Artificial Intelligence and Clinical Decision Support

The study explores the potential of artificial intelligence-driven generative language models (GLMs), specifically ChatGPT-3.5 and GPT-4, to adjust the complexity of medical information according to patients’ education levels. The aim is to address low health literacy. By using input templates related to type II diabetes and hypertension, the readability of clinical vignettes was modified for different education levels. This was assessed using the Flesch Reading Ease score and the Flesch-Kincaid grade level.

The findings indicate that GLMs can effectively alter the readability of text, categorizing the outputs into three broad tiers: easy, medium, and difficult, depending on the education level. The study highlights the need for further research to ensure the reliable personalization of medical texts, which would ultimately help improve health literacy.

Publication

• Evaluation of Generative Language Models in Personalizing Medical Information: Instrument Validation Study. Aidin Spina, Saman Andalib, Daniel Flores, Rishi Vermani, Faris F Halaseh, Ariana M Nelson. JMIR AI. 2024 Aug 13:3:e54371. doi: 10.2196/54371.

This study evaluates the effectiveness of GPT-4, a large language model, in simplifying medical literature to improve patient understanding in glaucoma care. GPT-4 was used to transform abstracts from glaucoma journals and patient education materials to a 5th-grade reading level. Additionally, it generated new educational content at six different education levels. Readability was assessed using the Flesch-Kincaid Grade Level and Flesch Reading Ease scores, while content consistency was evaluated through latent semantic analysis. The findings indicate that GPT-4 effectively simplifies medical information about glaucoma, making it more accessible while maintaining content consistency. This supports the model’s potential to enhance patient education across various educational levels.

Publication

• Tailoring glaucoma education using large language models: Addressing health disparities in patient comprehension. Aidin C Spina, Pirooz Fereydouni, Jordan N Tang, Saman Andalib, Bryce G Picton, Austin R Fox. Medicine (Baltimore). 2025 Jan 10;104(2):e41059. doi: 10.1097/MD.0000000000041059.

This project evaluates the ability of large language models (LLMs) like GPT-4, GPT-3.5, Claude 2, and Llama 2 to simplify complex orthopedic patient education materials (PEMs), aiming to make them more accessible for patients with diverse health literacy levels. By applying these models to 48 orthopedic PEMs, the study assessed readability improvements using Flesch-Kincaid metrics, with results showing that GPT-4 most effectively reduced reading complexity to a seventh-grade level. The analysis also explored how initial text characteristics influence transformation success, providing insight into optimizing AI applications in health literacy. This work underscores the potential of LLMs to enhance patient comprehension, a crucial step towards improving health outcomes in orthopedic care.

Publication

• Source Characteristics Influence AI-Enabled Orthopaedic Text Simplification: Recommendations for the Future. Saman Andalib, Sean S Solomon, Bryce G Picton, Aidin C Spina, John A Scolaro, Ariana M Nelson. JB JS Open Access. 2025 Jan 8;10(1):e24.00007. doi: 10.2106/JBJS.OA.24.00007. eCollection 2025 Jan-Mar.

This study used ChatGPT-4 to simplify complex abstracts in neurology and neurosurgery, as well as patient education materials, from a high reading level (12th/13th grade) down to a 5th grade level. The Flesch-Kincaid and Flesch Reading Ease scores showed significant improvements in readability, while Latent Semantic Analysis (LSA) and evaluations by experts indicated that the content was mostly preserved. However, LSA was found to be less reliable for longer texts, which suggests that expert validation is still essential. These findings highlight the potential of AI-driven text simplification to address low health literacy and emphasize the need for improvements in transformation accuracy and validation methods in future research.

Publication

• Assessing AI Simplification of Medical Texts: Readability and Content Fidelity. Bryce Picton, Saman Andalib, Aidin Spina, Brandon Camp, Sean S Solomon, Jason Liang, Patrick M Chen, Jefferson W Chen, Frank P Hsu, Michael Y Oh. Int J Med Inform. 2025 Mar:195:105743. doi: 10.1016/j.ijmedinf.2024.105743. Epub 2024 Dec 1.

This project explores the dissemination of orthopedic patient education materials (PEMs) on social media, specifically examining how text complexity affects engagement metrics like views, likes, and time spent on posts. The study utilized two accounts: one with original content from the American Academy of Orthopaedic Surgeons (AAOS) and the other featuring simplified, AI-processed versions. A total of 31 PEMs focused on topics related to the spine, hand, upper extremity, hip, and knee were analyzed.

 The findings revealed that while the simplified content reached a similar audience size, the more complex posts received significantly more likes and retained viewers for longer periods. Additionally, spine-related content consistently generated the highest levels of engagement, indicating that audiences prefer detailed information in this area of orthopedic care. These insights emphasize the need to strike a balance between making content accessible and maintaining depth in order to improve patient education on social media platforms.

This study examines the use of artificial intelligence (AI) to bridge language gaps for Spanish-speaking orthopedic patients by assessing the accuracy of large language models (LLMs) in translating and simplifying medical texts. Through BLEU analysis, the research found that AI tools, including ChatGPT, achieved translation accuracy comparable to that of Google Translate. Additionally, these tools showed potential for simplifying complex medical information into accessible Spanish. These findings emphasize the importance of AI in enhancing communication with Spanish-speaking patients; however, native language review remains crucial for ensuring accuracy in clinical settings.

This ongoing study collects data from four consecutive cohorts of medical students to evaluate trends in their understanding of and interest in artificial intelligence (AI). By identifying changes in knowledge and enthusiasm, the project aims to refine and enhance medical education strategies related to AI, ensuring that future physicians are well-prepared for technology-driven advancements in healthcare.