in a tertiary hospital for at least 10 years, are tasked with stratifying cholelithiasis-related questions and answers according to their level of professionalism. They divide the questions into general and professional categories. General questions are defined as those that do not require specialized knowledge and are easily understood by the general public. Professional questions are defined as those requiring specialized knowledge, difficult for the general public to understand, and comprehensible only to doctors of related specialties. In cases of disagreement, the final decision is made after discussion between the two researchers.

Get answers from ChatGPT

Several studies have demonstrated that ChatGPT-4.0 outperforms other large language models^,^16,17 prompting its use in this study. Initially, questions are posed directly to ChatGPT-4.0 in Chinese to record answers. To check for repetition of answers and to eliminate the interference of other factors, questions are also asked in English and responses are recorded. When posing questions, a background condition is set before each query: “Please answer as a chief doctor of hepatobiliary and pancreatic surgery who has been working for at least 40 years and is familiar with the latest cholelithiasis research trends domestically and internationally:” to ensure contextual accuracy. Additionally, to avoid biases related to timing and user interaction, other accounts are used to pose questions and record answers 10 days after the first session, with the language of the question being Chinese. Complete answers are recorded in Microsoft Word.

Get answers from doctors

To adapt to clinical scenarios, where the diagnosis and treatment process often involves a multi-person medical team, we recruited five doctors (hepatobiliary and pancreatic surgeons with 5–40 years of experience in tertiary hospitals) to form a team for answering questions, ensuring comparability with ChatGPT's responses. The process was conducted in a double-blind format: the team was unaware of the standard answers and the study's purpose beforehand. The preformulated questions were answered using keywords, and each keyword was recorded only after agreement by at least three doctors. The entire process was documented by a surgeon assistant using Microsoft Excel.

Evaluate the quality and accuracy of answers

The quality and accuracy of responses from ChatGPT and the experts were evaluated by two independent researchers. In cases of differing opinions, the final decision was made through discussion.

For quality assessment, the DISCERN tool, a reliable health information quality assessment tool, was employed.¹⁸ It includes 16 questions across three sections evaluating the reliability of information, details about treatment, and overall assessment. Each question is rated on a scale of 1 to 5, where “No” scores 1 point, “Partially” scores 3 points, and “Yes” scores 5 points. The total score is the sum of all scores across the 16 questions, with ≤26 indicating very poor quality, 27–38 indicating poor quality, 39–50 indicating average quality, 51–61 indicating good quality, and ≥62 indicating excellent quality.

Accuracy was assessed using two methods to ensure thorough evaluation. The first method employed a modified matching scoring system¹⁹: 11 point indicates the keyword is completely incorrect; 2 points indicate a mix of correct and incorrect keywords; 3 points indicate the keyword is correct or similar but insufficient; 4 points indicate an exact match of the keyword. Additionally, the number of keywords in the standard answers and the number of correct keywords identified were tallied.

To address potential shortcomings in the modified matching scoring system, especially when a response includes only one correct keyword while the standard answer contains multiple, a keyword proportion evaluation system was implemented. This system rates the accuracy based on the percentage of correct keywords from the standard answer: (a) 0–25% is rated as very poor; (b) ＞25%–50% as poor; (c) ＞50%–75% as average; (d) ＞75%–100% as good.

Data analysis

Statistical analysis was conducted using SPSS software (version 26.0). For normally distributed measurement data, descriptive statistics were reported as mean ± SD, and comparisons between groups were made using the t-test. For data with a skewed distribution, descriptive statistics were presented as the median (±interquartile range), and group comparisons were conducted using the Mann–Whitney U test. Categorical data were reported as the number of cases (percentage), and group comparisons were performed using the chi-squared test. A P-value of less than .05 was considered statistically significant. Data visualization was performed using R packages such as “pheatmap” and “ggplotify,” as well as GraphPad Prism (version 9.5.1) and Origin Pro 2024.

Results

In this study, a total of 65 questions were developed according to the clinical practice guidelines for cholelithiasis published by EASL and JSGE, with each question containing keywords as standard answers (Table 1). After question stratification, 33 were classified as general questions (50.77%) and 32 as professional questions (49.23%). For ChatGPT, varying the background, times, languages, and accounts when asking questions resulted in answers that were differently expressed but involved almost identical keywords, indicating high repetitiveness in ChatGPT's responses. The study predominantly used Chinese to pose questions directly. Across the 65 questions, 177 keywords were identified, with 96 (54.24%) from general questions and 81 (45.76%) from professional questions. ChatGPT's responses included 242 keywords, of which 117 (48.35%) matched the standard answers. The doctors’ responses contained 190 keywords, with 140 (73.68%) matching the standard answers (Figure 2), demonstrating higher accuracy despite fewer total keywords, with a P-value of <.001, indicating a statistically significant difference.

Figure 2.

Number of keywords in the question.

Table 1.

Part of the questions and keywords display.

Lamination	Question	Keywords for the answer
General questions	1. Can gallstone disease be prevented?	Yes
	2. What are the common symptoms of gallbladder stones?	Asymptomatic; abdominal pain; nausea; vomiting; fever
	3. Is cholecystectomy recommended for asymptomatic gallbladder stones?	No
	4. What are the surgical options for choledocholithiasis?	Open; laparoscopic
	5. Are gallbladder stones a risk factor for gallbladder cancer?	Unknown
Professional questions	1. What are the possible complications of laparoscopic cholecystectomy?	Bile duct injury; bleeding; organ damage; bile leak; infection; shoulder pain; subcutaneous emphysema
	2. Is early surgical treatment superior to conservative treatment in patients with severe acute cholecystitis?	Yes
	3. Should patients with Mirizzi syndrome be treated surgically or endoscopically?	Surgery
	4. Who has better postoperative outcomes comparing one-stage and two-stage surgery in patients with gallbladder stones combined with choledochal stones?	Similar
	5. What are the surgical guidelines for hepatectomy in patients with intrahepatic ductal stones?	Intrahepatic cholangiocarcinoma; liver atrophy; non-surgical failure

Comparison between ChatGPT and doctors in general questions

Among the 33 general questions, ChatGPT scored 4 (3,4) points in the modified matching score system, while the doctors scored 4 (4,4) points, showing similar performance with a P-value of .856, which was not statistically significant (Figure 3). In terms of keyword proportion evaluation, ChatGPT had 3 (9.1%) questions rated as Very Poor, 2 (6.1%) as Poor, 3 (9.1%) as Average, and 25 (75.7%) as Good. The doctors had 2 (6.1%) questions rated as Very Poor, 4 (12.1%) as Poor, 3 (9.1%) as Average, and 24 (72.7%) as Good (Figure 4). The evaluations were similar, with a P-value of .829, indicating no statistically significant difference (Table 2).

Figure 3.

The scores of general and professional questions are in the modified matching score system. *: P > .05; ***: P < .05.

Figure 4.

The scores of general and professional questions are in the keyword proportion scores.

Table 2.

Keywords the score in the modified matching score system and keyword proportion evaluation.

	All questions			General questions			Professional questions
Score system	ChatGPT (n = 65)	Doctor (n = 65)	P-value	ChatGPT (n = 33)	Doctor (n = 33)	P-value	ChatGPT (n = 32)	Doctor (n = 32)	P-value
Modified matching score system	4(3,4)	4(3,4)	.024	4(3,4)	4(4,4)	.856	3(1,4)	4(3,4)	.004
Keyword proportion evaluation			.136			.829			.016
Very poor	13(20)	5(7.7)		3(9.1)	2(6.1)		10(31.1)	3(9.3)
Poor	9(13.8)	6(9.2)		2(6.1)	4(12.1)		7(21.9)	2(6.3)
Average	5(7.7)	5(7.7)		3(9.1)	3(9.1)		2(6.3)	2(6.3)
Good	38(58.5)	49(75.4)		25(75.7)	24(72.7)		13(40.7)	25(78.1)

Comparison between ChatGPT and doctors on professional questions

In the evaluation of 32 professional questions, ChatGPT scored 3 (1, 4) in the modified matching score, while the doctors scored 4 (3, 4), demonstrating a higher performance by the doctors with a P-value of .004, indicating a statistically significant difference (Figure 3). Regarding keyword proportion scores, ChatGPT had 10 (31.1%) questions rated as Very Poor, 7 (21.9%) as Poor, 2 (6.3%) as Average, and 13 (40.7%) as Good. Conversely, doctors had 3 (9.3%) questions rated as Very Poor, 2 (6.3%) as Poor, 2 (6.3%) as Average, and 25 (78.1%) as Good (Figure 4). The expert evaluation was superior, with a P-value of .016, also showing a significant difference (Table 2).

ChatGPT complete answer quality analysis

According to the DISCERN score, for general questions, the median score was 59, with the highest at 63 and the lowest at 46 points. For professional questions, the median score was 51, with the highest at 63 and the lowest at 34 points. The quality evaluation revealed that 4 (12.1%) of the general questions were rated as Average, 19 (57.6%) as Good, and 10 (30.3%) as Excellent, with no answers classified as Very Poor or Poor. For professional questions, 2 (6.3%) were rated as Poor, 11 (34.3%) as Average, 16 (50%) as Good, and 3 (9.4%) as Excellent (Table 3). Most of ChatGPT's answers exhibited good quality, and the answers to general questions were of higher quality compared to professional ones, with statistical significance (P < .001, P = .026).

Table 3.

The complete answer of ChatGPT is scored in DISCERN.

	General questions (n = 33)	Professional questions (n = 32)	P-value
Score			<.001
Median	59	51
Max	63	63
Min	46	34
Evaluation			.026
Very poor	0	0
Poor	0	2(6.3)
Average	4(12.1)	11(34.3)
Good	19(57.6)	16(50)
Excellent	10(30.3)	3(9.4)

Discussion

Although LLMs have been applied across various fields demonstrating significant potential,^20–22 ChatGPT emerges as the most notable chatbot among them. It has shown its utility in providing information to patients and doctors before and after treatment, potentially playing a crucial role. However, the content available specifically in the field of cholelithiasis is quite limited, and few studies have evaluated LLMs’ efficacy in addressing specialized clinical issues. In this study, 65 questions were developed based on the clinical practice guidelines for cholelithiasis, covering pathogeny, clinical manifestations, risk factors, and treatment options, and classified to quantitatively assess the response performance of ChatGPT and clinical experts on general and professional questions. ChatGPT processes these questions using advanced algorithms from a vast database, which contrasts with the more limited, independent answering capabilities of clinical experts. In practice, diagnosis and treatment often involve collaborative efforts from a multi-person medical team. To mimic this real clinical environment, a team of five clinical experts was recruited to ensure comparability with ChatGPT. Most existing studies assessing ChatGPT's performance involve experts directly evaluating the chatbot's responses,^23,24 which may introduce bias due to subjectivity or memory bias. To counter these limitations, this study utilized clinical practice guidelines to formulate standard answers and employed keyword comparison for evaluation, ensuring the feasibility, objectivity, and accuracy of the research.

ChatGPT's answers in this study involved 242 keywords, yet the accuracy rate was only 48.35%. This result highlights a characteristic of ChatGPT: while it can provide extensive information, it does not prioritize or emphasize the importance of specific answers, aligning with findings from Rachel and Walker.^25,26 This trait allows users to access a broader range of information, yet it does not deliver definitive guidance. For instance, when questioned about the common clinical manifestations of gallstones, ChatGPT included both common and rare symptoms; similarly, when developing treatment plans, ChatGPT enumerated various options without recommending a singular, preferred treatment modality. Given these characteristics, ChatGPT seems better suited as an educational tool rather than a decision-making aid in clinical settings.

Several recent studies have shown that ChatGPT's responses align closely with those of experts,^24,27,28 However, these studies may overlook potential biases such as subjective judgment, knowledge confusion, or memory lapses during the evaluation process by experts. Therefore, this study used the clinical practice guidelines for cholelithiasis from EASL and JSGE to develop questions and standard answers, focusing solely on keyword comparison to minimize the limitations observed in previous studies. Two scoring systems were employed to evaluate the consistency of ChatGPT's responses with those of experts. Overall, ChatGPT showed a positive accuracy, with a high median score on the modified matching score (4) and over 60% of responses rated as fair or good based on keyword proportion. However, the critical finding of this study concerns ChatGPT's performance on professional questions about cholelithiasis, where it demonstrated significant weaknesses. ChatGPT scored notably lower than experts in the modified matching score (3(1,4) vs 4(3,4), P = .004), with up to 53% of questions rated as very poor or poor in the keyword proportion evaluation. This suggests that ChatGPT has substantial limitations in addressing professional medical inquiries, often providing vague responses. For example, when asked whether “patients with Mirizzi syndrome should be treated surgically or endoscopically,” despite guidelines favoring endoscopy, ChatGPT would discuss both methods’ pros and cons without recommending a clear preference, which was counted as an error in this study. Additionally, some of ChatGPT's dichotomous answers were completely contrary to the guidelines. Another aspect to consider is ChatGPT's reliance on a vast corpus of internet text data, as acknowledged by its creator, OpenAI.²⁹ This dependency means that responses are based on historical internet information,³⁰ which may not reflect the most current clinical guidelines that are regularly updated, potentially leading to outdated answers.

Despite its limitations, ChatGPT's capacity to popularize science among the general public cannot be overlooked. In general questions that do not require specialized knowledge, ChatGPT's performance is comparable to that of experts, as evidenced by a P-value of .856 in the modified matching score test and a P-value of .829 in the keyword proportion rating. With a high median modified matching score (4) and 75.7% of questions rated as good, ChatGPT demonstrates its utility for patient education. Recent studies corroborate this, showing ChatGPT's effectiveness in educating patients about joint replacements, thyroid nodules,^31,32 aligning with the findings of this study.

Furthermore, the quality of information provided by ChatGPT's answers is high. Although there is a disparity in quality between responses to general and professional questions, the DISCERN scores for both are above 50, with no questions rated as Very Poor and 87.95% and 59.4% of questions classified as Good or Excellent, respectively. ChatGPT offers a comprehensive context for each answer and outlines the advantages and disadvantages of different options, facilitating a more informed decision-making process for patients. However, a significant limitation noted in ChatGPT's responses relates to the “Evidence Source” category in the DISCERN evaluation, where ChatGPT received its lowest scores. For all answers, ChatGPT fails to indicate the source and quality of the evidence, which is a critical aspect. The general and extensive nature of the responses makes it challenging for patients or doctors to make reliable judgments based on the answers provided.

Limitation

This study has several limitations. First, the standard answers were formulated according to the clinical practice guidelines for cholelithiasis issued by EASL and JSGE. However, clinicians often respond based on their actual clinical experience or the clinical practice guidelines of their own countries, which may differ from these international guidelines. Second, this study aimed to evaluate the performance of ChatGPT in a medical environment comprehensively and in detail, thus employing the method of keyword comparison for research. While this approach allows for a nuanced analysis, it is more challenging to implement than binary classification and can introduce subjective bias. Efforts were made to minimize this bias through pre-developed keyword definitions, but more scientific methods may be needed for future keyword formulation.

Conclusions

The findings of this study suggest that ChatGPT performs similarly to clinical experts in answering general questions related to cholelithiasis. It demonstrates excellent accuracy and provides high-quality information for general questions, meeting the information needs of the general public effectively. Nevertheless, in more specialized areas, careful attention and continuous evaluation will be necessary to ensure its accuracy, reliability, and safety in the medical field.

Footnotes

Acknowledgments

The authors acknowledge gratitude to all the staff who participated in the study.

ORCID iD

Tianyang Mao

Author contributions/CRediT

Tianyang Mao: conceptualization, methodology, resources, writing—original draft, writing—review and editing. Xin Zhao: conceptualization, methodology, resources, writing—original draft, writing—review and editing, and contributed equally as the first author. Kangyi Jiang: investigation, formal analysis. Qingyun Xie: investigation, formal analysis. Manyu Yang: investigation. Ruoxuan Wang: investigation. Fengwei Gao: project administration, investigation, supervision, writing.

Ethical considerations

As no patients are involved in this study, approval from the ethics committee is not required.

Informed consent

No clinical data, human specimens, or laboratory animals were used in this study. All information was obtained from public resources, and none of the data involved personal privacy concerns. In addition, this study did not involve any interaction with users; therefore, no informed consent is required.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

References

Farthing

Roberts

Samuel

, et al. Survey of digestive health across Europe: final report. Part 1: the burden of gastrointestinal diseases and the organisation and delivery of gastroenterology services across Europe. United European Gastroenterol J 2014; 2: 539–543.

Ahmed

Hudson

Innes

, et al. Effectiveness of conservative management versus laparoscopic cholecystectomy in the prevention of recurrent symptoms and complications in adults with uncomplicated symptomatic gallstone disease (C-GALL trial): pragmatic, multicentre randomised controlled trial. Br Med J 2023; 383: e075383.

Kim

Donahue

. Laparoscopic cholecystectomy. Jama 2018; 319: 1834.

Williams

Beckingham

El Sayed

, et al. Updated guideline on the management of common bile duct stones (CBDS). Gut 2017; 66: 765–782.

Manes

Paspatis

Aabakken

, et al. Endoscopic management of common bile duct stones: European society of gastrointestinal endoscopy (ESGE) guideline. Endoscopy 2019; 51: 472–491.

Zhou

Yao

, et al. Safety and efficacy of laparoscopic common bile duct exploration for the patients with difficult biliary stones: 8 years of experiences at a single institution and literature review. Surg Endosc 2022; 36: 718–727.

Kida

Yoshio

Minami

, et al. Successful endoscopic treatment of multiple large intrahepatic bile duct stones with benign choledochojejunal anastomotic stenosis. Endoscopy 2023; 55: E914–e915.

Zhang

Liu

, et al. Strategies of minimally invasive treatment for intrahepatic and extrahepatic bile duct stones. Front Med 2017; 11: 576–589.

Eysenbach

Kohler

. What is the prevalence of health-related searches on the world wide web? Qualitative and quantitative analysis of search engine queries on the internet. AMIA Annu Symp Proc 2003; 2003: 225–229.

10.

Lubowitz

. ChatGPT, an artificial intelligence chatbot, is impacting medical literature. Arthroscopy 2023; 39: 1121–1122.

11.

Cadamuro

Cabitza

Debeljak

, et al. Potentials and pitfalls of ChatGPT and natural-language artificial intelligence models for the understanding of laboratory medicine test results. An assessment by the European federation of clinical chemistry and laboratory medicine (EFLM) working group on artificial intelligence (WG-AI). Clin Chem Lab Med 2023; 61: 1158–1166.

12.

Liu

Wang

Liu

. Utility of ChatGPT in clinical practice. J Med Internet Res 2023; 25: e48568.

13.

Lee

. The rise of ChatGPT: exploring its potential in medical education. Anat Sci Educ 2023; 17: 926–931.

14.

Deng

Wang

Yangzhang

, et al. Evaluation of large language models in breast cancer clinical scenarios: a comparative analysis based on ChatGPT-3.5, ChatGPT-4.0, and Claude2. Int J Surg 2024; 110: 1941–1950.

15.

Yeo

Samaan

, et al. Assessing the performance of ChatGPT in answering questions regarding cirrhosis and hepatocellular carcinoma. Clin Mol Hepatol 2023; 29: 721–732.

16.

Lim

Pushpanathan

Yew

SME

, et al. Benchmarking large language models’ performances for myopia care: a comparative analysis of ChatGPT-3.5, ChatGPT-4.0, and google bard. EBioMedicine 2023; 95: 104770.

17.

Giannakopoulos

Kavadella

Aaqel Salim

, et al. Evaluation of the performance of generative AI large language models ChatGPT, google bard, and microsoft bing chat in supporting evidence-based dentistry: comparative mixed methods study. J Med Internet Res 2023; 25: e51580.

18.

Charnock

Shepperd

Needham

, et al. DISCERN: an instrument for judging the quality of written consumer health information on treatment choices. J Epidemiol Community Health 1999; 53: 105–111.

19.

Cakir

Caglar

Yildiz

, et al. Evaluating the performance of ChatGPT in answering questions related to urolithiasis. Int Urol Nephrol 2024; 56: 17–21.

20.

Grünebaum

Chervenak

Pollet

, et al. The exciting potential for ChatGPT in obstetrics and gynecology. Am J Obstet Gynecol 2023; 228: 696–705.

21.

Hristidis

Ruggiano

Brown

, et al. ChatGPT vs google for queries related to dementia and other cognitive decline: comparison of results. J Med Internet Res 2023; 25: e48966.

22.

Jin

Dobry

. ChatGPT for healthcare providers and patients: practical implications within dermatology. J Am Acad Dermatol 2023; 89: 870–871.

23.

Suárez

Díaz-Flores García

Algar

, et al. Unveiling the ChatGPT phenomenon: evaluating the consistency and accuracy of endodontic question answers. Int Endod J 2024; 57: 108–113.

24.

Patel

Hermann

Growdon

, et al. ChatGPT accurately performs genetic counseling for gynecologic cancers. Gynecol Oncol 2024; 183: 115–119.

25.

Walker

Ghani

Kuemmerli

, et al. Reliability of medical information provided by ChatGPT: assessment against clinical guidelines and patient information quality instrument. J Med Internet Res 2023; 25: e47479.

26.

Goodman

Patrinely

Stone

Jr. , et al. Accuracy and reliability of chatbot responses to physician questions. JAMA Netw Open 2023; 6: e2336483.

27.

Cakir

Caglar

Sekkeli

, et al. Evaluating ChatGPT ability to answer urinary tract infection-related questions. Infect Dis Now 2024; 54: 104884.

28.

Xue

Zhang

Gan

, et al. Quality and dependability of ChatGPT and DingXiangYuan forums for remote orthopedic consultations: comparative analysis. J Med Internet Res 2024; 26: e50882.

29.

Sinsky

Shanafelt

Ripp

. The electronic health record inbox: recommendations for relief. J Gen Intern Med 2022; 37: 4002–4003.

30.

Tan

Xin

. ChatGPT in medicine: prospects and challenges: a review article. Int J Surg 2024; 110: 3701–3706.

31.

Mika

Martin

Engstrom

, et al. Assessing ChatGPT responses to common patient questions regarding total hip arthroplasty. J Bone Joint Surg Am 2023; 105: 1519–1526.

32.

Campbell

Estephan

Sina

, et al. Evaluating ChatGPT responses on thyroid nodules for patient education. Thyroid 2024; 34: 371–377.