Characterization of Drug Use Patterns Using Process Mining and Temporal Abstraction Digital Phenotyping

Abstract

Understanding and identifying executed patterns, activities and processes for patients of different characteristics provides medical experts a deep understanding of which tasks are critical in the provided care, and may help identify ways to improve them. However, extracting these events and data for patients with complex clinical phenotypes is not a trivial task. This paper provides an approach to identifying specific patient cohorts based on complex digital phenotypes as a starting point to apply process mining tools and techniques and identify patterns or process models. Using temporal abstraction-based digital phenotyping and pattern matching, we identified a cohort of patients with sepsis from the MIMIC II database, and then apply process mining techniques to discover medication use patterns. In the case study we present, the use of temporal abstraction digital phenotyping helped us discover a relevant patient cohort, aiding in the extraction of the data required to generate drug use patterns for medications of different types such as vasopressors, vasodilators and systemic antibacterial antibiotics. For sepsis patients, combining the use of temporal abstraction digital phenotyping and process mining tools and techniques, was proven to help extract accurate cohorts of patients for health care process mining.