AmsterdamUMCdb

AmsterdamUMCdb provides de-identified health data from tens of thousands of intensive care unit admissions for machine learning research and scientific inquiry in critical care medicine.

Key Features:

• Comprehensive Clinical Data: The database contains approximately 1 billion clinical data points encompassing demographics, vital signs, laboratory tests, and medications.
• Large Patient Cohort: Data spans 23,106 admissions of 20,109 patients admitted between 2003 and 2016.
• Risk-Based Deidentification: A risk-based deidentification strategy maintains data utility while preserving patient privacy.
• Regulatory Compliance: The database complies with both U.S. Health Insurance Portability and Accountability Act and European General Data Protection Regulation privacy regulations.
• Multidisciplinary Governance: Contractual and governance processes, communication strategies, and audits by patient organizations, supporting hospitals, and experts on ethics and privacy were implemented.

Scientific Applications:

• Machine Learning in Critical Care: The database supports machine learning approaches to improve outcomes for critically ill patients through analysis of vast ICU-generated datasets.
• Privacy-Preserving Data Sharing: AmsterdamUMCdb serves as a model for responsible ICU patient data sharing worldwide under the Society of Critical Care Medicine and European Society of Intensive Care Medicine Joint Data Science Collaboration.
• Critical Care Outcomes Research: The database enables scientific inquiry to improve critical care outcomes through analysis of large-scale combined ICU datasets.

Methodology:

Data were collected from ICU patients admitted between 2003 and 2016 at a university hospital ICU. A risk-based deidentification strategy was applied to maintain data utility while preserving privacy. Contractual and governance processes, and a communication strategy were implemented. Patient organizations, supporting hospitals, and experts on ethics and privacy audited these processes and the database.