Leveraging artificial intelligence in the peer review of neurosurgical research articles.

Abstract

The traditional peer review process is time-consuming and can delay the dissemination of critical research. This study evaluates the effectiveness of artificial intelligence (AI) in predicting the acceptance or rejection of neurosurgical manuscripts, offering a possible solution to optimize the process. Neurosurgical preprints from Preprint.org and medRxiv.org were analyzed. Published preprints were compared to those presumed not accepted after remaining on preprint servers for over 12 months. Each article was uploaded to ChatGPT 4o, Gemini, and Copilot with the prompt: "Based on the literature up to the date this article was posted, will it be accepted or rejected for publication following peer review? Please provide a yes or no answer." AI predictive accuracy and journal metrics were assessed between preprints that were accepted or presumed to be not accepted. A total of 51 preprints (31 skull base, 20 spine) were included, with 28 published and 23 presumed not accepted. The average impact factor and cite score for accepted preprints were 4.36 ± 2.07 and 6.38 ± 3.67 for skull base and 3.48 ± 1.08 and 4.83 ± 1.37 for spine topics. Across all AI models, there were no significant differences in journal metrics between preprints predicted to be accepted or not accepted (p > 0.05). Overall, AI models had significantly low performance, with accuracy ranging from 40 to 66.67% (p < 0.001). Current AI models exhibit moderate accuracy in predicting peer review outcomes. Future AI models, developed in collaboration with journals and with authors' consent, could access a more balanced dataset, enhancing accuracy and streamlining the peer review process.