gms | German Medical Science

Introduction: On intensive care units(ICUs) the problem of alarming devices is well-known. The alarms are often not clinically relevant with rates as low as 20%-1% [1] and with up to 350 alarms per day and patient occurring [2]. As a result many ICUs are affected by “Alarm Fatigue” [3], characterized by inadequate responses to alarms, seriously impacting patient safety. In the AlarmRedux-project (www.alarmredux.de) members from different fields, e.g. medical informatics, clinician and medical device manufacturers work together to reduce the alarm load on intensive care units.

For improving the alarm management on ICUs there are two main problems. On the one hand there is currently a lack of measurements for the alarm management on ICUs [4] and on the other hand there is the demand for a data driven way to handle alarm fatigue [2].

In this study the alarm data from two ICUs was gathered and transferred into a data warehouse. The data warehouse was connected to a data analysis platform enabling clinicians to discover patterns in the alarm data for their specific ICU in a self-service manner.

Methods: The alarm data from two surgical ICUs from different sites was included. The data was gathered for 24 hours over 7 days on both ICUs. A data interface of the patient data monitoring system was used. The interface provided data in a HL7 format. For privacy, only the message header (MSH) and the observation result (OBX) segment was processed. The OBX segment was used to gather the alarm related information. As a second data source and for additional information an audit log from the patient monitoring central was used. Both data sources were combined as a data basis for the data warehouse. The data warehouse was connected to a self-service business intelligence platform. Specific analytic views for alarm data analysis were provided for clinicians.

Results: The HL7 data interface of the patient data monitoring system and the data from the clinical audit log were the data basis for the developed alarm data warehouse. The combination of both data sources resulted in information about the alarm data, alarm pausing, alarm setting changes and physiological values in the moment of an occurring alarm condition. For providing a platform to enable explorative data analysis, a data warehouse was set up and the HL7 messages were transformed into a multidimensional model. When extracting the operational data and transforming it we needed to filter and normalize the data. In the filter step the data was harmonized to correct syntactically or semantically varying data. In this step we also added some metadata like “language” and “number of beds per unit.

Conclusion and Outlook: The alarm data from two ICUs was transferred into a data warehouse and the data is ready for an explorative data analysis. We will test the implemented platform in a workshop with clinicians from the AlarmRedux-project. The feedback from them will be used for further development of the platform. We will work on more data analysis views specific to the alarm management on ICUs.

Competing interests: D. Hüske Kraus is employed by Philips Health Care. The other authors state that they have no conflict of interest.

The authors declare that an ethical vote was obtained for this study (Medizinische Ethik-Kommission der Universität Oldenburg, Nr. 052/2016, Chair: Prof. Dr. Frank Griesinger).