How to Remove Bias from Hospital Data: Automatic Preprocessing of Patient Pathways for Data Analysis (Preprint)

Abstract

Optimisation of patient care pathways is crucial for hospital managers in a context of a scarcity of medical resources. Assuming unlimited capacities, the pathway of a patient would only be governed by pure medical logic, to meet at best patient’s needs. However, logistical limitations (e.g., resources such as inpatient beds) are often associated with delayed treatments and may ultimately affect patient pathways. This is especially true for unscheduled patients: when a patient at the emergency department needs to be admitted to another medical unit without disturbing the flow of planned hospitalisations.

In this study, we propose a new framework to automatically detect activities in patient pathways which may be unrelated to patients’ needs and rather induced by logistical limitations.

The scientific contribution lies in a method that turns a database of history pathways with bias into two databases: (i) a labelled pathways database where each activity is labelled as relevant (related to patient’s need) or irrelevant (induced by logistical limitations) and (ii) a corrected pathways database where each activity corresponds to the activity that would occur assuming unlimited resources. The labelling algorithm is assessed through medical expertise. Two case studies quantify the impact of our preprocessing method of healthcare data by using respectively process mining and discrete event simulation.

Focusing on unscheduled patient pathways, we collected data covering 12 months of activity at the Groupe Hospitalier Bretagne Sud in France. Our algorithm has an error of 13% and has demonstrated its usefulness to preprocess traces and obtain a clean database. The two case studies show the importance of our preprocessing step before any analysis.

Patient pathways data reflect the actual activity of hospitals, governed by medical requirement and logistical limitation. Before any use of these data, these limitations should be identified and corrected. We anticipate the generalisation of our approach to obtain unbiased analyses of patient pathways for other hospitals.

The study was approved by the French Data Protection Authority (CNIL) under the number 922243.