Author: Nadali. Prepared with AI assistance.
A new publication, not a new launch
On October 30, JAMA Network Open published a study of AI-drafted patient replies at Columbia University Irving Medical Center. It analyzed first responses from 2023–2025; the Epic tool was rolled out in 2024.
The journal’s announcement reports that adoption was associated with longer turnaround, while using a draft for a particular message was associated with shorter turnaround. No overall reduction in after-hours responses was found.
This single-center observational evidence establishes neither causation nor payroll savings. Time until a first reply is not staff working time; the study does not cover all subsequent exchanges.
Who benefits
Clinic owners or operations managers assessing a messaging-assistance pilot. Nadali’s practical takeaway: do not approve expansion on a speed metric alone.
What to add to your checks
This is an editorial checking sequence, not the study protocol or medical guidance:
- Separate three groups: before adoption, after adoption without using a draft, and after adoption with draft use. Compare enquiries of similar complexity.
- Record patient waiting time, active preparation and review time, and replies outside the agreed schedule separately.
- Include corrections and follow-up enquiries. A quickly sent response does not necessarily resolve the issue.
- Decide in advance who assesses quality and authorizes further use. Do not treat this study as grounds for automatically sending medical replies.
Our small-pilot worksheet can help organize measurement. Keep your working records free of unnecessary personal information and within an approved environment.
Availability and limits
The research article is publicly accessible. It does not establish access to the described tool, integration or terms in your clinic. Nadali has not tested the product or assessed patient records; sources were read on November 3, 2026.
The cover is a reused conceptual Nadali AI illustration: fictional records, not study results or medical data.
