Abstract
Background
Describing complex orthopaedic surgical procedures and diagnoses to pediatric patients across a large age range and at varying levels of health literacy can be quite challenging. Generative artificial intelligence (GenAI) offers a potential solution by providing real-time age specific explanations for pediatric patients. The use of GenAI large language models (LLMs) could potentially allow providers to more effectively communicate complex medical problems in an age-appropriate and comprehensible manner. This study aims to validate the readability, accuracy, and comprehensiveness of responses produced by the ChatGPT-4 LLM (OpenAI, San Francisco, CA) when prompted to explain a specific pediatric orthopaedic procedure indicated for a given diagnosis to patients of 3 different age groups.
Methods
ChatGPT-4 explanations were analyzed for 43 pediatric orthopaedic procedures and tests across prompts for 3 different age groups: 6 years old, 10 years old, and 14 years old. Responses were assessed for readability using 6 quantitative readability indices. The accuracy and comprehensiveness of the responses were rated by 7 board-certified pediatric orthopaedic surgeons, who also compared them to their own explanations used in clinical practice.
Results
Each of the 6 indices showed a significant difference in readability between the 6-, 10-, and 14-year-old age group prompts (P < .001). The 6-year-old prompts were more readable than the 10- and 14-year-old prompts (P < .001), and the 10-year-old prompts were more readable than the 14-year-old prompts (P < .001) . Out of a 5-point scale, the mean (standard deviation) attending-rated accuracy for the artificial intelligence responses was 4.15 (0.81), with no significant differences in rated accuracy between procedures (P = .051) or age groups (P = .27). The mean attending-rated comfort was 3.95, with inter-rater reliability coefficients for accuracy and comfort of 0.92 and 0.94, respectively.
Conclusions
GenAI LLMs can generate age-specific explanations for common surgical procedures, treatments, and diagnostic markers in pediatric orthopaedics. Physicians validated the accuracy of its responses and demonstrated a high level of comfort using the GenAI explanations in clinical settings.
Key Concepts
-
(1)
An artificial intelligence (AI) language model adjusted its explanations based on the child’s age, providing simpler language for younger children and more advanced explanations for older children.
-
(2)
Across 6 established readability measures, explanations written for younger children were consistently easier to understand than those created for older age groups.
-
(3)
Seven board-certified pediatric orthopaedic surgeons found the AI-generated explanations to be highly accurate, with an average rating of 4.15 out of 5.
-
(4)
Physicians were generally comfortable with the use of AI-generated educational explanations in practice, and their ratings were highly consistent with one another.
-
(5)
When reviewed by a physician, generative AI could be a helpful tool for improving patient education, health literacy, and shared decision-making in pediatric orthopaedic care.
Level of Evidence
III/IV
Keywords: ChatGPT, Patient eduction, Healthcare literacy, Age-specific education
Introduction
Effective communication between physicians and patients is foundational to high-quality health care. In pediatric populations, interpersonal and communication skills are crucial to the effectiveness of the doctor-patient and doctor-caregiver relationship. Notably, the manner in which physicians explain a child’s medical condition significantly influences patient and family satisfaction [1,2]. Successful communication, particularly of surgical management, is associated with a higher likelihood of clinical success [3]. When children understand their medical condition and anticipated interventions, they are better equipped to cope with perceived threats related to illness, injury, hospitalization, or surgery [4]. Literature dating back to the 1970s consistently demonstrates that structured preparation for medical encounters positively influences children’s behavioral and emotional responses to hospitalization and surgical procedures [4,5].
Health literacy remains a critical barrier in pediatric care. At least 1 in 3 adolescents and young adults has low health literacy, with rates increasing among younger children. Furthermore, most pediatric health information is written above a 10th-grade reading level [6], even though the median (interquartile range) age of pediatric orthopaedic surgery patients in the United States is 10 [4,5,[7], [8], [9], [10], [11], [12], [13]] years [12], which is well below the average age of a 10th-grade student. This mismatch between reading level and patient comprehension limits children’s ability to meaningfully engage in their care.
Communicating effectively with pediatric patients presents unique challenges. The information presented must be conveyed to multiple audiences while adapting explanations to the child’s developmental stage. This challenge is amplified in specialties such as pediatric orthopaedics, where diagnoses, procedures, and diagnostic tools can be particularly complex.
A potential solution to the health literacy gap is the use of generative artificial intelligence (GenAI) large language models (LLMs) that can deliver age-tailored explanations in real time. GenAI refers to the broad category of artificial intelligence (AI) that produces new content in the form of text, images, audio, or code. LLMs fall under the umbrella of GenAI as they specifically generate text via statistical predictions that stem from analyzed language patterns. Currently, clinicians use GenAI to synthesize information, perform administrative tasks, and draft notes in the electronic medical record to optimize time spent with patients [13]. Herein, we sought to validate the performance of a GenAI LLM in generating meaningful, age-appropriate explanations for pediatric patients and to critically assess the readability, accuracy, and comprehensiveness of its outputs.
Materials and methods
A list of common pediatric orthopaedic conditions, diagnostic tests, and procedures was generated from our institution’s case log over a 90-day period. GenAI explanations for 3 different age level prompts, 6 years old, 10 years old, and 14 years old, across 43 different pediatric orthopaedic surgery procedures or diagnostic tests were collected and analyzed for readability, accuracy, and comprehensiveness. ChatGPT-4 (OpenAI, 2025), a GenAI LLM, was used to develop patient-directed explanations.
The GenAI prompts were constructed as follows: “Explain (specific procedure or diagnostic test) for (indicated diagnosis or condition) to a (6, 10, or 14) year old patient.” A complete list of procedures used to construct GenAI prompts is shown in Appendix I. The generated responses were not edited before analysis. This study did not need institutional review board approval as no patient-level data were utilized.
The readability of the collected GenAI responses was determined using 6 different quantitative readability indices, as previously described in the literature [[7], [8], [9], [10]], namely the Flesch Reading Ease Score, Flesch Kincaid Grade Level, Gunning Fog Index, Simple Measures of Gobbledygook (SMOG) Index, Coleman-Liau Index, and Automated Readability Index. In short, each index determines the readability of written text based on the total number of words, the number of words per sentence, the total number of sentences, and the number of syllables per word. A lower numerical score for each index, except the Flesch Reading Ease Score, indicates a response that is more readable.
The accuracy of the GenAI responses for 5 common procedures (Appendix I) was rated by 7 board-certified pediatric orthopaedic surgeons. The surgeon reviewers were also asked to rate their level of trust in GenAI to answer patient questions based on each response. This was conducted on a 5-point Likert scale and reported as a mean and standard deviation (SD). For example, a rating of 5 indicated perfect comfort, 4 indicated more comfort than discomfort, and 3 indicated indifference. The quantitative readability scores were reported as mean and standard error. The quantitative indices and attendings’ ratings were compared using 1-way analysis of variance and Fisher’s least significant difference post hoc tests, where applicable, with a significance level of ɑ = 0.05. The reliability of the attendings’ survey was determined by using the intraclass correlation coefficient. The analyses were performed using Microsoft Excel (Microsoft Corporation, Redmond, WA) and IBM SPSS (IBM Corporation, Armonk, NY).
Results
The quantitative readability indices scores are summarized in Table 1. Each of the 6 indices showed a significant difference in readability between the 6-, 10-, and 14-year-old age-group prompts (P < .001). The 6-year-old prompts were more readable than the 10- and 14-year-old prompts (P < .001), and the 10-year-old prompts were more readable than the 14-year-old prompts (P < .001) (Fig. 1). An example prompt and outputs are provided in Table 2.
Table 1.
Quantitative reading indices scores.
| 6-year-old prompts | 10-year-old prompts | 14-year-old prompts | F (sig) | |
|---|---|---|---|---|
| Flesch Reading Ease Score, mean (SE) | 77.2 (0.8) | 73.4 (1.0) | 68.2 (1.0) | 23.41 (<0.001) |
| Flesch Kincaid Grade Level, mean (SE) | 6.6 (0.2) | 7.4 (0.2) | 8.2 (0.2) | 17.04 (<0.001) |
| Gunning Fog Index, mean (SE) | 8.8 (0.2) | 10.1 (0.2) | 11.1 (0.2) | 29.57 (<0.001) |
| SMOG Index, mean (SE) | 8.6 (0.1) | 9.6 (0.2) | 10.6 (0.2) | 41.06 (<0.001) |
| Coleman-Liau Index, mean (SE) | 6.3 (0.1) | 7.0 (0.2) | 7.9 (0.2) | 29.86 (<0.001) |
| Automated Readability Index, mean (SE) | 6.0 (0.2) | 7.1 (0.2) | 7.8 (0.2) | 18.13 (<0.001) |
SE, standard error.
Figure 1.
GenAI LLM (ChatGPT 4) generates age-specific information when prompted to explain common pediatric orthopaedic procedures to children, as measured by (A) Flesch Kincaid Reading Ease Score and (B) five other readability indices. ∗P < .05, ∗∗P < .01. GenAI, generative artificial intelligence; LLM, large language model; SMOG, Simple Measure of Gobbledygook.
Table 2.
Example GenAI LLM prompts and outputs.
| Prompt | Output |
|---|---|
| Explain Posterior spinal instrumentation and fusion to a 6 year old patient. | Sure. So imagine you have a very tall and wobbly tower of blocks, and in order to make sure it doesn’t fall over, you need to add some extra support—maybe more blocks or maybe some tape. Now let’s think about your back, which is kind of like that tower of blocks. It is made up of many small pieces called vertebrae. If your back is like a wobbly tower, the doctor needs to add some support so it doesn't hurt or cause problems when you move. This operation, called posterior spinal instrumentation and fusion, does just that. The doctor uses special tools, kind of like extra blocks and tape, these ones are called screws and rods. They help hold the vertebrae together so they can grow into one solid piece. This is like gluing blocks together to make our tower strong and straight. This helps your back stay in a good shape, prevents more wobbling, and stops any pain or problems with moving. And don’t worry; you won’t feel anything during the operation because you’ll be sleeping under the safe watch of your doctor and his team. |
| Explain Posterior spinal instrumentation and fusion to a 10 year old patient. | Okay, sure! So, imagine your spine is like a tall stack of coins. It helps you stand up straight, bend, and move around. Sometimes, there’s a problem with the stack—like a coin is damaged or out of place. When that happens, your back can hurt a lot or you might have trouble moving. Now, think of “Posterior spinal instrumentation and fusion” like a super special repair job for your stack of coins. The doctor uses tools, which we can think of as clips and stickers to help those troubled areas. “Posterior” means it’s happening at the back side of your spine. The “instrumentation” part is when the doctor places these special clips (which are really screws and rods) to hold your spine in place. “Fusion” is a neat part—it’s like magic glue, but it’s actually a graft, kind of like a super-healing putty. This is placed around the problem area on your spine and over time, it helps your body grow new bone. This new bone sticks or ‘fuses’ some of your spine’s coins together so they don’t move out of place and cause problems anymore.” This repair job might sound a bit scary but remember, the doctors who do this are like expert toy builders, and they make sure that you’re sleeping and can’t feel anything during the procedure. Their goal is to help you feel better and do all your favorite things without any trouble or pain. |
| Explain Posterior spinal instrumentation and fusion to a 14 year old patient. | Sure, I understand that you might be scared or confused about this process, but I’ll break it down for you so that you can understand it easily. Imagine you have a wobbly tower made out of blocks, and it’s your job to stop it from falling over or wobbling. The term “posterior spinal instrumentation and fusion” refers to a surgery that’s a little like fixing that wobbly tower. But in this case, the tower is your spine.
You’ll be asleep under general anesthesia during the whole procedure, and when you wake up, the doctors will be there to ensure the process went smoothly and you’re comfortable. Afterward, you’ll need to rest and recover, and there will be follow-up visits to the doctor’s office to make sure everything is healing as it should. |
GenAI, generative artificial intelligence; LLM, large language model.
The mean (SD) accuracy rating for the ChatGPT responses was 4.15 (0.81) out of 5, with no significant differences in rated accuracy between procedures (P = .051) and age groups (P = .27). The mean (SD) comfort rating was 3.95 (1.01) out of 5, with no significant differences between procedures (P = .51) and age groups (P = .31). The inter-rater reliability was excellent, with intraclass correlation coefficients for accuracy and comfort of 0.92 and 0.94, respectively.
Discussion
This study provides a focused evaluation of GenAI’s potential in delivering age-appropriate explanations of clinical diagnosis, treatment modalities, and plans of care to facilitate patient understanding and autonomy in the pediatric orthopaedic setting. The need to educate both parents and young patients in pediatric surgery is unique to this area of medicine. GenAI can enhance patient health literacy and improve patient-physician communication by tailoring information on an age-specific level.
The main findings of our study showed that across 5 different readability indexes, a GenAI LLM increased the complexity of its language output in an age-specific manner, suggesting its potential as a supportive clinical tool in pediatric care. The high level of physician inter-rater reliability and intraclass correlation coefficients for accuracy (0.92) and comfort (0.94) demonstrates a reassuringly high level of trust 7 attending surgeons have in its ability to provide accurate and age-specific health information.
These findings are consistent with other studies that have evaluated GenAI’s ability to produce age-appropriate health information. For example, ChatGPT’s ability to simplify existing health information was assessed to reduce an average grade reading level of 12.8 to 11, with the complexity of language decreased from 22.8% to 14.4% while maintaining, on average, 80% of the key messages from the original text [14]. The use of AI in clinical practice has previously been explored in the context of assisting physicians in various aspects of care, including diagnostic assistance, triaging patients, streamlining documentation, and enhancing patient health literacy [15]. In a comparison study between LLMs, ChatGPT-3.5 and ChatGPT-4 achieved better-tailored grade-level outputs with a broader range, in comparison to their Microsoft Bing and Google counterparts [16]. ChatGPT’s ability to translate complex medical terminology into an individually tailored language level will make health information more accessible to a broader patient population. Now that physician approval has been established, a future study should assess patient-reported satisfaction and comprehension in response to the generated outputs.
This study has several limitations. First, the findings are limited by the specific prompts used when conducting the surveys as it is unclear how changing the parameters to less prevalent health conditions will affect the language output. Second, physician comfort and accuracy were assessed by 7 attending surgeons at a single institution, which may limit generalizability. Third, readability indices are proxy measures of comprehension and may not fully capture how pediatric patients or caregivers interpret AI-generated content. Additionally, discrepancies between patients’ ages and cognitive abilities were not accounted for. In future applications, the AI prompt may require additional context or age adjustment, tailored to the specific cognitive abilities of the patient or patient’s family. Fourth, this study did not evaluate clinical outcomes such as patient adherence or satisfaction. In a diverse patient population, the cultural bias of GenAI LLMs poses a probable communication barrier. Tao et al. [17] found that multiple models of ChatGPT tended to express cultural values from English-speaking countries, but with intentional “cultural prompting,” the text output can be manipulated to align with the specified culture. Future studies should be conducted across multiple specialties and institutions to validate utility across the health care spectrum and to assess real-world patient outcomes. The high level of physician trust and comfort in this study demonstrates the potential for GenAI to be utilized in clinical settings. However, human oversight remains essential to ensure the information’s safety, accuracy, and completeness.
The integration of GenAI into patient care with the specific aim of improving health literacy represents a significant stride in leveraging AI. One could envision using age-targeted GenAI to improve how information is communicated with patients through handouts, patient portals, after-visit summaries, online content, and child life material. Incorporating GenAI into pediatric orthopaedic care holds significant promise, offering the potential to enhance patient education, promote shared decision-making, and improve health care outcomes.
GenAI LLMs demonstrate the ability to generate age-calibrated, readable explanations for common surgical procedures, treatments, and diagnostic markers in pediatric orthopaedics, with language complexity increasing significantly across age groups on 5 validated readability indices. Physicians validated the accuracy of its responses and demonstrated a high level of comfort using the GenAI explanations in clinical settings.
Additional links
- •
-
•
Explanation: This article directly supports our study by establishing a validated precedent, in https://www.jposna.com/article/S2768-2765(24)00251-7/fulltext JPOSNA® itself, for using pediatric orthopaedic surgeons to evaluate the accuracy of large language model (LLM)–generated responses to common pediatric orthopaedic questions, benchmarked against Pediatric Orthopaedic Society of North America (POSNA)’s own OrthoKids platform. Their finding that ChatGPT responses agreed with specialty-society material in 93% of cases confirms the clinical plausibility of artificial intelligence (AI)–generated patient education in our field, but their study was limited to a single, generic reading level and did not address whether responses could be tailored to the developmental stage of the child or caregiver audience. Our study builds directly on this foundation by extending the evaluation to age-specific explanations and by adding a readability dimension alongside surgeon-rated accuracy and usefulness, addressing the exact next-step question their work leaves open.
- •
-
•
Explanation: This is the landmark, high-impact study that established the foundational rationale for investigating LLMs as a tool for patient education, blinded health care professionals preferred ChatGPT’s responses to verified physicians' responses 79% of the time, with markedly higher ratings for both quality (3.6 fold) and empathy (9.8 fold). This paper is essential to our justification because it provides top-tier, JAMA-level evidence that LLMs are capable of producing clinically meaningful patient-facing communication, which is the premise our study depends on before moving to the more specific questions of age-tailoring and pediatric orthopaedic applicability. Citing it anchors our work to the broader, high-credibility literature demonstrating that AI-generated patient education is not merely a novelty but a rigorously studied modality, reinforcing why systematically evaluating it in our pediatric orthopaedic population is both warranted and timely.
Declaration of competing interests
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Ethics approval and consent
Patient consent was not required for this study because no patients were enrolled, no patient information was used, and all evaluations were performed by physician reviewers. The authors declare that no patient consent was necessary as no images or identifying information is included in the article.
Funding
None declared.
Author contributions
Cole T. Payne: Writing – original draft, Methodology, Investigation, Data curation, Conceptualization. Andrew K. Morse: Writing – review & editing, Writing – original draft, Data curation, Conceptualization. Brendan T. O’Reilly: Writing – review & editing, Writing – original draft, Methodology, Data curation, Conceptualization. Giselle K. Henry: Writing – review & editing, Writing – original draft, Methodology, Investigation, Conceptualization. Lindsay M. Crawford: Writing – review & editing. Brennan Roper: Writing – review & editing. Jessica Traver: Writing – review & editing. Rohini Vanodia: Writing – review & editing, Writing – original draft, Project administration, Conceptualization. Alfred A. Mansour: Writing – review & editing, Writing – original draft. Surya N. Mundluru: Writing – review & editing. Shiraz A. Younas: Writing – review & editing. Timothy C. Borden: Writing – review & editing, Writing – original draft, Supervision, Formal analysis, Conceptualization.
Footnotes
Investigation performed at McGovern Medical School at UTHealth Houston.
Appendix I
List of procedures analyzed. ∗ Represents prompts asked for physician survey on accuracy and comfortability.
| Spica cast removal |
| Right distal femoral epiphysiodesis∗ |
| Posterior spinal instrumentation and fusion∗ |
| Tibia and fibula lengthening |
| Percutaneous pinning of right elbow∗ |
| Bilateral femur hardware removal∗ |
| Left talocalcaneal coalition excision |
| Left anterior tibialis tendon transfer |
| Bilateral percutaneous tendo achilles lengthening |
| Scar revision on left forearm |
| Deep implant removal of left forearm |
| Cast application for humeral shaft fracture |
| Left open adductor tenotomy |
| Left hip arthrogram |
| Left hip arthroscopy with labral repair |
| Osteoplasty |
| Periacetabular osteotomy |
| Right knee scope and anterior cruciate ligament repair∗ |
| Medial meniscus repair |
| Knee scope retrieval of loose body |
| Right hip osteotomy |
| Left hip arthroscopic femoroplasty |
| Open reduction and internal fixation of left distal fibula |
| Closed reduction and percutaneous pinning of left supracondylar fracture |
| Removal of foreign body and irrigation and debridement of right great toe |
| Bilateral lateral distal femur and proximal tibia physeal tethering |
| Mehta casting for scoliosis |
| Getting an MRI scan |
| Getting a CT scan |
| Undergoing general anesthesia |
| Getting an X-ray |
| Undergoing local anesthesia |
| Ankle foot orthosis |
| Getting an electromyogram |
| Undergoing external fixation |
| Thoracolumbosacral orthosis |
| Syndactyly release |
| Elbow dislocation reduction |
References
- 1.Mărginean C.O., Meliţ L.E., Chinceşan M., Mureşan S., Georgescu A.M., Suciu N., et al. Communication skills in pediatrics - the relationship between pediatrician and child. Medicine. 2017;96(43) doi: 10.1097/MD.0000000000008399. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Benson B.J. Domain of competence: interpersonal and communication skills. Acad Pediatr. 2014;14(2 Suppl):S55–S65. doi: 10.1016/j.acap.2013.11.016. [DOI] [PubMed] [Google Scholar]
- 3.Stewart M.A. Effective physician-patient communication and health outcomes: a review. CMAJ. 1995;152(9):1423–1433. [PMC free article] [PubMed] [Google Scholar]
- 4.Wolfer J.A., Visintainer M.A. Pediatric surgical patients' and parents' stress responses and adjustment as a function of psychologic preparation and stress-point nursing care. Nurs Res. 1975;24(4):244–255. [PubMed] [Google Scholar]
- 5.Melamed B.G., Siegel L.J. Reduction of anxiety in children facing hospitalization and surgery by use of filmed modeling. J Consult Clin Psychol. 1975;43(4):511–521. doi: 10.1037/h0076896. [DOI] [PubMed] [Google Scholar]
- 6.Sanders L.M., Federico S., Klass P., Abrams M.A., Dreyer B. Literacy and child health: a systematic review. Arch Pediatr Adolesc Med. 2009;163(2):131–140. doi: 10.1001/archpediatrics.2008.539. [DOI] [PubMed] [Google Scholar]
- 7.Abu-Heija A.A., Shatta M., Ajam M., Abu-Heija U., Imran N., Levine D. Quantitative readability assessment of the internal medicine online patient information on Annals.org. Cureus. 2019;11(3) doi: 10.7759/cureus.4184. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Bange M., Huh E., Novin S.A., Hui F.K., Yi P.H. Readability of patient education materials from RadiologyInfo.org: has there been progress over the past 5 years? AJR Am J Roentgenol. 2019;213(4):875–879. doi: 10.2214/AJR.18.21047. [DOI] [PubMed] [Google Scholar]
- 9.Watad A., Bragazzi N.L., Brigo F., Sharif K., Amital H., McGonagle D., et al. Readability of Wikipedia pages on autoimmune disorders: systematic quantitative assessment. J Med Internet Res. 2017;19(7) doi: 10.2196/jmir.8225. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Skalitzky M.K., Gulbrandsen T.R., Lorentzen W., Gao B., Shamrock A.G., Weinstein S.L., et al. Health literacy in clubfoot: a quantitative assessment of the readability, understandability and actionability of online patient education material. Iowa Orthop J. 2021;41(1):61–67. [PMC free article] [PubMed] [Google Scholar]
- 11.Desai P.P., Pandya S.V. Communicating with children in healthcare settings. Indian J Pediatr. 2013;80(12):1028–1033. doi: 10.1007/s12098-013-0969-z. [DOI] [PubMed] [Google Scholar]
- 12.Varas-Rodriguez Emil, Hanson K., Habermann E.B., Grigoriou E., Milbrandt T.A., Larson A.N., et al. Adverse outcomes in pediatric orthopaedic surgery: insights from the NSQIP pediatric database. J Pediatr Orthop. 2025;45(10):e1022–e1029. doi: 10.1097/BPO.0000000000003063. [DOI] [PubMed] [Google Scholar]
- 13.Scott I.A., Reddy S., Kelly T., Miller T., van der Vegt A. Using generative artificial intelligence in clinical practice: a narrative review and proposed agenda for implementation. Med J Aust. 2025;223(11):664–672. doi: 10.5694/mja2.70057. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Ayre J., Mac O., McCaffery K., McKay B.R., Liu M., Shi Y., et al. New frontiers in health literacy: using ChatGPT to simplify health information for people in the community. J Gen Intern Med. 2024;39(4):573–577. doi: 10.1007/s11606-023-08469-w. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Xiao D., Meyers P., Upperman J.S., Robinson J.R. Revolutionizing healthcare with ChatGPT: an early exploration of an AI language model's impact on medicine at large and its role in pediatric surgery. J Pediatr Surg. 2023;58(12):2410–2415. doi: 10.1016/j.jpedsurg.2023.07.008. [DOI] [PubMed] [Google Scholar]
- 16.Amin K.S., Mayes L., Khosla P., Doshi R. ChatGPT-3.5, ChatGPT-4, Google Bard, and Microsoft bing to improve health literacy and communication in pediatric populations and beyond. arXiv preprint arXiv:231110075. 2023 [Google Scholar]
- 17.Tao Y., Viberg O., Baker R.S., Kizilcec R.F. Cultural bias and cultural alignment of large language models. PNAS Nexus. 2024;3(9):346. doi: 10.1093/pnasnexus/pgae346. [DOI] [PMC free article] [PubMed] [Google Scholar]