Truace
Subscribe

(AI)lephant in the Room: Utility of Generative AI to Develop Discussion Guides for Cardiac Surgery Morbidity and Mortality Conferences

Objective: Morbidity & Mortality Conferences (M&Ms) are forums to assess adverse outcomes and to analyze their root causes. We sought to utilize a large language model (LLM) to generate discussion materials for cardiac surgery M&Ms, based upon data available in the electronic medical record (EMR). Design: EMR data, including information from notes, bloodwork, vital sign monitoring, and the medication administration record, were input into a private instance of GPT-5. A prompt was iterated to accomplish three tas…

By the Truvace desk2 min read1 sourcePeer-reviewedTRV-2026-1666· cite & verify
(AI)lephant in the Room: Utility of Generative AI to Develop Discussion Guides for Cardiac Surgery Morbidity and Mortality Conferences
Material for the study of surgery, the surgery of the kidney and its adnexa : being a series of cases selected for class-room study by Holmes, Bayard, 1852-1924. Public domain

In brief

Researchers in a high-volume cardiothoracic surgery department fed electronic medical record data from fourteen cardiac surgery cases into a private instance of GPT-5. With an iterated prompt, the model produced a case outline, a table flagging deviations from evidence-based recommendations, and discussion questions for morbidity and mortality conferences.

The work suggests language models could reduce the preparation burden of M&M conferences while keeping the focus on systems-based discussion of adverse events. The evidence is small in scale, though, and rests on a handful of trainee preferences rather than on any measured change in patient outcomes or conference quality.

Main points

  1. Inputs included notes, bloodwork, vital sign monitoring and the medication administration record from fourteen cardiac surgery cases.
  2. Median outline length was similar: 453.5 words for LLM versus 487.5 for trainees (p = 0.68).
  3. LLM outlines had a higher Flesch-Kincaid Grade Level, 13.9 versus 11.6 (p = 0.003).
  4. Nine trainees completed a forced-choice survey; baseline information in LLM outlines was preferred 16 of 27 times.

The gain

A private GPT-5 instance produced case outlines, deviation tables and discussion questions for cardiac surgery M&Ms from EMR data, matching trainee outline length and earning high perceived clinical applicability.

The rundown

The study tasked the model with three outputs: a case outline, a discussion table to identify deviations from evidence-based recommendations, and discussion questions. The setting was a department with both residency and fellowship training programs.

Readability was assessed with the Flesch-Kincaid Grade Level, and the LLM-generated outlines scored significantly higher, meaning they were written at a more difficult reading level than trainee-generated outlines.

In the two-alternative forced-choice survey, trainees favored the LLM outline for baseline information at 59.3 percent versus 40.7 percent, while preferences for operative and perioperative information were more evenly split. The authors concluded the model effectively synthesized EMR information to outline cases.

Sources

  1. Peer-reviewedJournal of Surgical Education2026-10-13

The debate