Grant: Project Grant

In healthcare, patient safety is often quantified by the occurrence of adverse and unintended events that may befall them. Capturing adverse events is a critical component of understanding the quality of health care and has been widely studied in many countries. These events are used to create patient safety indicators, which are specific safety- related adverse events. Adverse events are largely under-reported in hospital data and health surveillance systems. Thus, despite various efforts to identify hundreds of patient safety indicators, multi-specialty panels recommended only 26 indicators as the 'accepted' set of patient safety indicators. Electronic medical records (EMRs) contain detailed information regarding various aspects of patient care and have been widely implemented across Canada. As such, EMRs have become a valuable resource for the timely detection of various adverse events. However, the implementation of EMRs has not yet been coupled with powerful computational tools that automatically detect adverse events. To bridge this gap, we propose the current study, which aims to create EMR-based algorithms that detect adverse events for defining patient safety indicators. We will use artificial intelligence techniques (i.e. natural language processing) to automate the detection of a comprehensive list of adverse events using linked EMR and discharge abstract data from Alberta. We will then assess the validity of the developed algorithms by comparing them with clinical expert reviews conducted on a dataset of 10,000 EMR charts from Calgary. The developed EMR-based algorithms can be used to routinely detect adverse events for patient safety surveillance and monitoring and facilitate decision-making.