Standardized Description of the Feature Extraction Process to Transform Raw Data Into Meaningful Information for...

Area of Science:

• Biomedical Informatics
• Data Science
• Observational Studies

Background:

• Data reuse in research faces challenges due to raw data complexity.
• Information often requires computation from raw data for algorithmic use.

Purpose of the Study:

• Standardize feature extraction steps in retrospective observational studies.
• Propose methods for storing extracted features in a data warehouse schema.

Main Methods:

• Collected and analyzed study cases on automatic and secondary data use for feature extraction.
• Standardized common steps and transformations across study cases.
• Identified suitable tables within the OMOP Common Data Model (CDM) for feature storage.

Main Results:

• Defined two states: 'track' (time-dependent signal) and 'feature' (time-independent information).
• Identified two transformations: 'track definition' and 'track aggregation'.
• Proposed 'TRACK' and 'FEATURE' tables to extend the OMOP CDM for storing extracted variables.

Conclusions:

• A standardized feature extraction process, divided into track definition and aggregation, manages complexity.
• Standardizing tracks requires data expertise but enables complex transformations.
• Enhanced reproducibility of retrospective studies through standardized feature extraction.