Getting There: Evidence-Based Decision-Making in Road Trauma Prehospital Transport and Care in Queensland

Abstract

Introduction:Process mining, a branch of data science, aims at deriving an understanding of process behaviors from data collected during executions of the process. In this study, we apply process mining techniques to examine retrieval and transport of road trauma patients in Queensland. Specifically, we use multiple datasets collected from ground and air ambulance, emergency department, and hospital admissions to investigate the various patient pathways and transport modalities from accident to definitive care.Aim:The project aims to answer the question, “Are we providing the right level of care to patients?” We focus on (i) automatically discovering, from historical records, the different care and transport processes, and (ii) identifying and quantifying factors influencing deviance from standard processes, e.g. mechanisms of injury and geospatial (crash and trauma facility) considerations.Methods:We adapted the Cross-Industry Standard Process for Data Mining methodology to Queensland Ambulance Service, Retrieval Services Queensland (aero-medical), and Queensland Health (emergency department and hospital admissions) data. Data linkage and “case” definition emerged as particular challenges. We developed detailed data models, conduct a data quality assessment, and preliminary process mining analyses.Results:Preliminary results only with full results are presented at the conference. A collection of process models, which revealed multiple transport pathways, were automatically discovered from pilot data. Conformance checking showed some variations from expected processing. Systematic analysis of data quality allowed us to distinguish between systemic and occasional quality issues, and anticipate and explain certain observable features in process mining analyses. Results will be validated with domain experts to ensure insights are accurate and actionable.Discussion:Preliminary analysis unearthed challenging data quality issues that impact the use of historical retrieval data for secondary analysis. The automatically discovered process models will facilitate comparison of actual behavior with existing guidelines.