Enhancing Data Reuse through Feature Extraction: Proposal of Concepts, Methods and Data Model (Preprint)

Abstract

Despite the many opportunities data reuse offers, its implementation presents many difficulties and raw data cannot be reused directly. Information are not always directly available in the source database, and have to be computed afterwards with raw data in defining an algorithm.

The purpose of this article is to present the generalization of feature extraction concepts and methods when conducting retrospective observational studies. We also suggest a complementary table to store the features in the Observational Medical Outcomes Partnership (OMOP) common data model (CDM).

This study involved three main steps : the collection of relevant study cases related to feature extraction and based on the automatic and secondary use of data; the definition of a higher-level abstraction, with the concepts, their characteristics, and the methods which were common to the study cases; the proposal of a table to store the features in the OMOP CDM.

We interviewed 10 researchers from 3 French university hospitals and a national institution who were involved in 8 retrospective and observational studies. Based on these studies, two concepts – the track and the feature, and two operations – the track definition and the feature formalization, have emerged. The track is a time-dependent signal or period of interest, defined by a statistical unit, a type of track, a value or a list of values. The feature is a time-independent high-level information with dimensionality identical to the statistical unit of the study, defined by a label and a value. Time dimension has become implicit in the value or name of the variable. We proposed two tables to store variables resulting from feature extraction and extend the OMOP CDM.

We propose a definition of the feature extraction process, in order to transform heterogeneous, multidimensional and time-dependent raw data into valuable information when conducting observational retrospective studies. The process combined two steps, the track definition and the feature formalization. By dividing the feature extraction into these two steps, the difficulty is managed during the track definition. The standardization of the tracks requires a great expertise of the data but allows to apply an infinite number of complex transformations. On the contrary, the feature formalization becomes a very simple operation with a finite number of possibilities.