Growth, Methodological Evolution, and Clinical Maturity of Artificial Intelligence Research in Major Medical Journals, 2018–2025: A Longitudinal Bibliometric Study

This is the authors' abstract. We don't add key points for this paper.

Objectives: To quantify changes in the volume, methods, and clinical maturity of artificial intelligence (AI) and machine learning (ML) research in major medical journals from 2018 through 2025.Methods: We conducted a longitudinal bibliometric study of PubMed-indexed records from 99 leading clinical and general medical journals. The search was conducted on July 2026. Records were deduplicated and classified as empirical AI/ML research, evidence synthesis, AI discourse, or unclassified. Empirical studies were categorized by model family and clinical maturity. Journal-year Poisson generalized estimating equations used total substantive publications as an offset and clustered observations by journal.Results: Among 6,891 unique publications, 4,205 (61%) were empirical AI/ML studies. Empirical articles increased from 213 in 2018 to 903 in 2025. The adjusted publication rate rose by 20.5% annually (incidence rate ratio, 1.205; 95% confidence interval, 1.166–1.245; p < 0.001). Blinded validation yielded 92.7% agreement (Cohen's kappa, 0.854). Traditional ML remained dominant, while transformer, large language model, and generative AI studies rose sharply after 2022. Overall, 81.9% of empirical studies were retrospective development studies, and 17.4% reached external validation, prospective evaluation, or real-world deployment.Conclusion: The AI/ML research became a larger component of major-journal output, but clinical translation lagged behind publication growth. Progress should be assessed through external validation, prospective evaluation, real-world effectiveness, equity, and transparent reporting.