Scoping review finds code sharing in 12.2% of clinical prediction model studies

A scoping review published in Nature Medicine assessed analytical code sharing in clinical prediction model research to inform development of TRIPOD-Code, a reporting guideline extension. Researchers used a large-language-model-assisted pipeline to screen articles citing TRIPOD or TRIPOD+AI, extract repository links and assess repositories against 14 predefined reproducibility-related features. The review included 3,967 eligible articles identified through searches current to 11 August 2025.

Overall, 482 articles (12.2%) shared code, including 75 through appendices or supplementary material. Sharing rose from 6.3% in 2015 (1/16) to 14.5% in 2024 (119/820) and 15.8% in 2025 (74/468). Articles citing TRIPOD+AI shared code more frequently than those citing TRIPOD (29.2% versus 11.4%); the association persisted after adjustment for publication year (adjusted odds ratio, 2.61; 95% confidence interval, 1.79–3.82). Sharing also varied by journal and country.

Repository assessment found substantial heterogeneity in reproducibility features. The findings support clearer expectations for documentation, dependency specification and executable structure, rather than code availability alone. For clinicians assessing prediction models, access to analytical code can facilitate independent methodological scrutiny but does not itself establish clinical validity or generalizability. The sample was restricted to articles citing the TRIPOD statements, and 1,407 articles were not retrievable through the PMC Open Access API, limiting representativeness. The association with TRIPOD+AI citation does not establish that the guideline caused greater sharing.