AI helps researchers investigate how thousands of medical studies share code

MIT AI News
Read full story· 5 min
Researchers from MIT and Dana-Farber Cancer Institute used large language models to analyze nearly 4,000 medical studies, finding that code sharing is rare and often lacks sufficient documentation for reuse. Their findings support the development of TRIPOD-Code, a guideline to improve code sharing in clinical prediction research.

What happened

Researchers from MIT and Dana-Farber Cancer Institute used large language models to review nearly 4,000 clinical prediction studies available in PubMed Central. Their goal was to assess how often and how well these studies shared their analytical code. The team found that code sharing was infrequent and, when present, often lacked the documentation needed for others to reuse it. Their findings are being used to inform the development of TRIPOD-Code, a new guideline for code sharing in clinical prediction research.

Key details

  • Nearly 4,000 studies were analyzed using large language models.
  • The review included studies citing TRIPOD or TRIPOD+AI guidelines.
  • Code repositories were checked for documentation, software requirements, and reuse terms.
  • Code sharing rates varied widely between journals, from 0% to over 70%.

Why it matters

Researchers, journal editors, and clinical prediction model developers are directly affected, as they rely on access to well-documented code for transparency and reproducibility. The lack of consistent code sharing and documentation means that other scientists may struggle to verify or build on published work. The introduction of the TRIPOD-Code guideline will give researchers and journals clearer standards for what to share, potentially improving the reliability and utility of future studies.


More in Healthcare