receipt
Response tool adoption was associated with an 8.6% (95% CI, 4.6%-12.9%; P < .001) increase in overall message turnaround time.Abstract, ResultsJAMA Network Open: AI-Generated Draft Response Tool, Message Turnaround Time, and After-Hours Replies Among Health Care Professionals (Jeanselme et al.), PMC · pmc.ncbi.nlm.nih.gov
Clinicians at Columbia who adopted Epic's AI-drafted replies to patient messages answered 8.6% more slowly afterwards, and their after-hours replying did not fall, across 983,154 responses.
receipt
Response tool adoption was associated with an 8.6% (95% CI, 4.6%-12.9%; P < .001) increase in overall message turnaround time.Abstract, ResultsJAMA Network Open: AI-Generated Draft Response Tool, Message Turnaround Time, and After-Hours Replies Among Health Care Professionals (Jeanselme et al.), PMC · pmc.ncbi.nlm.nih.gov
Response tool use was not associated with an overall reduction of after-hours messaging burdenAbstract, ResultsJAMA Network Open: AI-Generated Draft Response Tool, Message Turnaround Time, and After-Hours Replies Among Health Care Professionals (Jeanselme et al.), PMC · pmc.ncbi.nlm.nih.gov
The analysis included 983 154 responses from 1542 cliniciansAbstract, ResultsJAMA Network Open: AI-Generated Draft Response Tool, Message Turnaround Time, and After-Hours Replies Among Health Care Professionals (Jeanselme et al.), PMC · pmc.ncbi.nlm.nih.gov
A cohort study (JAMA Network Open, October 2026) followed 1,542 clinicians at Columbia University Irving Medical Center before and after their first use of Epic's generative-AI draft-reply tool, with clinician and calendar-month fixed effects; adoption was associated with longer time to answer patients, and using a draft was not associated with less after-hours replying overall.
Sources
- primaryJAMA Network Open: AI-Generated Draft Response Tool, Message Turnaround Time, and After-Hours Replies Among Health Care Professionals (Jeanselme et al.), PMCpmc.ncbi.nlm.nih.gov
Editor's notes
Filed under not-because on the adoption estimate, the one that reflects having the tool at all; the 29.5% faster figure for messages answered with a draft compares messages clinicians selected, which the authors themselves flag as prone to selection bias. Both are shown. Dates: the analysis keeps clinicians who first adopted the tool on or after 12 September 2024 (the final prompt update), and the data end on 15 May 2025; enterprise rollout was 17 December 2024. Window pre. Compare ucla-chart-summarizer-rct and ambient-scribes-nejm-ai.
Related
History
- Added. Weekly run; read in full from the PMC copy of the JAMA Network Open article.
Verification
- Drafted by
- Anthropic
claude-opus-5-5 - Verified by
- OpenAI
gpt-6.1-sol· pass - How sure
- Strict
- Record
- data/verifications/columbia-epic-ai-draft-replies-null/2026-11-09.json · The source supports the reported associations, sample counts, and dates, including 8.6% longer turnaround after adoption and no significant overall after-hours reduction. Its observational design does not establish that AI caused these outcomes.
Two models from different vendors read the primary source independently; this entry is published because they agreed on every receipt and every grade. The models grading an AI registry have an interest in the answer. Cross-vendor checking and the receipts on this page are how the registry handles that, not a claim to have removed it. How the gate works · Dispute this entry
Cite this entry
“Clinicians at Columbia who adopted Epic's AI-drafted replies to patient messages answered 8.6% more slowly afterwards, and their after-hours replying did not fall, across 983,154 responses.” Because of AI, updated Nov 9, 2026. https://becauseofai.org/entry/columbia-epic-ai-draft-replies-null@misc{becauseofai_columbia_epic_ai_draft_replies_null,
title = {Clinicians at Columbia who adopted Epic's AI-drafted replies to patient messages answered 8.6% more slowly afterwards, and their after-hours replying did not fall, across 983,154 responses.},
howpublished = {\url{https://becauseofai.org/entry/columbia-epic-ai-draft-replies-null}},
note = {Because of AI registry, updated Nov 9, 2026},
year = {2026}
}