Heart failure & cardiomyopathiesCohort study
A No-Code, Guideline-Based Custom Generative Pretrained Transformer (GPT) Outperforms Cardiologists in Response Quality for Cardiac Amyloidosis
- prospective, blinded, comparative study
Authors' conclusion
A no-code, guideline-based custom GPT delivered superior response quality to that of cardiologists for CA questions. This approach allows clinicians without programming skills to build disease-specific large language models, potentially supporting equitable care where specialist access is limited. However, further studies are needed to evaluate potentially inaccurate outputs such as hallucinations.
Study context
- Population
- Clinical questions regarding cardiac amyloidosis generated by nonspecialist physicians
- Participants
- 71 unique clinical questions
- Exposure
- AmyloGPT vs board-certified cardiologist answers
Main finding
As reported in the abstract
Preferred by nonspecialist evaluators
81.1% (95% CI: 78.1%-83.8%)
Preferred by cardiologist evaluators
83.6% (95% CI: 78.6%-88.6%)
Observational data: associations, not effects.
Further findings
Median ratings for intent alignment and clinical usefulness (nonspecialist evaluators)
- Higher with AmyloGPT
- both P < 0.001
Median ratings for accuracy, consistency, validity, completeness, and absence of bias (cardiologist evaluators)
- Higher with AmyloGPT
- all P < 0.001
Original publication
Summarised from the published abstract; every number is checked against it