Analysis of Critical Incident Reports Using Natural Language Processing.

A Critical Incident Reporting System (CIRS) collects anecdotal reports from employees, which serve as a vital source of information about incidents that could potentially harm patients. OBJECTIVES: To demonstrate how natural language processing (NLP) methods can help in retrieving valuable information from such incident data. METHODS: We analyzed frequently occurring terms and sentiments as well as topics in data from the Swiss National CIRRNET database from 2006 to 2023 using NLP and BERTopic modelling. RESULTS: We grouped the topics into 10 major themes out of which 6 are related to medication. Overall, they reflect the global trends in adverse events in healthcare (surgical errors, venous thromboembolism, falls). Additionally, we identified errors related to blood testing, COVID-19, handling patients with diabetes and pediatrics. 40-50% of the messages are written in a neutral tone, 30-40% in a negative tone. CONCLUSION: The analysis of CIRS messages using text analysis tools helped in getting insights into common sources of critical incidents in Swiss healthcare institutions. In future work, we want to study more closely the relations, for example between sentiment and topics.

[Critical Incident Reporting on medication errors regarding units of measurement and calculations]. / Critical Incident Reporting zu Medikationsfehlern mit Masseinheiten und Berechnungen.

INTRODUCTION: Medication safety is an important hot spot of patient safety. This fact prompted the Swiss Patient Safety Foundation to conduct its annual CIRRNET (Critical Incident Reporting & Reacting NETwork) reporting month on the topic of "Units of measurement of medicines and calculation errors in prescription/dispensing (mg, g, mmol etc.)". METHODS: During the CIRRNET reporting month 2018, the CIRS (Critical Incident Reporting System) managers of the health institutions participating in CIRRNET forwarded all incident reports on the topic to the CIRRNET database. The archive was also searched for usable reports on the same topic. The reports were analysed according to the type of medication involved (type of application and medication class according to ATC codes), the process stage affected (prescription, preparation, administration and documentation) and the error category. RESULTS: Of the 110 reports analysed, 27 reports originated directly from the reporting month, 83 reports were additionally extracted from the archive. 67 reports (60 %) concerned intravenously administered medications and analgesics (opioids and non-opioids, such as NSAIDs) were the most frequently affected/involved medication class with 21 reports (18 %). The process step preparation/calculation was the one where most of the errors analysed occurred. With a total of 48 reports (44 %), most of the errors analysed were classified here. Regarding error categories, it becomes apparent that the categories "Derivation of the amount to be administered (volume/number of single doses) from strength or concentration" with 33 (30 %), "Unit of measurement ml - mg" with 20 (18 %) and "Dose/volume per time" with 14 reports (13 %) were reported most frequently. CONCLUSION: Errors concerning units of measurement or the calculation of medication have been numerously reported by the CIRRNET institutions. Often, these were cases in which, for example, the volume of a drug to be administered was derived incorrectly. Such calculations are demanding, prone to error and represent a system problem. Therefore, good solution strategies need to be implemented in order to sustainably improve medication safety and thus patient safety.