Background: Spanish-speaking patients face persistent barriers to health information due to limited availability and poor readability of online Spanish-language educational resources. Generative artificial intelligence (AI) presents a promising solution to address this gap. This study modeled a practical, scalable AI intervention to generate accessible Spanish-language orthopedic educational materials that institutions can readily adopt to address language disparities.

Methods: We conducted a cross-sectional analysis of the 203 U.S. academic orthopedic institutions listed on the 2024 ERAS directory to assess the availability of Spanish-language patient resources on each institution’s website. We then modeled an AI intervention using 30 articles from Institution’s orthopedic patient library using ChatGPT-4o, as currently no Spanish resources are available, with the goal of improving readability and translation quality. Prompt development was performed using proven prompt engineering strategies including context/persona building, few-shot prompting, structured chain prompting, and iterative testing until satisfactory outputs were achieved. Readability was assessed using validated readability tests in both languages. A modified DISCERN tool scored by two Spanish-speaking orthopedic surgeons assessed content and translation quality. Scores before and after ChatGPT intervention were compared with paired t-tests while correlation between changes in reading grade level and quality was assessed with Pearson’s correlation (p<0.05).

Results: Of the 203 institutions included in our search, 142 (70%) provided patient educational resources in English on their websites. Of these 142 institutions, only 17 (12%) provided Spanish-language resources. The cumulative average reading grade level of our model’s original articles was 10.7±2.3 in English and 10.1±1.7 in Spanish (“original” Spanish articles were approximated with Google Translate). ChatGPT significantly reduced the reading level in both languages, achieving an average of 6.9±1.5 in English (p<0.01) and 6.8±1.2 in Spanish (p<0.01). ChatGPT-generated Spanish articles demonstrated improvement in DISCERN content quality scores according to Grader 1 (p<0.05), and noninferiority according to Grader 2 (p=0.37). The quality of translation was scored significantly higher by both graders for ChatGPT content (p<0.05). No significant negative correlation was observed between improvements in readability and content quality scores for either grader (Grader 1: r=0.34, p=0.23; Grader 2: r=0.12, p=0.67).

Conclusion: Our model demonstrates the effectiveness of AI for enhancing the accessibility, readability, and quality of online Spanish-language education materials, representing a cost-effective and practical approach institutions may implement in real-time to address language barriers and promote equity in patient education.