Answers are grounded in this paper and cite their sources.
Rationale Clinical decision support (CDS) tools require timely availability of accurate data in the electronic health record (EHR). In order to facilitate timely and accurate intervention, these tools require updated and accurate information which can be limited by data entry errors and critical information not contained in structured data fields, such as data contained in clinical notes. Interruptive alerts may be a method for acquiring timely and accurate data that is otherwise unavailable in the EHR. We evaluated the timeliness and accuracy of otherwise unstructured data acquired from clinicians via interruptive alert to inform a machine learning model for research. Methods Between August 21st, 2025 and October 22nd, 2025, a provider-facing interruptive alert was implemented in an Intensive Care Unit (ICU) at Vanderbilt University Medical Center to acquire data needed for a machine learning algorithm within the EHR to predict the optimal oxygen target for patients receiving mechanical ventilation. The alert was triggered when the clinicians opened the encounter for a patient receiving mechanical ventilation. Clinicians answered “yes” or “no” to the presence of 6 diagnoses (see Figure 1). To determine the impact of alert timing, on September 24 the alert was modified to not allow the clinicians to “delay” responding until the next time the patient chart was opened. The primary outcome was the accuracy of the diagnoses compared to manual review of the health record by a critical care fellow. The secondary outcome was time from first alert to complete data entry defined as selecting a response for all six ICU diagnoses. Chi-squared testing was used to analyze accuracy and a Students T-test was used to compare timing. Results Clinicians provided data on 110 patients through the interruptive alert during the study period. Chart review was conducted on 52 patients in the Delay and 48 in the No Delay cohorts. During the period with the defer button, diagnoses were entered on average within 175 minutes of the initial interruptive alert firing compared to 7 minutes without the defer button (p < 0.001). The average number of diagnoses correct per patient was 5.6 in the Delay Cohort versus 5.6 out of 6 in the No Delay Cohort (p = 0.67). Conclusions An interruptive alert can be used to acquire accurate data for CDS tools for clinical care or research. The addition of a defer option led to a longer time to completion without improving accuracy.
No takes yet. Share an insight, caveat, or question.
LaNasa et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: