Back

Evaluation of Closed and Open Large Language Models in Pediatric Cardiology Board Exam Performance

Nikolovski, N.; Morgan, C. T.; Gritti, M. N.

2025-06-30 cardiovascular medicine
10.1101/2025.06.28.25330485 medRxiv
Show abstract

IntroductionLarge language models (LLMs) have gained traction in medicine, but there is limited research comparing closed- and open-source models in subspecialty contexts. This study evaluated ChatGPT-4.0o and DeepSeek-R1 on a pediatric cardiology board-style examination to quantify their accuracy and discuss clinical and educational utility. MethodsChatGPT-4.0o and DeepSeek-R1 were used to answer 88 text-based multiple-choice questions across 11 pediatric cardiology subtopics from a Pediatric Cardiology Board Review textbook. DeepSeek-R1s processing time per question was measured. Statistical analyses for model comparison were conducted using an unpaired two-tailed t-test, and bivariate correlations were assessed using Pearsons r. ResultsChatGPT-4.0o and DeepSeek-R1 achieved 70% (62/88) and 68% (60/88) accuracy, respectively (p=0.79). Subtopic accuracy was equal in 5 of 11 chapters, with each model outperforming its counterpart in 3 of 11. DeepSeek-R1s processing time negatively correlated with accuracy (r = -0.68, p = 0.02). ConclusionChatGPT-4.0o and DeepSeek-R1 approached the passing threshold on a pediatric cardiology board examination, with comparable accuracy and potential for open-source models to enhance clinical and educational outcomes while supporting sustainable AI development.

Matching journals

The top 9 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.