Federated Timeline Synthesis: Scalable and Private Methodology For Model Training and Deployment

TL;DR

The paper introduces Federated Timeline Synthesis (FTS), a privacy-preserving, scalable framework for training generative models on distributed healthcare time series data, enabling effective clinical predictions and simulations without sharing sensitive patient information.

Contribution

FTS is a novel federated learning approach that encodes patient histories as tokenized timelines, allowing decentralized training and synthesis of large clinical datasets for healthcare applications.

Findings

Models trained on synthetic data perform comparably to real data models.

FTS provides strong privacy guarantees and scalability across institutions.

Enables diverse healthcare prediction and simulation tasks.

Abstract

We present Federated Timeline Synthesis (FTS), a novel framework for training generative foundation models across distributed timeseries data applied to electronic health records (EHR). At its core, FTS represents patient history as tokenized Patient Health Timelines (PHTs), language-agnostic sequences encoding temporal, categorical, and continuous clinical information. Each institution trains an autoregressive transformer on its local PHTs and transmits only model weights to a central server. The server uses the generators to synthesize a large corpus of trajectories and train a Global Generator (GG), enabling zero-shot inference via Monte Carlo simulation of future PHTs. We evaluate FTS on five clinically meaningful prediction tasks using MIMIC-IV data, showing that models trained on synthetic data generated by GG perform comparably to those trained on real data. FTS offers strong privacy guarantees, scalability across institutions, and extensibility to diverse prediction and simulation tasks especially in healthcare, including counterfactual inference, early warning detection, and synthetic trial design.