VentiProtect ARDS, vibe coding lung protection: a five-minute clinician-built ventilator tool as proof-of-concept for physician prompt engineering.

Authors

Sacha Rozencwajg

Published in

Anaesthesia, critical care & pain medicine. Pages 101934. Oct 16, 2026. Epub Oct 16, 2026.

Abstract

Lung-protective ventilation in acute respiratory distress syndrome (ARDS) remains under-applied at the bedside, partly due to the absence of integrated, context-aware decision-support tools. Large language models (LLMs) with advanced code generation capabilities have created a new possibility: physicians can now build the bedside tools they need without depending on software vendors or programming expertise. This practice, broadly termed "vibe coding", has entered medical discourse but has not been formally evaluated as a clinical competency. We report the development of VentiProtect ARDS, a real-time ventilator titration assistant built by a single physician in a single session of 4 minutes 25 seconds, as a proof-of-concept for physician prompt engineering: the structured translation of clinical knowledge into a natural-language specification that constrains LLM code generation. A structured eight-block prompt was developed through an iterative pre-prompt reasoning process and submitted to Claude Opus 4.6 via the Antigravity agentic development environment. The generated application computes predicted body weight, tidal volume, driving pressure, respiratory system compliance, mechanical power, and oxygenation index, with real-time colour-coded alerts and an active suggestion engine that calculates exact corrective ventilator settings. Formula accuracy was verified using an independent automated test suite of 100 scenarios (1,400 individual checks) with no computational errors identified. The tool runs locally with no installation, network dependency, or patient data transmission. This proof-of-concept suggests that physician prompt engineering may represent a potential clinical competency that does not require programming experience, although a single self-performed case cannot establish it across physicians. The LLM generates syntax. The physician generates medicine.

PMID:
42887688
Bibliographic data and abstract were imported from PubMed on 17 Oct 2026.

Read full publication at:
Please sign in to see all details.

Stats

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.