Abdul Kerim BUĞRA
Journal of Health Sciences and Medicine - 2026;9(3):721-726
Aims: This study aimed to systematically evaluate the clinical accuracy and patient accessibility of ChatGPT-4-generated cardiovascular surgery patient education content across Turkish and English languages, examining language-dependent performance patterns and their implications for clinical practice. Methods: Twelve cardiovascular surgery specialists with minimum 5 years clinical experience conducted blinded evaluation of 52 ChatGPT-4 responses (26 Turkish, 26 English), generating 624 total expert assessments, across six cardiovascular surgery patient education domains using validated 7-point Likert scales. Simultaneously, computational accessibility analysis was performed using language-specific readability indices. Statistical analysis included effect size calculations and correlation analysis. Results: ChatGPT-4 demonstrated high clinical accuracy across both languages (mean satisfaction: 6.40/7.0, 91.4%), with 88.7% of responses receiving high expert ratings. However, profound language-dependent accessibility disparities emerged; Turkish responses achieved near-universal patient accessibility (mean score: 77.7+/-4.2), while English responses required university-level education (mean score: 30.2+/-6.1), creating a 47.5-point accessibility gap (Cohen's d=9.07, p<0.001). Turkish showed balanced clinical accuracy-patient accessibility optimization, while English exhibited systematic optimization prioritizing expert preferences over patient accessibility. Conclusion: Artificial Intelligence (AI)-generated cardiovascular surgery patient education shows equivalent clinical accuracy but profound accessibility disparities across languages. Turkish-speaking patients may benefit from superior accessibility in AI-generated health information, with potential impact on preoperative education quality, informed consent processes, and postoperative patient engagement.
