LLMs for Atrial Fibrillation Patient Education: Cardiologist vs. Non-Medical Professional Perceptions of ChatGPT and Google Gemini Quality
What the Researchers Tested
This pilot study employed a straightforward comparative analysis design. The core methodology involved submitting a standardized clinical prompt, "Explain atrial fibrillation and how to treat it to a patient," to two prominent large language models (LLMs): ChatGPT and Google Gemini. The generated text responses were then disseminated to a diverse participant group via a Google Forms survey. The survey instrument was meticulously crafted, incorporating items derived from established tools for assessing patient education materials, specifically the Patient Education Materials Assessment Tool for Printable Materials (PEMAT-P) and the DISCERN criteria. These criteria encompass aspects such as reliability, clarity, and organizational structure. Participant evaluation involved rating each aspect on a 5-point Likert scale. The study cohort consisted of 28 individuals, segmented into two distinct groups: medical professionals (cardiologists and cardiology fellows, n=11) and non-medical professionals (college graduates not employed in the medical field, n=17). Statistical analysis, utilizing the Wilcoxon signed-rank test, was performed to compare mean ratings for various quality indicators between the two AI platforms and across the different participant groups, thereby identifying significant perceptual differences in the assessment of AI-generated educational content.
What the Study Found
No significant overall quality differences were observed between ChatGPT and Google Gemini for explaining atrial fibrillation and its treatment to patients.
Cardiologists rated AI-generated responses significantly higher than non-medical professionals for perceived bias (mean 3.82 vs mean 4.33; P=.04).
Cardiologists rated AI-generated responses significantly higher for explaining consequences of no treatment (mean 2.85 vs mean 1.86; P=.005) compared to non-medical professionals.
Non-medical professionals rated structural and organizational features more favorably than cardiologists, including visual cues (mean 3.97 vs mean 2.48; P<.001), informative headers (mean 4.77 vs mean 3.95; P=.001), and concise sections (mean 4.71 vs mean 4.29; P=.04).
Non-medical professionals also rated identifiable patient actions (mean 4.21 vs mean 3.19; P=.04) and references to additional resources (mean 2.65 vs mean 1.71; P=.006) more favorably than cardiologists.
Why This Matters to Clinical Practice
The increasing integration of AI into healthcare necessitates a rigorous evaluation of its utility, particularly in patient education. Accurate and comprehensible patient understanding of conditions like atrial fibrillation is paramount for adherence to treatment, lifestyle modifications, and early recognition of complications such as stroke and heart failure. LLMs offer a scalable and accessible means to disseminate health information, but their efficacy hinges on the quality and contextual appropriateness of their output. This study highlights a critical disconnect: what medical experts deem 'quality' in health information (e.g., accuracy, bias, consequence explanation) differs substantially from what laypersons prioritize (e.g., structural clarity, visual aids). For practicing clinicians and department heads, these findings underscore the importance of not solely relying on AI-generated content without clinical oversight and customization. While LLMs can provide a foundational draft for patient education, specialists must critically review and adapt such materials to ensure accuracy, address potential biases, and frame information in a way that resonates with patient learning styles and priorities. Overlooking these nuanced differences could lead to suboptimal patient comprehension, reduced adherence, and potentially adverse health outcomes, thereby undermining the very goal of leveraging AI for improved patient engagement.
Clinical & Workflow Takeaways
Clinicians, particularly cardiologists and those involved in patient education, should view AI-generated content as a starting point rather than a final product for patient materials. While LLMs can efficiently produce initial drafts, a thorough review is crucial to ensure clinical accuracy, unbiased presentation, and clear articulation of treatment consequences. When developing patient education resources, it is advisable to involve both clinical experts and patient representatives (or those with similar non-medical backgrounds) in the review process. This multidisciplinary approach can help balance clinical rigor with user-friendliness, addressing both the informational needs identified by medical professionals and the structural preferences highlighted by non-medical individuals. Department heads should consider implementing guidelines for the use of AI in patient education, advocating for human-in-the-loop validation to maintain high standards of patient communication and mitigate potential misunderstandings.
Limitations and What Remains Uncertain
This was a pilot study with a relatively small sample size (n=28), which limits the generalizability of the findings. The evaluation was based on a single prompt and specific versions of the AI models (ChatGPT and Google Gemini), meaning results may not apply to other LLMs, different prompts, or future iterations of these platforms. The assessment relied on self-reported perceptions via a Likert scale, which can be subjective. Furthermore, the study did not evaluate the actual impact of these AI explanations on patient comprehension, retention, or clinical outcomes, focusing solely on perceived quality. Future research needs to involve larger, more diverse cohorts and longitudinal studies to assess real-world efficacy and long-term effects on patient behavior and health status.
Explore the Work Behind Clinical AI
MedXchange Research Fellowship
Publish Clinical AI Research With Us
Our 10-week fellowship mentors physicians and researchers to build foundation models, analyze multimodal health datasets, and co-author peer-reviewed clinical studies.
Related Clinical AI Stories
LLM Performance in PMOS/PCOS Information: DeepSeek-V4 Outperforms GPT-5.5 and Gemini 3.5 in Clinician-Rated Quality and Empathy
A recent study evaluated the information quality, readability, transparency, and perceived empathy of responses from three widely accessible large language models (LLMs)—DeepSeek-V4, GPT-5.5, and Gemini 3.5—to clinician-formulated questions regarding polyendocrine metabolic ovarian syndrome (PMOS), formerly known as polycystic ovary syndrome (PCOS). The study found significant differences in performance among the LLMs, with DeepSeek-V4 consistently receiving higher clinician ratings across multiple quality and empathy metrics. Thirty-three distinct English questions, crafted by clinicians and using established PCOS terminology, were submitted to each LLM. Two independent clinicians then rigorously assessed the 99 archived responses using standardized tools such as the DISCERN instrument for information quality, EQIP for transparency, and JAMA benchmarks. Readability was quantitatively analyzed using six different indices, while perceived empathy was subjectively rated. The methodological approach involved matched comparisons using Friedman tests, paired signed-rank permutation comparisons, and Holm correction to account for multiple comparisons.
Mitigating AI Hallucinations in LLMs: A Scoping Review for Clinical and Educational Integration
Large Language Models (LLMs) are increasingly integrated into medical education and healthcare, yet their utility is significantly hampered by 'AI hallucinations' – outputs that are plausible but factually incorrect, unsupported, or fabricated, including erroneous citations. These inaccuracies pose substantial risks to educational integrity, clinical decision-making, and patient safety. A comprehensive scoping review, adhering to Joanna Briggs Institute and Arksey and O'Malley methodologies, has systematically mapped the evidence on the nature, detection, mitigation, and governance implications of these hallucinations. The review identified 21 relevant sources from a pool of 2,051 records published between January 2018 and July 2026. This selection included five empirical studies evaluating LLM performance and 16 conceptual or review-based articles. The synthesis categorized findings into five major themes: conceptualization of hallucinations, detection and validation methods, mitigation strategies, educational implications, and governance frameworks. Key strategies to reduce hallucination risks include retrieval-augmented generation (RAG), structured prompt engineering, robust output verification, and essential human oversight, emphasizing that these are not just technical issues but multidimensional challenges requiring a comprehensive, multilayered approach for safe implementation.
LLM Chatbot Offers Promising, Yet Flawed, Allergen Avoidance Guidance in Contact Dermatitis
A recent quantitative evaluation study explored the utility of a Large Language Model (LLM) chatbot, specifically GPT-4, in providing allergen-free product recommendations for patients with contact dermatitis. The study assessed the LLM's ability to recommend products free of 30 common allergen groups found in international standard patch tests. While the chatbot demonstrated a high proportion of valid responses, particularly for cosmetics and medications, clinicians must recognize its limitations, especially concerning complex allergens prone to cross-reactivity. Over a period from September 2023 to January 2024, GPT-4 was prompted to suggest five allergen-free products for each allergen group, utilizing three distinct question formats. Dermatologists manually reviewed and categorized 2250 responses for validity, invalidity, cross-reactions, unknown status, and unresponsiveness. The findings indicate a significant potential for LLM tools to assist in allergen avoidance counseling, albeit with a crucial need for human verification and oversight.