The Acceptance of Interruptive Medication Alerts in an Electronic Decision Support System Differs between Different Alert Types

Abstract

Background Through targeted medication alerts, clinical decision support systems (CDSS) help users to identify medication errors such as disregarded drug–drug interactions (DDIs). Override rates of such alerts are high; however, they can be mitigated by alert tailoring or workflow-interrupting display of severe alerts that need active user acceptance or overriding. Yet, the extent to which the displayed alert interferes with the prescribers' workflow showed inconclusive impact on alert acceptance. Objectives We aimed to assess whether and how often prescriptions were changed as a potential result of interruptive alerts on different (contraindicated) prescription constellations with particularly high risks for adverse drug events (ADEs). Methods We retrospectively collected data of all interruptive alerts issued between March 2016 and August 2020 in the local CDSS (AiDKlinik) at Heidelberg University Hospital. The alert battery consisted of 31 distinct alerts for contraindicated DDI with simvastatin, potentially inappropriate medication for patients > 65 years (PIM, N = 14 drugs and 36 drug combinations), and contraindicated drugs in hyperkalemia (N = 5) that could be accepted or overridden giving a reason in free-text form. Results In 935 prescribing sessions of 500 274 total sessions, at least one interruptive alert was fired. Of all interruptive alerts, about half of the sessions were evaluable whereof in total 57.5% (269 of 468 sessions) were accepted while 42.5% were overridden. The acceptance rate of interruptive alerts differed significantly depending on the alert type (p <0.0001), reaching 85.7% for DDI alerts (N = 185), 65.3% for contraindicated drugs in hyperkalemia (N = 98), and 25.1% for PIM alerts (N = 185). Conclusion A total of 57.5% of the interruptive medication alerts with particularly high risks for ADE in our setting were accepted while the acceptance rate differed according to the alert type with contraindicated simvastatin DDI alerts being accepted most frequently.

Modulators Influencing Medication Alert Acceptance: An Explorative Review

“…The large majority of the 35 quantitative factors ( n = 22) were studied once, whereas nine factors were investigated twice (i.e., alert display, 33 64 filtering, clustering or deactivation of alerts, 46 61 interruptive alerts, 45 56 alert frequency, 33 54 inclusion of patient-specific context factors, 60 64 care provider's professional status, 62 63 laboratory value, 49 57 weekday, 39 41 and drug triggering the alert 57 62 ). Two factors were analyzed three times (i.e., tiering of alerts according to severity 33 45 64 and care provider's department 39 57 62 ), one factor four times in two different articles (i.e., assessment of alert relevance by care provider 50 52 ), and one factor was studied seven times in seven independent articles (i.e., alert type 39 40 41 48 54 62 65 ).…”

“…After the removal of 110 duplicates and the exclusion of 276 articles following title and abstract screening, a total of 153 full texts were read. In compliance with the inclusion and exclusion criteria, 31 articles reporting quantitative and qualitative parameters 13 33 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 and 29 articles reporting exclusively qualitative parameters of alert acceptance 1 6 12 17 32 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 were included in the analysis.…”

“…However, the alert type as one single factor was analyzed in seven different settings with six different CDSS software vendors (SafeRx ® -CDSS, in-house system of Brigham Integrated Clinical Information System, Epic ® , Cerner, DARWIN's CDSS, and the stand-alone system AiD Klinik ® ) and had equal impact on alert acceptance. 39 40 41 48 54 62 65 Due to the fact that DAI alerts were accepted more often than DDI alerts in four settings 41 48 54 62 and less often in only one setting, 40 it can be discussed whether study and implementation settings varied too much to merge these parameters into one factor. Moreover, both studies using Cerner's CDSS software and compared DDI and DAI, achieved higher acceptance rates for DAI.…”

Trends in Drug Overdose Deaths Among US Adolescents, January 2010 to June 2021

“…However, the frequent alerts led to a fatigue phenomenon, and physicians eventually ended up not taking them into account. 2 Likewise, one can imagine observing a fatigue alert linked to the wayfinding provided by next-generation artificial intelligence. Therefore, it is vital to set the recommendation and intervention level of the system.…”

Supporting Diagnosis With Next-Generation Artificial Intelligence

“…The strategy of alerting physicians initially seemed relevant and necessary to help reduce prescription errors. However, the frequent alerts led to a fatigue phenomenon, and physicians eventually ended up not taking them into account . Likewise, one can imagine observing a fatigue alert linked to the wayfinding provided by next-generation artificial intelligence.…”