gms | German Medical Science

Introduction: Alarm floods occur when operators receive more alarms in a period of time than they can manage appropriately [1]. In the intensive care unit (ICU) setting this can lead to alarm fatigue, where staff gets desensitized by a high number of often (72-99%) irrelevant alarms, which endangers patients [2]. In industrial plant alarm management alarm floods are interpreted as ≥ 10 alarms in 10 minutes [3]. This interpretation was previously adapted in the medical field [4] although its suitability in this setting is unknown. We aim to explore different applications of alarm floods based on the ISO definition [1] and their association with alarm relevance.

Methods: Following IRB approval (EA1/127/18), we retrospectively analyzed annotated alarm data [5] from July-September 2022 (21-bed ICU, 92 days, 210 patients, 175,804 alarms). An alarm is considered actionable and relevant if followed by a related medical intervention. An alarm flood is defined as ≥ x alarms in a y-minute interval per bed, with x y, 2*y and y 5,10,15. y-minute intervals as (a) sliding bins starting at each (actionable) alarm and (b) fixed overlapping bins shifted by y/2. Actionable alarm floods refer to only actionable alarms. Metrics include mean and standard deviation of (actionable) alarm floods per day.

Results: The mean number of alarm floods per day varies immensely by approach, whereas the proportion of actionable alarm floods remains similar (Table 1 [Tab. 1]). The daily mean of non-actionable alarm floods across explored applications is 86-92% of the daily mean of all alarm floods. Sliding bins measure an order of magnitude more alarm floods compared to fixed bins.

Discussion: All approaches show a non-actionable alarm flood rate of 86-92%, similar to the rate of irrelevant alarms [2]. Measuring more alarm floods when using sliding bins as compared to fixed bins suggests actionable and non-actionable alarms often occur clustered. This is plausible as a single event can produce multiple alarms. However, it results in the sliding bin approach multi-counting one alarm into multiple alarm floods, overestimating the occurrence of alarm floods and thereby the alarm load. We propose an approach that limits multi-counting, such as fixed bins. ICU alarms should be analyzed in the context of the overall alarm situation and the temporal dimension as alarm clusters. Limitations of this research include disregarding the available number of alarm-managing operators and patients.

Conclusion: This research demonstrates analyzing ICU alarms as alarm clusters by interpreting the ISO definition of alarm floods and suggests a fixed bin approach to measure them. Non-actionable alarm floods account for 86-92% of all alarm floods, highlighting the need for alarm flood reduction to alleviate alarm fatigue. Further research is required to investigate a more precise interpretation of alarm floods in medical settings, considering the number of alarm-managing operators and patients, as well as approaches to minimize their occurrence such as mechanisms to individualize and automatically adjust alarm thresholds in contexts where they lead to non-actionable alarm floods.

Grant: BMBF 16SV8559.

The authors declare that they have no competing interests.

The authors declare that a positive ethics committee vote has been obtained.