TL02: Proprietary Algorithms in Clinical Risk Prediction: Responsibilities of Consulting Biostatisticians

Thomas M. Menino Convention & Exhibition Center 

The boom in accessibility of artificial intelligence and machine learning software over the last decade has generated many new challenges for collaborative biostatisticians. One such challenge is the incorporation of proprietary algorithms (often industry-sponsored) into clinical risk prediction analyses. While a proprietary approach to data analysis methods is in direct conflict with the transparency and reproducibility principles that statisticians often advocate for, modern biostatisticians consulting in clinical settings must accept that proprietary models are likely to only increase in popularity and be prepared to assess a project proposal with a proprietary component. In this roundtable, we will discuss pros and cons of proprietary methods, current academic and regulatory guidance, recommended reporting guidelines, and potential quality control actions. Participants will leave with resources outlining key considerations for determining scope of biostatistician responsibility on such collaborative projects.

Clinical Risk Prediction

Statistical Consulting

Regression Modeling

Machine Learning

Artificial Intelligence

Transparent Reporting 

Section on Statistical Consulting