An Evaluation of Clinical Decision Support and Use of Machine Learning to Reduce Alert Fatigue

Abstract

Therapeutic duplication alert is one of the Clinical Decision Support Systems (CDSS) that was implemented to help physicians and other healthcare providers in making clinical judgements about the patients' management of therapy and decreasing medication errors. However, there were high override rates of these alerts by physicians as they were deemed to be of non-clinical significance. The quantity of the alerts fired by the system was high leading to "alert fatigue". Thus, the hospital administrators reached an agreement to deactivate it. To assess the validity of this decision, the aim of the study was to analyze the impact of therapeutic duplication alert deactivation on medication errors' rate. This study retrospectively screened a total of 593 electronic Medication Administration Records (e-MAR) of hospitalized patients with 297 e-MARs in the pre-therapeutic duplication alert deactivation period and 296 e-MARs in the post-therapeutic duplication alert deactivation period in a tertiary care hospital in Saudi Arabia. The number and type of duplicate medication errors in each period was documented to determine whether there was a significant difference between the two periods. The results detected 51 out of 297 e-MARs with medication errors in the pre-therapeutic duplication alert deactivation period versus 47 out of 296 in the post alert deactivation therapeutic duplication. Chi square test showed that there was no significant difference in the incidence of medication errors detected among the two periods with a p-value of 0.672. Therefore, we concluded that there was no significant difference on the medication error after the therapeutic duplication alert deactivation. An integration of machine learning into the clinical decision support design was recommended to filter the duplicated and unimportant alerts and reduce the alert fatigue of physicians.

Alerts in Clinical Decision Support Systems (CDSS): A Bibliometric Review and Content Analysis

“…Thus, alert implementation will not increase the clinical decision value equivalently [58,59]. A context-aware solution based on machine learning should be conducted in future studies [60,61].…”

Untitled

“…[10] Similarly, another study in Saudi Arabia found no significant difference in the number of medication errors discovered when one type of CDSS alert was activated versus when it was deactivated. [11] In other words, CDS systems are intended to aid in the sifting of massive volumes of digital data to recommend the next steps for treatments, alert physicians about available information they may not have noticed, or detect potential problems such as dangerous prescription combinations.…”

How artificial intelligence can help us ‘Choose Wisely’

“…AI algorithms, such as the Pulmonary Embolism Result Forecast Model, can be deployed within CDS software enabling the use of case-specific recommendations; if, in the case of a patient presenting to the emergency department with intermediate pretest probability of pulmonary embolism, the patient-specific likelihood of a positive CT Pulmonary Angiogram study is now known, the patient and provider may engage in decision-making that could reduce low value imaging in a sizable portion of patients (Banerjee et al, 2019 ). More accurately matching recommendations to patient contexts can help reduce the overall number of alerts for clinicians and mitigate alert fatigue (Chen et al, 2020 ; Khreis et al, 2019 ).…”