Exploring Suitability of Low-Severity Rating Hospital Incident Reports for Machine Learning.

Abstract

Electronic incident reporting is a key quality and a safety process for healthcare organizations that assists in evaluating performance and informing quality improvement initiatives. Although it is mandatory for high-severity incident reports to be investigated, the majority, classified as low severity, are seldom examined due to the large volume of reports, constraints of human cognitive capacity to process such large amounts of data, and the limited resources available in healthcare organizations. The purpose of this study was to investigate low-severity incident reports for suitability of future machine learning to identify actionable interventions for harm prevention. This qualitative descriptive study used a yearlong dataset of low incident severity rating reports to model the incident reporting documentation workflow and explored findings with five nursing and healthcare quality and safety experts. Incident severity reports were reported to have multiple conflicting issues including information duplication, subjective data, too many selection options, and absence of contextual information resulting in a lack of usefulness of information for machine learning. Next steps include analysis of a dataset for machine learning suitability. Recommendations include end-user involvement in system redesign to ensure hospital incident reports are comprised of meaningful data.