Grant Details
| Grant Number: |
1R01CA316177-01 Interpret this number |
| Primary Investigator: |
Ning, Jing |
| Organization: |
University Of Tx Md Anderson Can Ctr |
| Project Title: |
Advancing Transfer Learning Using Surrogates, Family History, and Prior Models. |
| Fiscal Year: |
2026 |
Abstract
Project Summary:
Accurate clinical decision-making increasingly depends on the integration of diverse data
resources, including established risk prediction models, surrogate outcomes, and family
medical history. Transfer learning offers a powerful tool to adaptively leverage these
heterogeneous sources, improving model performance when target data are limited or differ
substantially from prior studies. However, existing transfer learning approaches often rely on
restrictive assumptions, such as identical risk factors and measurement scales across studies
or require complex joint modeling that is impractical in many real-world settings. In addition,
family history data pose challenges due to recall bias, missingness, and sparsity, limiting their
predictive utility.
The objective of this proposal is to address these challenges by developing rigorous, flexible,
and robust methods for knowledge transfer, tailored to the unique characteristics of each
setting. Specifically, we will: 1) Develop rank-based transfer learning approaches that leverage
relative risk rankings from existing models, enabling integration of new variables such as novel
biomarkers while addressing population heterogeneity and preventing negative transfer; 2)
Create systematic frameworks for incorporating surrogate data to improve the efficiency of
primary endpoint analyses without requiring burdensome joint modeling, while adjusting for
covariate and label shifts; and 3) Propose empirical likelihood approaches with novel bias-
correction methods to integrate family health history data, explicitly accounting for recall bias
to enhance risk assessment and prediction accuracy.
These methodological innovations will be applied to motivating studies to demonstrate gains in
statistical efficiency and clinical utility. To ensure broad dissemination and maximize impact,
we will provide user-friendly, open-access software tools with comprehensive documentation.
By bridging methodological innovation with real-world application, this project aims to
significantly advance the use of transfer learning in biomedical research and support more
precise, evidence-based clinical decision-making.
Publications
None