Online Program

Collaborating to Develop Prediction Models: Advice from a Recent Project (309967)

Keywords: collaboration, prediction modeling

Developing prediction models is a common task for statisticians and data scientists, and the process usually involves collaborating with content area experts. But, working together can be a challenge because of differing interdisciplinary perspectives, especially in a new collaboration. In this presentation, I will give practical advice for collaborations that involve developing prediction models based on a recent project. Our aim was to develop prediction models for fall-risk in older adults, and my collaborators included medical doctors and epidemiologists. Advice arising from this project includes: 1) clearly define outcomes and predictors, 2) figure out each other’s jargon, 3) discuss modeling goals (interpretation versus performance) and openness to complex models (e.g., machine learning), 4) determine a strategy for model validation (split sampling versus independent data), and 5) ensure your code is quickly, easily and consistently able to be re-run as the models are refined. Prediction modeling is an important and fun role for statisticians and data scientists, and effectively collaborating with content area experts is essential for developing useful and impactful models.