The AI footprint in cardiology: rapid rise of large language models-associated vocabulary in ESC congress abstracts (2019–2025)
F Buendia Santiago, J M Andreu Cayuelas, A Meseguer Hernandez, M I Banos Maturano, J Merino Romero, M B Villamarin Heredia, L Risco Eres, I Gonzalez Fernandez, C Azcondo San Segundo, F Selles Lopez, M Gimenez Pinana, J M Cubillas Para, J C Bonaque Gonzalez, P Cano Vivar, J A Castillo MorenoAbstract
Introduction
Since the release of ChatGPT in late 2022, Large Language Models (LLMs) have transformed scholarly writing. Recent evidence suggests that LLMs have an unprecedented impact on scientific vocabulary; however, the extent of this "AI footprint" in the field of cardiology remains largely unquantified.
Objective
To describe the evolution in the frequency of use of specific "LLM-style" keywords in abstracts accepted to the European Society of Cardiology (ESC) Congresses from 2019 to 2025.
Methods
We conducted an observational study analysing the frequency of five keywords previously identified by various authors as markers of LLM-assisted writing (significant, additionally, enhance, comprehensive, and crucial) in ESC Congress abstracts as of October 2025. We compared the frequency of use between the pre-LLM baseline period (2019–2022) and the individual years 2024, and 2025 using Pearson's chi-squared tests. Year 2023 was analysed descriptively as an early post-release transitional period but was not included in the primary inferential comparisons. A Cochran–Armitage test was performed to evaluate linear trends in word frequency across the entire study period. This analysis evaluated linguistic patterns and does not allow confirmation of direct LLM use by individual authors.
Results
A total of 26,874 abstracts were analysed. All five keywords showed a significant increase in frequency in the 2024 and 2025 congresses compared to the 2019–2022 baseline (all p < 0.005, results remained significant after Bonferroni correction). Notably, the word "significant" increased from 53.3% to 63.1% in 2025, while lower-frequency words showed even more dramatic relative growth: "crucial" nearly tripled (2.1% to 6.1%) and "comprehensive" more than doubled (3.6% to 8.3%). The Cochran–Armitage test confirmed a highly significant upward linear trend for all analysed words (p < 0.001), with the most pronounced shift occurring after the 2022 congress (Figure).
Conclusions
The marked post-2022 increase in LLM-associated vocabulary in ESC abstracts is consistent with the growing availability and use of LLM tools in the research workflow within cardiology. This linguistic “AI footprint” is consistent with progressive integration of LLMs into the research workflow. While such tools may improve drafting efficiency, their potential impact on linguistic diversity and clinical nuance warrants further investigation.Table 1.Abstracts and words per yearFigure 1.Evolution of AI-Associated Vocab