Issa Dahabreh, MD, ScD
Awards
Advance-CTR Pilot Projects Program (2020)
"Transportability of Risk Prediction Models"
CO-PI: Jon Steingrimsson, PhD (Contact PI)
When the source data used to build a prediction model differs on important characteristics from the target population the model is intended to be used in, models built using the source data are not directly applicable to the target population. Such differences between the study and target population are common (e.g., due to data coming from different geographic locations or healthcare systems). Motivated by applications in breast cancer, this proposal develops and evaluates methods for tailoring risk prediction models and associated measures of model performance to a target population the model is intended for.
With medical data, the outcome of interest is often time to some event, such as overall or recurrence free survival. Such outcomes are often only partially observed due to individuals dropping out of the study or not experiencing the outcome of interest before the end of the study. This proposal will develop methods that can tailor risk prediction models and associated measures of model performance to a target population when some outcomes are only partially observed.