Related Experiment Video
Updated: May 13, 2026
Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness
Published on: December 6, 2024
Ambient Artificial Intelligence Scribes in Pediatric Hematology-Oncology: Early Implementation of DAX Copilot
Molly Talman1,2, Lamia Alam2, Dara Mize2,3
1Department of Pediatrics, Vanderbilt University Medical Center, Nashville, Tennessee, United States.
Background:
Ambient artificial intelligence scribes are increasingly integrated into electronic health records to reduce documentation burden, but limited data describe their performance in high-complexity pediatric subspecialties.
Objectives:
This study aimed to describe the early implementation of Dragon Ambient Experience (DAX) Copilot in a pediatric hematology-oncology division and evaluate patterns of use, documentation time, and workflow effect among providers who adopted the tool.
Methods:
We reviewed outpatient encounters from January to July 2025 and identified those using DAX. Analyses focused on providers who used DAX more than twice, given low overall adoption. Documentation time (minutes spent in active editing sessions) and note-closure timeliness were compared between DAX and non-DAX encounters. Implementation processes, barriers, and accuracy concerns were qualitatively summarized.
Results:
Of 11,544 outpatient encounters, 427 (3.7%) involved DAX; 10 of 29 providers used DAX at least once, while 6 (20.7%) used it more than twice. Among repeat users, for shorter encounters (≤40 minutes), median documentation time was lower for DAX versus non-DAX (11 [interquartile range, (IQR): 13] vs. 24 [IQR: 30] minutes), corresponding to an approximately 30% lower documentation time (95% confidence interval [CI], 19.9-38.4%; p < 0.001). For longer encounters (>40 minutes), documentation time did not differ significantly. Although unadjusted analyses showed lower rates of same-day and 7-day note closure with DAX, these differences were not significant after adjustment in mixed-effects logistic regression models. Providers reported improved engagement with families and narrative drafting; limitations were related to workflow alignment rather than transcription accuracy.
Conclusion:
For repeat users, DAX reduced active documentation time in shorter visits but showed limited benefit for longer encounters and did not improve the timeliness of note closure. Adoption remained modest, underscoring the importance of workflow fit and usability considerations in AI-scribe deployment.