A federated data analysis framework for surrogate endpoint evaluation
Principal Investigator
Name
Edmund Njeru Njagi
Degrees
PhD
Institution
The University of Warwick
Position Title
Associate Professor in Quantitative Analysis of Breast Cancer Screening
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-1527
Initial CDAS Request Approval
Aug 10, 2026
Title
A federated data analysis framework for surrogate endpoint evaluation
Summary
The need for faster access to innovative healthcare technologies has seen regulatory and health technology assessment agencies increasingly make decisions on the use of these technologies based on evidence from surrogate endpoints. Statistical methodology for evaluating surrogates before they are deployed in clinical research relies on analysis of data from multiple trials, requiring access to data from many different organisations. The current data sharing environment is however prohibitive, as sharing of data by many organisations continues to be impeded due to concerns including data security, ownership and control, and patient privacy. Federated networks, which enable organisations to analyse data from other organisations in the network without any data being shared, transferred, or pooled, are an innovative solution. The analyst submits the analysis task to each data server in the network, which is executed locally under the respective organisation’s data security mechanisms without any data physically leaving the source. Results from the different sources are sent back to the analyst to combine. While this facilitates access, it poses complex statistical challenges because principled methods for deriving valid combined inferences need to be developed. This project aims to develop methodology for evaluating surrogates using federated data approaches. The NLST data are requested to study the practical viability of the developed methodology in real-world application. Our methodology, whose application spans a vast array of disease areas and trial types, will advance surrogate evaluation by tackling the unresolved and critical problem of data access barriers through a cutting-edge methodology. This will optimise surrogate endpoint evaluation, providing robust evidence that will ultimately optimise the huge investments put in trials and guide policy-makers make sound decisions to benefit population at earlier time points.
Aims
Our aims are to:
1. Derive estimators for joint models for different types of endpoints in federated data analysis and use simulation studies to evaluate their performance. We will consider the following combinations of endpoints: time-to-event clinical endpoint and time-to-event potential surrogate endpoint, time-to-event clinical endpoint and binary potential surrogate endpoint, and time-to-event clinical endpoint and categorical (i.e. multicategory) potential surrogate endpoint.
2. Use the joint model estimators developed in (1) to derive surrogate endpoint evaluation measures for the different combinations of endpoints, under federated data analysis, and use simulation studies to investigate their performance.
3. Implement and disseminate the derived methodology in open-source R software.
4. Study the practical viability of the developed methodology in real-world application, by analysing the NLST data.
Collaborators
Edmund Njeru Njagi The University of Warwick
Sian Taylor-Phillips The University of Warwick
Anna Mae Scott The University of Warwick
Geert Molenberghs Katholieke Universiteit Leuven
Ariel Alonso Abad Katholieke Universiteit Leuven