E-PIX

E-PIX performs identity management and probabilistic record linkage to detect duplicate patient registrations and link records across large-scale biomedical research datasets.

Key Features:

• Probabilistic Record Linkage: Implements the Fellegi-Sunter algorithm to calculate match probabilities and link patient records.
• Error Tolerance: Uses the Levenshtein distance to account for typographical errors in data entry during matching.
• Scalability: Demonstrated capability to manage datasets exceeding 30 million patients on existing hardware configurations.
• Performance Benchmarking: Evaluated by simulating registration of at least 20 million patients, achieving registration rates of over 16,000 patients per day.

Scientific Applications:

• Multi-site identity management: Prevents duplicate registrations and enables consistent patient identification when recruiting from multiple research sites.
• Large-scale data aggregation: Supports aggregation of medical data across locations for centralized biobanks such as the Central Biomaterial Bank at Charité - Universitätsmedizin Berlin.
• Dataset integrity assurance: Detects and minimizes duplicate entries to maintain data integrity in multi-million patient studies.

Methodology:

E-PIX employs probabilistic algorithms (Fellegi-Sunter) combined with error-tolerant techniques (Levenshtein distance) and uses simulations of the registration process with modified real patient data to evaluate runtime, memory usage, and processor utilization.