Abstract
Objective
Many caregivers of children with cleft lip and palate experience a high level of anxiety throughout their child's medical and surgical care. We aim to evaluate artificial intelligence (AI) as a tool to mitigate these feelings and can aid clinicians in the development of robust pediatric educational materials for caregivers and families.
Design
Thirteen of the most common postoperative questions following cleft lip and/or palate repair were developed by an expert panel of senior Pediatric Plastic Surgeons and were posed to ChatGPT. Professional answers from the expert panel were provided and compared to responses from ChatGPT. A literature review was also conducted to generate a new support model for caregivers with children undergoing a surgical procedure.
Setting
Department of Pediatric Plastic Surgery at a metropolitan Children's Hospital.
Participants
Senior Pediatric Plastic Surgeons at a metropolitan Children's Hospital.
Interventions
None.
Main Outcome Measure
The primary outcome was to determine the ability of ChatGPT to respond to common postoperative questions and to develop a model for AI assistance in family-centered perioperative care.
Results
ChatGPT had a postoperative question response accuracy rate of 69% when compared with subject matter expert responses, with its greatest errors being information errors. An extensive literature search revealed that AI can assist in multiple traditional perioperative strategies to reduce caregivers and patient anxiety.
Conclusions
Artificial Intelligence can help to reduce the burden of generating patient education materials as well as support caregivers in multiple aspects and perioperative care.
Introduction
Cleft lip and palate (CLP) occurs in 1 out of 700 live births per year globally.1,2 Primary surgical repair of the cleft lip and/or palate is the most common treatment modality for this congenital difference and on average occurs between 6 to 12 months of age.3,4 The young age at which this procedure occurs, compounded with caregiver separation anxiety, lack of control over the environment, and the risk of complications can lead to high levels of anxiety in caregivers with children with CLP.5–8 Furthermore, CLP families often feel a sense of inadequate medical support and low self-esteem in their caregiving capabilities.2,9 Preoperative anxiety is a common phenomenon surrounding pediatric procedures and affects 74% of caregivers.10,11 High levels of parental stress or anxiety can drastically impact the child's experience and even lead to higher levels of postoperative pain.12–14
A critical tool to mitigate caregiver anxiety is through the use of comprehensive patient educational materials. Studies have shown that an increase in written communication provided to caregivers prior to their child's surgery can significantly decrease levels of caregiver anxiety.15–18 In addition, low caregiver health literacy is associated with higher pediatric emergency department medical visits and may increase medication errors.19,20 Therefore, it is fundamental to maintain and provide high quality, informative, and easy-to-read patient education materials to caregivers when a child requires medical or surgical care. As a department-wide effort, the Department of Pediatric Plastic Surgery at a metropolitan Children's Hospital launched a quality improvement project in February 2023 to improve postoperative patient education materials for pediatric patients and their families. Improving patient education materials can be labor intensive and this provided an opportunity to apply and evaluate the use of AI in this context.
Chat Generative Pre-Trained Transformer, or better known as ChatGPT (Open AI, San Francisco, CA, USA) is an artificial intelligence (AI) chatbot that can provide multifaceted responses across multiple disciplines. 21 In the short time since its inception in November 2022, ChatGPT has become a major disrupter in the technology industry. In the medical field, many researchers have sought to determine how extensive ChatGPT's abilities may be. Kung et al. and Gilson et al. have found that ChatGPT possesses the ability to pass the United States Medical Licensing Exam.22,23 This finding generated a vast interest and spawned many studies in academic medicine which aimed to determine the extent of this powerful technology's abilities and the potential applications of this tool in medicine.
This study aimed to determine if AI can potentially provide appropriate responses to the most commonly asked postoperative questions of parents or guardians of children who underwent cleft lip and/or palate repair. This study also aimed to propose an AI-assisted supportive framework for caregivers with children with cleft lip/palate. Both could serve as potential tools for clinicians to develop patient education materials and mitigate caregiver anxiety with the assistance of AI.
Methods
The need for ethics approval and the need to obtain consent was not required for this non-interventional study.
Content Selection and First Draft of Patient Education Material
As a pilot project, CLP repair was chosen as CLP is the most common craniofacial malformation. The most common postoperative questions were compiled by an expert panel at a metropolitan Children's Hospital. Thirteen of the most asked postoperative questions were selected, and two senior Pediatric Plastic Surgeons drafted their professional answers to generate patient education materials (IG, MSG).
ChatGPT Responses to Postoperative Questions
The same thirteen questions were posed to ChatGPT (ChatGPT 3.5 February 13 Version) and its answers were recorded.
Evaluation of ChatGPT Responses
The answers generated by ChatGPT were evaluated for accuracy by the same two senior Pediatric Plastic Surgeons (IG, MSG). To prevent any bias, the experts were asked to evaluate the answers provided by ChatGPT independently and were blinded to the professional answers written during the expert panel.
Answer generated by ChatGPT that were found to be erroneous were categorized into three groups: information error, statistical error, or multiple errors using previously published methodology. 22 Information error was defined as the ChatGPT material not conveying correct information. Statistical error was defined as ChatGPT providing incorrect numbers, with or without calculation. Multiple errors were defined as errors that have a combination of both information and statistical error.
Literature Search to Develop an AI Assisted, Caregiver-Centered Support Model
In parallel to determining the efficacy of an AI chatbot in developing patient education materials, an extensive PubMed and Google Scholar literature search was conducted to identify and incorporate AI into a supportive model for caregivers of children who require surgical treatment. Terms queried included “pediatric surgery”, “cleft palate surgery”, “cleft/lip repair” or/and “caregiver burden”, “parents with cleft/palate children”, or/and “support framework”. AI associated materials were searched using terms “AI” or/and “reduce caregiver burden”, “communication”, “pediatric surgery”, “support framework”. Literature which could help develop the support model was reviewed, incorporated, synthesized, and is presented with our results.
Results
Accuracy and Quality of ChatGPT Responses
The responses generated by ChatGPT were recorded and compared to the professional responses listed in Table 1. Overall, ChatGPT's answers were verbose yet accurate with some notable incongruencies. ChatGPT correctly normalized some common postoperative scenarios such as postoperative bleeding (#1, #7), postoperative halitosis (#2), or changes in postoperative appearance (#8). However, ChatGPT seemed to have limited knowledge of postoperative care such as avoiding using spoons or forks for feeding (#4). In addition, while surgeons recommended caregivers to gently put nasal stents and tubes back in if they fall out, ChatGPT did not recommend doing so (#12). Lastly, ChatGPT seemed to lack knowledge of various specific surgical techniques of cleft lip/palate repair. For example, when ChatGPT was asked about an “open hole” developing after the surgery, it did not realize these could be lateral relaxing incisions rather than surgical dehiscences or postoperative fistulas (#3). ChatGPT also confused cleft lip repair with cleft palate repair (#8).
ChatGPT Responses Compared to Professional Answers.
Overall, ChatGPT's responses provided excessive and unnecessary information leading to much lengthier responses than a professional answers. The accuracy of ChatGPT generated responses was 69%, with most errors being information errors (Table 2).
Analysis of ChatGPT-Generated Responses.
AI-assisted Perioperative Support Model
A proposed model of AI-assisted pediatric perioperative care to reduce the burden on caregivers was generated and is represented in Figure 1. The traditional model was adapted from previous work by Chorney et al. and Santapuram et al.8,24 In Chorney et al., to support caregivers of pediatric patients who required surgical treatment, the family-centered pediatric perioperative care system can be categorized into (1) family factors, (2) preoperative factors, (3) intraoperative factors, (4) postoperative factors, and (5) provider/system factors.

Schematic of the traditional perioperative support model versus our proposed AI-assisted model.
Details of the Components of the Traditional Perioperative Support Model and the AI-Assisted Model.
Discussion
This study explored ChatGPT's ability to produce patient education materials and an AI-assisted perioperative support model for cleft lip/palate patients. Though ChatGPT struggled to compute and understand granular differences in surgical techniques and its answers were lengthy and provided unnecessary, and sometimes irrelevant, information, its answers to the postoperative questions were overall accurate. This aligns with the current literature which ascribes some level of medical knowledge to ChatGPT.22,23,25 As suggested by our model, there is an increase in utility (or at least the consideration of the utility) of AI in the medical field. This posits that AI might be able to decrease caregivers’ emotional burden through multiple avenues including AI-assisted patient education material generation, audiovisual interpretation, therapeutic play, emotional support and coping skills (via a chatbot that provides tailored supportive conversations to manage anxiety and stress for caregivers 24/7), and communication between providers and caregivers.8,24 AI also has the capability in language translation for patients and caregivers and translation of complex medical information into varying reading levels.26,27
Ensuring a high level of accessibility and quality of educational materials distributed to patients and caregivers by providers or healthcare systems requires an extensive amount of time from experts to develop and maintain these materials. Additionally, given the nature of AI models constantly updating its algorithm, more advanced versions of ChatGPT will be able to provide more accurate medical knowledge across more fields, with some studies already investigating the use of ChatGPT to provide quality and empathic patient messages.28–30 Furthermore, with ChatGPT 4 providing the capability to analyze images, this unlocks a substantial number of possibilities for ChatGPT to aid surgeons in postoperative surgical concerns for patients with limited access to healthcare.
This pilot study contains several strengths. The ChatGPT-generated answers were evaluated in terms of information accuracy, statistical accuracy, and a combination of both, allowing for an improved ability to identify any erroneous elements. This study also offered a unique caregiver-centered model which incorporates AI to reduce caregiver experienced anxiety and postoperative burden of care. Additionally, the postoperative questions generated by our expert panel align closely with previously published literature, offering a general application for all healthcare providers in CLP management. 31 Moreover, answers provided by our expert panel align with the general consensus of other published patient education resources.32,33 The senior craniofacial surgeons included in the expert panel also have standardized postoperative practices as a part of their cleft team consensus practices. Therefore, inaccuracies in the answers provided by ChatGPT can be accredited to its lack of medical knowledge rather than discourse in the literature.
Despite these strengths, there are some limitations to consider regarding the use of AI in patient education materials. To begin, ChatGPT's algorithm is constantly evolving. Our study did not use the newest version of ChatGPT (ChatGPT 4) as it is currently a paid service and ChatGPT4 was published after inception of this study. 34 As the authors do not have access to change the algorithm of ChatGPT (to increase the accuracy of the answers), we realized that pursing a high accuracy rate of ChatGPT generated materials was not the goal of this study, but rather it was a proof of concept. In addition, it is important to note that the purpose of using ChatGPT is not to replace healthcare providers in generating educational materials, but rather to supplement and reduce the burden of this often time-consuming process. Tone, body language, and other nonverbal communication also contributes to quality communication between a provider and patient/caregiver and subsequently, the lowering of caregiver and child perioperative anxiety. 35 Given that AI cannot currently match this level of communication, this limitation should be considered when using AI in caregiver-centered perioperative care. Further validation methods are needed to truly determine the effect of this interventional method.
As the field of AI continues to grow, it is important for medical researchers to elucidate how this technology can be applied to different facets of healthcare. With virtual capabilities becoming more popular in healthcare, more companies are exploring the application of AI in patient care to provide immediate support to caregivers and patients, ultimately increasing the accessibility.36,37 A large language AI model such as ChatGPT can be a critical tool for clinicians to efficiently provide high quality care for patients and their caregivers. With consideration to its limitations, we believe this tool could unlock a sizable number of new applications for surgeons to explore and provide much needed support to both physicians and caregivers.
Conclusion
With the substantial incidence of CLP and the high levels of stress and anxiety caregivers can experience when their child undergoes surgical repair, it is important for clinicians to seek ways to reduce this anxiety in caregivers. Educational materials and multifaceted support models have shown to improve caregiver anxiety and improve outcomes in pediatric procedures. AI could supplement the current traditional methods for caregiver-centered perioperative care and support clinicians in multiple aspects for pediatric surgical procedures.
Footnotes
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.
