Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MyDigiTwin: A Privacy-Preserving Framework for Personalized Cardiovascular Risk Prediction and Scenario Exploration

Héctor Cadavid, Hyunho Mo, Bauke Arends, Katarzyna Dziopa, Esther E. Bron, Daniel Bos, Sonja Georgievska, Pim van der Harst

TL;DR

MyDigiTwin tackles the challenge of enabling proactive primary prevention for cardiovascular disease under strict privacy constraints by combining health digital twins with personal health environments in a privacy-preserving federated learning framework. The approach hinges on a FHIR-based data harmonization layer and the Personal Health Train/Vantage6 infrastructure to train models across distributed cohorts without sharing raw data, while enabling patients to explore personalized scenarios. The authors demonstrate end-to-end feasibility through a proof-of-concept using Lifelines and WHAS data, showing successful data harmonization to a ZIB-FHIR profile and improved discrimination (e.g., $C$-statistic$=$0.764 to 0.788 on Lifelines) with FedAvg. Collectively, the work presents a scalable pathway for privacy-preserving, personalized cardiovascular risk prediction and scenario exploration in real-world healthcare, with future work aimed at real deployment, broader data modalities, and enhanced transparency.

Abstract

Cardiovascular disease (CVD) remains a leading cause of death, and primary prevention through personalized interventions is crucial. This paper introduces MyDigiTwin, a framework that integrates health digital twins with personal health environments to empower patients in exploring personalized health scenarios while ensuring data privacy. MyDigiTwin uses federated learning to train predictive models across distributed datasets without transferring raw data, and a novel data harmonization framework addresses semantic and format inconsistencies in health data. A proof-of-concept demonstrates the feasibility of harmonizing and using cohort data to train privacy-preserving CVD prediction models. This framework offers a scalable solution for proactive, personalized cardiovascular care and sets the stage for future applications in real-world healthcare settings.

MyDigiTwin: A Privacy-Preserving Framework for Personalized Cardiovascular Risk Prediction and Scenario Exploration

TL;DR

MyDigiTwin tackles the challenge of enabling proactive primary prevention for cardiovascular disease under strict privacy constraints by combining health digital twins with personal health environments in a privacy-preserving federated learning framework. The approach hinges on a FHIR-based data harmonization layer and the Personal Health Train/Vantage6 infrastructure to train models across distributed cohorts without sharing raw data, while enabling patients to explore personalized scenarios. The authors demonstrate end-to-end feasibility through a proof-of-concept using Lifelines and WHAS data, showing successful data harmonization to a ZIB-FHIR profile and improved discrimination (e.g., -statistic0.764 to 0.788 on Lifelines) with FedAvg. Collectively, the work presents a scalable pathway for privacy-preserving, personalized cardiovascular risk prediction and scenario exploration in real-world healthcare, with future work aimed at real deployment, broader data modalities, and enhanced transparency.

Abstract

Cardiovascular disease (CVD) remains a leading cause of death, and primary prevention through personalized interventions is crucial. This paper introduces MyDigiTwin, a framework that integrates health digital twins with personal health environments to empower patients in exploring personalized health scenarios while ensuring data privacy. MyDigiTwin uses federated learning to train predictive models across distributed datasets without transferring raw data, and a novel data harmonization framework addresses semantic and format inconsistencies in health data. A proof-of-concept demonstrates the feasibility of harmonizing and using cohort data to train privacy-preserving CVD prediction models. This framework offers a scalable solution for proactive, personalized cardiovascular care and sets the stage for future applications in real-world healthcare settings.
Paper Structure (18 sections, 10 figures, 4 tables, 2 algorithms)

This paper contains 18 sections, 10 figures, 4 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Personal Health Environments and the Dutch healthcare system
  3. MyDigiTwin system architecture
  4. Harmonization of federated cohort studies
  5. Proof of concept
  6. Research infrastructure setup
  7. Model development
  8. Federated learning algorithm
  9. Prediction models
  10. Datasets
  11. Worcester Heart Attack Study (WHAS) Dataset
  12. Lifelines
  13. Validation
  14. Discussion
  15. Health digital twins: adoption challenges and MyDigiTwin
  16. ...and 3 more sections

Figures (10)

  • Figure 1:
  • Figure 2: Overview of MyDigiTwin's end-user early detection and simulation services. (1) The end-user provides scenarios of health indicators for his/her DigiTwin to be simulated. (2) The personal health environment (PGO) gets access to the end-user indicators available from (MedMij-compliant) health care providers. (3) The prediction model that fits better the end-user is selected from the set of pre-trained prediction models. This model is fed with both modified and actual indicators based on the model's metadata -which defines its expected input-. (4) The risk estimation, calculated by the prediction model for the given scenario, is sent back to the end-user.
  • Figure 3: MyDigiTwin framework high-level system architecture. The organizations involved on research environment (bottom left) agreed on which variables the vantage6 node will have access to through a set of FHIRPath expressions. A researcher request the execution of an algorithm for performing a federated analysis or model training (bottom right), also considering these "canonical" FHIRPath expressions. On the end-user environment (top), when the model is required for a given user prediction or simulation, the same FHIRPath expressions, that are part of the model metadata, are used to get (consistently with the training process) the indicators that will be used as input.
  • Figure 4: Left: Lifelines datafile format, where the assessments of each data collection phase (a wave) are available on a separate file. Center: Intermediate representation format used by the general-purpose data-harmonization tools. Right: Generated FHIR resources. Lifelines2CDF: tool for transforming Lifelines data files into the CDF format. CDF2FHIR: tool for transforming CDF records into FHIR resources, given a set of pairing rules.
  • Figure 5: Pairing rules development workflow. Top: the git-based collaborative environment for developing pairing rules following a test-driven approach. Bottom: the pairing rules, once defined, are used on a batch transformation process --creating the target FHIR dataset-- running where the data resides.
  • ...and 5 more figures