KETOS

KETOS provides a reproducible framework for developing, training, and deploying patient-level prediction and decision support models in clinical settings.

Key Features:

• Docker Virtualization: Leverages Docker to create isolated, reproducible data analysis and development environments.
• Jupyter Notebook Integration: Hosts Jupyter Notebooks within those environments for statistical analysis and model development.
• Modular Architecture: Employs a modular architecture that supports integration with web services for extensibility.
• FHIR Standard Compliance: Interfaces with HL7 Fast Healthcare Interoperability Resources (FHIR) to access and exchange patient data.
• OMOP-CDM Database Utilization: Uses the OMOP Common Data Model (OMOP-CDM) for standardized input data and vocabularies.
• End-to-End Research Lifecycle Support: Supports creation of analysis environments through to deployment of models in hospital settings.
• Privacy-Preserving Data Analysis: Integrates with the open-source DataSHIELD architecture to enable distributed analysis and model training across institutions while preserving patient privacy.

Scientific Applications:

• Hemoglobin Reference Intervals: Evaluated by establishing an analysis application for hemoglobin reference intervals at the University Hospital Erlangen.
• Colorectal Cancer Prediction Models: Used to load a colorectal cancer dataset into an OMOP database and develop machine learning models predicting patient outcomes, with models exposed via web services.

Methodology:

Uses Docker virtualization, the OMOP Common Data Model (OMOP-CDM), HL7 FHIR data exchange, and integration with the open-source DataSHIELD architecture for distributed, privacy-preserving analysis and model training.