ODVICE: An Ontology-Driven Visual Analytic Tool for Interactive Cohort Extraction
This addresses the challenge of robust EHR analysis for researchers studying uncommon diseases, but it is an incremental improvement as it applies data augmentation specifically to medical ontologies.
The paper tackles the problem of limited records in EHR cohorts for uncommon diseases by presenting ODVICE, a data augmentation framework that uses a medical concept ontology and a novel Monte-Carlo graph spanning algorithm, resulting in ~30% improvement in AUC over non-augmented datasets and other strategies.
Increased availability of electronic health records (EHR) has enabled researchers to study various medical questions. Cohort selection for the hypothesis under investigation is one of the main consideration for EHR analysis. For uncommon diseases, cohorts extracted from EHRs contain very limited number of records - hampering the robustness of any analysis. Data augmentation methods have been successfully applied in other domains to address this issue mainly using simulated records. In this paper, we present ODVICE, a data augmentation framework that leverages the medical concept ontology to systematically augment records using a novel ontologically guided Monte-Carlo graph spanning algorithm. The tool allows end users to specify a small set of interactive controls to control the augmentation process. We analyze the importance of ODVICE by conducting studies on MIMIC-III dataset for two learning tasks. Our results demonstrate the predictive performance of ODVICE augmented cohorts, showing ~30% improvement in area under the curve (AUC) over the non-augmented dataset and other data augmentation strategies.