Online Program

Return to main conference page
Wednesday, September 25
Wed, Sep 25, 9:45 AM - 10:30 AM
Marriott Foyer
Poster Session

Building a Predictive Model to Understand the Likelihood of Audit Findings at the Site Level (300929)

*Wen-Yaw (Alex) Hsieh, Pfizer Inc. 

Keywords: Predictive Model, Machine Learning, Clinical Development, Quality, Risk

Lack of insight to proactively predict, in a data-driven and objective manner, which clinical sites are at risk of experiencing potential audit findings is a problem in clinical development. We develop a predictive model to show the probability of a site could have critical or major audit findings to solve this problem. In this poster, we aim to introduce the probability of audit finding model and specific machine learning methodology used. The model uses study characteristics, site attributes, and measures associated with the site to predict which sites are more likely to have audit findings. By using this predictive model, business not only can understand the associated Good Clinical Practice (GCP) risk factors and use them to identify risk signals from active sites, but also can better allocate the resource from risk-based monitoring perspective.