FedECA: federated external control arms for causal inference with time-to-event data in distributed settings

Area of Science:

• * Clinical pharmacology and drug development.
• * Biostatistics and real-world evidence generation.
• * Health data privacy and security.

Background:

• * External control arms are crucial for drug development and regulatory approval.
• * Challenges exist in accessing and pooling real-world or historical clinical trial data due to privacy regulations.
• * Centralized data aggregation is often hindered by data protection requirements.

Purpose of the Study:

• * To develop a federated learning method for inverse probability of treatment weighting (IPTW) for time-to-event outcomes.
• * To enable the use of external control arms without centralizing sensitive patient data.
• * To facilitate the comparison of treatment effects across distributed datasets.

Main Methods:

• * Implementation of federated learning to perform IPTW on decentralized patient cohorts.
• * Application of the method in simulated and real-world settings of increasing complexity.
• * Validation using data from three separate patient cohorts for metastatic pancreatic cancer.

Main Results:

• * Demonstrated the feasibility of federated learning for IPTW on time-to-event data across separate cohorts.
• * Successfully applied the method to compare chemotherapy regimens in metastatic pancreatic cancer patients.
• * Showcased the potential for robust comparative effectiveness research without data pooling.

Conclusions:

• * Federated learning offers a viable solution for utilizing external control arms in drug development.
• * The developed method addresses data privacy concerns and facilitates collaborative research.
• * This approach can accelerate the generation of real-world evidence and support regulatory decision-making.