High sensitivity for peripartum cardiomyopathy among large language models during differential diagnosis consideration
Thomas Gausepohl, Melanie Ricke-Hoch, Denise Hilfiker-Kleiner, Johann Bauersachs, Tobias Jonathan PfefferBackground
Peripartum cardiomyopathy (PPCM) is a rare, potentially life threatening condition with nonspecific heart failure symptoms, often leading to delayed diagnosis. Large language models (LLMs) may support differential diagnosis suggestion.
Objectives
To assess whether commonly available LLMs can suggest PPCM and differential diagnoses based on typical symptoms.
Design
Comparative evaluation of three LLMs using standardized clinical scenarios.
Methods
ChatGPT (GPT-5), Gemini (2.5 Flash), and Claude (Sonnet 4.5) were queried with prompts describing a 34-year-old woman with PPCM-related symptoms, with and without postpartum context. Responses were analyzed for PPCM and key differential diagnoses recommended by the European Society of Cardiology.
Results
ChatGPT and Claude consistently identified PPCM. Gemini suggested PPCM mainly when dyspnea or edema were present and required postpartum context when palpitations were reported. Claude listed the most differential diagnoses.
Conclusion
LLMs are a valuable tool among differential diagnosis suggestion but performance varies and cannot replace clinical judgment.