Large Language Models (LLMs have the potential to revolutionize clinical medicine by enhancing healthcare access, diagnosis, surgical planning, and education. However, their utilization requires careful, prompt engineering to mitigate challenges like hallucinations and biases. Proper utilization of LLMs involves understanding foundational concepts such as tokenization, embeddings, and attention mechanisms, alongside strategic prompting techniques to ensure accurate outputs. For innovative healthcare solutions, it is essential to maintain ongoing collaboration between AI technology and medical professionals. Ethical considerations, including data security and bias mitigation, are critical to their application. By leveraging LLMs as supplementary resources in research and education, we can enhance learning and support knowledge-based inquiries, ultimately advancing the quality and accessibility of medical care. Continued research and development are necessary to fully realize the potential of LLMs in transforming healthcare.

Background: Large language models (LLMs) have many clinical applications. However, the comparative performance of different LLMs on orthopedic board style questions remains largely unknown. Methods: Three LLMs, OpenAI’s GPT-4 and GPT-3.5, and Google Bard, were tested on 189 official 2022 Orthopedic In-Training Examination (OITE) questions. Comparative analyses were conducted to assess their performance against orthopedic resident scores and on higher-order, image-associated, and subject category-specific questions. Results: GPT-4 surpassed the passing threshold for the 2022 OITE, performing at the level of PGY-3 to PGY-5 ( p = .149, p = .502, and p = .818, respectively) and outperforming GPT-3.5 and Bard ( p < .001 and p = .001, respectively). While GPT-3.5 and Bard did not meet the passing threshold for the exam, GPT-3.5 performed at the level of PGY-1 to PGY-2 ( p = .368 and p = .019, respectively) and Bard performed at the level of PGY-1 to PGY-3 ( p = .440, .498, and 0.036, respectively). GPT-4 outperformed both Bard and GPT-3.5 on image-associated ( p = .003 and p < .001, respectively) and higher-order questions ( p < .001). Among the 11 subject categories, all models performed similarly regardless of the subject matter. When individual LLM performance on higher-order questions was assessed, no significant differences were found compared to performance on first order questions (GPT-4 p = .139, GPT-3.5 p = .124, Bard p = .319). Finally, when individual model performance was assessed on image-associated questions, only GPT-3.5 performed significantly worse compared to performance on non-image-associated questions ( p = .045). Conclusion: The AI-based LLM GPT-4, exhibits a robust ability to correctly answer a diverse range of OITE questions, exceeding the minimum score for the 2022 OITE, and outperforming predecessor GPT-3.5 and Google Bard.

Background: Patient-education materials (PEMs) are essential to improve health literacy, engagement, and treatment adherence, yet many exceed the recommended readability levels. Therefore, individuals with limited health literacy are at a disadvantage. This study evaluated the readability of spine-related PEMs from the American Academy of Orthopaedic Surgeons (AAOS), the North American Spine Society (NASS), and the American Association of Neurological Surgeons (AANS), and examined the potential of artificial intelligence (AI) in optimizing PEMs for improved patient comprehension. Methods: A total of 146 spine-related PEMs from the AAOS, NASS, and AANS websites were analyzed. Readability was assessed using the Flesch-Kincaid Grade Level (FKGL) and Simple Measure of Gobbledygook (SMOG) Index scores, as well as other metrics, including language complexity and use of the passive voice. ChatGPT-4o was used to revise the PEMs to a sixth-grade reading level, and post-revision readability was assessed. Test-retest reliability was evaluated, and paired t tests were used to compare the readability scores of the original and AI-modified PEMs. Results: The original PEMs had a mean FKGL of 10.2 ± 2.6, which significantly exceeded both the recommended sixth-grade reading level and the average U.S. eighth-grade reading level (p < 0.05). ChatGPT-4o generated articles with a significantly reduced mean FKGL of 6.6 ± 1.3 (p < 0.05). ChatGPT-4o also improved other readability metrics, including the SMOG Index score, language complexity, and use of the passive voice, while maintaining accuracy and adequate detail. Excellent test-retest reliability was observed across all of the metrics (intraclass correlation coefficient [ICC] range, 0.91 to 0.98). Conclusions: Spine-related PEMs from the AAOS, the NASS, and the AANS remain excessively complex, despite minor improvements to readability over the years. ChatGPT-4o demonstrated the potential to enhance PEM readability while maintaining content quality. Future efforts should integrate AI tools with visual aids and user-friendly platforms to create inclusive and comprehensible PEMs to address diverse patient needs and improve health-care delivery.

Background: Sarcopenia increases postoperative complication and mortality rates in elderly patients. Although measurement of the psoas muscle area on computed tomography (CT) scans is traditionally used to diagnose sarcopenia, CT is not routine in orthopaedic practice and causes unnecessary radiation exposure. EOS, a low-dose full-body imaging modality, captures musculoskeletal structures in an upright position, offering an alternative for sarcopenia diagnosis. Methods: Patients ≥18 years of age were included in this retrospective study if they had undergone non-contrast CT spine and EOS imaging between May 2022 and May 2024. Psoas muscle measurements at L3 and L4 were made using non-contrast CT scans, while thigh muscle measurements were obtained with EOS imaging. Inter- and intra-rater reliabilities were assessed using intraclass correlation coefficients (ICCs). Predicted probabilities for L4-psoas sarcopenia were determined through logistic regression, controlling for demographic covariates and validated with an 80% to 20% train-validate split. Sarcopenia cutoffs for anteroposterior (AP) thigh thickness and lateral (LAT) quadriceps thickness were determined with use of the Youden index. Results: Sarcopenia was identified in 23.1% of 134 patients (85 female and 49 male; 121 White, 7 Black, and 6 Hispanic) on the basis of L4-psoas muscle index thresholds. EOS and CT measurements showed excellent ICCs (≥0.90). Multivariable regressions identified AP thigh thickness and LAT quadriceps thickness as significant predictors of psoas area and L4-psoas sarcopenia. The area under the receiver operating characteristic curve for identifying L4-psoas sarcopenia was 0.85 for AP thigh thickness and 0.77 for LAT quadriceps thickness. Cutoffs were 12.47 cm (males) and 10.68 cm (females) for AP thigh thickness, and 3.23 cm (males) and 2.20 cm (females) for LAT quadriceps thickness. In the validation cohort of 27 patients, the AP thigh thickness model showed 0.94 sensitivity and 0.89 specificity, while the LAT quadriceps thickness model showed 0.70 sensitivity and 1.00 specificity. Applying these cutoffs to the entire data set showed that 66.7% of males and 75.0% of females with measurements below both cutoffs had sarcopenia. Conclusions: EOS is a reliable alternative to CT for muscle mass assessment and sarcopenia diagnosis. EOS may be a valuable tool for assessing sarcopenia without a CT scan, as thigh muscle measurements via EOS correlate well with CT-derived psoas measurements. This imaging modality aids in early sarcopenia diagnosis, potentially enhancing preoperative planning and reducing radiation exposure, unnecessary costs, and resource utilization. Level of Evidence: Prognostic Level III. See Instructions for Authors for a complete description of levels of evidence.