Fully automated artificial intelligence-driven targeted literature review on clinical outcomes in neuromyelitis optica spectrum disorder

Evidence generation is a multistep process that remains labor-intensive and time-consuming. Although artificial intelligence (AI)-enabled platforms support literature reviews, their complete automation is scarce. We utilized an online AI suite, SymproAI, to perform a fully automated targeted literature review (TLR) on treatment outcomes in patients with aquaporin-4 (AQP4)-positive neuromyelitis optica spectrum disorder (NMOSD) and compared its performance with the manual process. This TLR summarizes the efficacy and safety of Food and Drug Administration-approved biologics or conventional immunosuppressants in patients with AQP4-positive NMOSD using SymproAI. Performance was validated using sensitivity, specificity, predictive values, overall accuracy, and overall speed of performance. SymproAI searches demonstrated 98.2% agreement with the manual retrieval (1,721 vs. 1,691 citations). For title/abstract and full-text screening, SymproAI achieved 94.4% and 80.5% accuracy, respectively. SymproAI generated TLR-identified 44 publications identical to those provided by the manual review. Overall, the TLR timeframe was estimated at ~120 min using SymproAI, compared with 96 work hours required for the manual process. The generative feature of SymproAI demonstrated good agreement with human findings and data interpretation. Both processes agreed on eculizumab with the strongest relapse-prevention efficacy (94.2% risk reduction), followed by satralizumab (74–79%) and inebilizumab (73%). Across conventional therapies, rituximab demonstrated superior relapse prevention compared with these oral immunosuppressants (mycophenolate mofetil and azathioprine), reducing relative relapse risk by ~60% ( p = 0.046), highlighting its greater efficacy. SymproAI enables rapid, accurate, and reproducible automation of TLRs without compromising study selection or clinical conclusions, supporting accelerated and methodologically rigorous evidence synthesis in time-sensitive clinical settings.