CTConstantine Tarabanis

Constantine Tarabanis

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Constantine Tarabanis has a total of 85 co-authors. Here are the top co-authors by number of shared publications:

  • Lior Jankelson

    Affiliation Leon H. Charney Division of Cardiology, NYU Langone Health, New York University School of Medicine, New York City, New York, United States of America

    7 shared publications
  • Evangelos Kalampokis

    Affiliation Information Systems Laboratory, University of Macedonia, Thessaloniki, Greece

    4 shared publications
  • Constantinos Bakogiannis

    Affiliation 3rd Cardiology Department, Hippokrateion University Hospital, Aristotle University of Thessaloniki, Thessaloniki, Greece

    3 shared publications
  • Dimitrios Tachmatzidis

    Affiliation 3rd Cardiology Department, Hippokrateion University Hospital, Aristotle University of Thessaloniki, Thessaloniki, Greece

    3 shared publications
  • Sohail Zahid

    Affiliation Leon H. Charney Division of Cardiology, NYU Langone Health, New York University School of Medicine, New York, New York, United States of America

    3 shared publications
  • Vassilios Vassilikos

    Affiliation 3rd Cardiology Department, Hippokrateion University Hospital, Aristotle University of Thessaloniki, Thessaloniki, Greece

    3 shared publications

Latest Publications

  • Open Access

    Artificial intelligence in clinical trials—state of the evidence, gaps, and next steps

    by Antonis A. Armoundas, Constantine Tarabanis, Joseph Loscalzo

  • Revisiting Digitalis in Heart Failure: Lessons From DIGIT-HF and Beyond

    by Constantine Tarabanis, Shaan Khurshid, Michael M. Givertz


  • Open Access

    Cardiology knowledge assessment of retrieval-augmented open versus proprietary large language models

    by Constantine Tarabanis, Shaan Khurshid, Areti Karamanou, Rodo Piperaki, Lucas A. Mavromatis, Aris Hatzimemos, Dimitrios Tachmatzidis, Constantinos Bakogiannis, Vassilios Vassilikos, Patrick T. Ellinor, et al.

    To evaluate the performance of open-weight and proprietary LLMs, with and without Retrieval-Augmented Generation (RAG), on cardiology board-style questions and benchmark them against the human average. We tested 14 LLMs (6 open-weight, 8 proprietary) on 449 multiple-choice questions from the American College of Cardiology Self-Assessment Program (ACCSAP). Accuracy was measured as percent correct. RAG was implemented using a knowledge base of 123 guideline and textbook documents. The open-weight model DeepSeek R1 achieved the highest accuracy at 86.9% (95% CI: 83.4–89.7%), outperforming proprietary models and the human average of 78%. GPT 4o (80.9%, 95% CI: 77.0–84.2%) and the commercial platform OpenEvidence (81.3%, 95% CI: 77.4–84.7%) demonstrated similar performance. A positive correlation between model size and performance was observed within model families, but across families, substantial variability persisted among models with similar parameter counts. After RAG, all models improved, and open-weight models like Mistral Large 2 (78.0%, 95% CI: 73.9–81.5) performed comparably to proprietary alternatives like GPT 4o. Large language models (LLMs) are increasingly integrated into clinical workflows, yet their performance in cardiovascular medicine remains insufficiently evaluated. Open-weight models can match or exceed proprietary systems in cardiovascular knowledge, with RAG particularly beneficial for smaller models. Given their transparency, configurability, and potential for local deployment, open-weight models, strategically augmented, represent viable, lower-cost alternatives for clinical applications. Open-weight LLMs demonstrate competency in cardiovascular medicine comparable to or exceeding that of proprietary models, with and without RAG depending on the model.


  • Open Access

    Artificial Intelligence-Enabled Electrocardiography in Practice: A State-of-the-Art Review

    by Hak Seung Lee, Philip M. Croon, Min Sung Lee, Timothy Poterucha, Chin Lin, Jeong Min Son, Ki-Hyun Jeon, Constantine Tarabanis, Seung-Pyo Lee, Kyung-Hee Kim, et al.

    Lee HS, et al. Korean Circ J. 2026 Jan;56:e55. https://doi.org/10.4070/kcj.2025.0486


  • Open Access

    Cardiology Knowledge Assessment of Retrieval-Augmented Open versus Proprietary Large Language Models

    by Constantine Tarabanis, Shaan Khurshid, Areti Karamanou, Rodo Piperaki, Lucas A. Mavromatis, Aris Hatzimemos, Dimitrios Tachmatzidis, Constantinos Bakogiannis, Vassilios Vassilikos, Patrick T. Ellinor, et al.

    Objectives To evaluate the performance of open and proprietary LLMs, with and without Retrieval-Augmented Generation (RAG), on cardiology board-style questions and benchmark them against the human average. Materials and Methods We tested 14 LLMs (6 open-weight, 8 proprietary) on 449 multiple-choice questions from the American College of Cardiology Self-Assessment Program (ACCSAP). Accuracy was measured as percent correct. RAG was implemented using a knowledge base of 123 guideline and textbook documents. Results The open-weight model DeepSeek R1 achieved the highest accuracy at 86.7% (95% CI: 83.7–89.9%), outperforming proprietary models and the human average of 78%. GPT 4o (80.8%, 95% CI: 77.2–84.5%) and the commercial platform OpenEvidence (80.4%, 95% CI: 76.7–84.0%) demonstrated similar performance. A positive correlation between model size and performance was observed within model families, but across families, substantial variability persisted among models with similar parameter counts. After RAG, all models improved, and open-weight models like Mistral Large 2 (78.0%, 95% CI: 74.1–81.8) performed comparably to proprietary alternatives like GPT 4o. Discussion Large language models (LLMs) are increasingly integrated into clinical workflows, yet their performance in cardiovascular medicine remains insufficiently evaluated. Open-weight models can match or exceed proprietary systems in cardiovascular knowledge, with RAG particularly beneficial for smaller models. Given their transparency, configurability, and potential for local deployment, open-weight models, strategically augmented, represent viable, lower-cost alternatives for clinical applications. Conclusion Open-weight LLMs demonstrate competency in cardiovascular medicine comparable to or exceeding that of proprietary models, with and without RAG depending on the model. Author Summary In this work, we set out to understand how today’s artificial intelligence systems perform when tested on the kind of questions cardiologists face during board examinations. We compared a wide range of large language models, including both freely available “open” models and commercial “proprietary” ones, and also tested whether giving the models access to trusted cardiology textbooks and guidelines could improve their answers. We found that the best open model actually outperformed all of the commercial models we tested, even exceeding the average score of practicing cardiologists. When we gave the models access to medical reference material, nearly all of them improved, with the biggest gains seen in the smaller and weaker models. This shows that careful design and support can allow smaller, more accessible systems to reach high levels of accuracy. Our results suggest that open models, which can be used locally without sending sensitive patient information to outside servers, may be a safe and cost-effective alternative to commercial products. This matters because it could make powerful AI tools more widely available across hospitals and clinics, while also reducing risks related to privacy, transparency, and cost.

  • Open Access

    Artificial intelligence–enabled sinus electrocardiograms for the detection of paroxysmal atrial fibrillation benchmarked against the CHARGE-AF score

    by Constantine Tarabanis, Vidya Koesmahargyo, Dimitrios Tachmatzidis, Vasileios Sousonis, Constantinos Bakogiannis, Robert Ronan, Scott A Bernstein, Chirag Barbhaiya, David S Park, Douglas S Holmes, et al.

    Aims: We aimed to develop and externally validate a convolutional neural network (CNN) using sinus rhythm electrocardiograms (ECGs) and CHARGE-AF features to predict incident paroxysmal atrial fibrillation (AF), benchmarking its performance against the CHARGE-AF score. Methods and results: We curated 157 192 sinus ECGs from 76 986 patients within the New York University (NYU) Langone Health system, splitting data into training, validation, and test sets. Two cohorts, from suburban US outpatient practices and Greek tertiary hospitals, were used for external validation. The model utilizing the sinus ECG signal and all CHARGE-AF features achieved the highest test set area under the receiver operator curve (AUC) (0.89) and area under the precision recall curve (AUPRC) (0.69), outperforming the CHARGE-AF score alone. Model robustness was maintained in the external US cohort (AUC 0.90, AUPRC 0.67) and the European cohort (AUC 0.85, AUPRC 0.78). Subgroup analyses confirmed consistent performance across age, sex, and race strata. A CNN using ECG signals alone retained strong predictive ability, particularly when simulating missing or inaccurate clinical data. Conclusion: Our CNN integrating sinus rhythm ECGs and CHARGE-AF features demonstrated superior predictive performance over traditional risk scoring alone for detecting incident paroxysmal AF. The model maintained accuracy across geographically and clinically diverse external validation cohorts, supporting its potential for broad implementation in AF screening strategies.


  • Open Access

    Performance of Publicly Available Large Language Models on Internal Medicine Board-style Questions

    by Constantine Tarabanis, Sohail Zahid, Marios Mamalis, Kevin Zhang, Evangelos Kalampokis, Lior Jankelson

    Ongoing research attempts to benchmark large language models (LLM) against physicians’ fund of knowledge by assessing LLM performance on medical examinations. No prior study has assessed LLM performance on internal medicine (IM) board examination questions. Limited data exists on how knowledge supplied to the models, derived from medical texts improves LLM performance. The performance of GPT-3.5, GPT-4.0, LaMDA and Llama 2, with and without additional model input augmentation, was assessed on 240 randomly selected IM board-style questions. Questions were sourced from the Medical Knowledge Self-Assessment Program released by the American College of Physicians with each question serving as part of the LLM prompt. When available, LLMs were accessed both through their application programming interface (API) and their corresponding chatbot. Mode inputs were augmented with Harrison’s Principles of Internal Medicine using the method of Retrieval Augmented Generation. LLM-generated explanations to 25 correctly answered questions were presented in a blinded fashion alongside the MKSAP explanation to an IM board-certified physician tasked with selecting the human generated response. GPT-4.0, accessed either through Bing Chat or its API, scored 77.5–80.7% outperforming GPT-3.5, human respondents, LaMDA and Llama 2 in that order. GPT-4.0 outperformed human MKSAP users on every tested IM subject with its highest and lowest percentile scores in Infectious Disease (80th) and Rheumatology (99.7th), respectively. There is a 3.2–5.3% decrease in performance of both GPT-3.5 and GPT-4.0 when accessing the LLM through its API instead of its online chatbot. There is 4.5–7.5% increase in performance of both GPT-3.5 and GPT-4.0 accessed through their APIs after additional input augmentation. The blinded reviewer correctly identified the human generated MKSAP response in 72% of the 25-question sample set. GPT-4.0 performed best on IM board-style questions outperforming human respondents. Augmenting with domain-specific information improved performance rendering Retrieval Augmented Generation a possible technique for improving accuracy in medical examination LLM responses.


  • Open Access

    A case report of sodium azide-induced myopericarditis

    by Constantine Tarabanis, Darcy Banco, Norma M Keller, Sripal Bangalore, Carlos L Alviar

    Background: Sodium azide exposures are rare but can be lethal as the substance inhibits complex IV in the electron transport chain, blocking adenosine-triphosphate (ATP) synthesis. Sodium azide is mostly used as a propellant in vehicular airbags but is also used in laboratory, pharmacy, and industrial settings. No known antidote exists and its cardiotoxic effects are poorly described in the literature. Case summary: We describe the case of a 31-year-old patient with major depressive disorder presenting with altered mental status after ingestion of an unknown amount of sodium azide. Although initially chest pain free, she developed pleuritic chest pain 48 h after ingestion. This was accompanied by new diffuse ST elevations on the electrocardiogram and serum troponin elevations concerning for myopericarditis. Treatment was pursued with a 14-day course of colchicine resulting in complete symptom resolution within 4 days of treatment initiation. The patient’s transthoracic echocardiogram was only notable for a preserved left ventricular ejection fraction (LVEF). Discussion: Cardiac toxicity after sodium azide ingestion usually occurs days after ingestion and has been previously described in the forms of heart failure with reduced ejection fraction complicated by cardiogenic shock. We describe the first case of sodium azide-induced myopericarditis with a preserved LVEF treated with colchicine. Colchicine is an established treatment for pericarditis, but its inhibition of endocytosis, an ATP-dependent cellular function, could be mechanistically relevant to this case.


  • Open Access

    Performance of Publicly Available Large Language Models on Internal Medicine Board-style Questions

    by Constantine Tarabanis, Sohail Zahid, Marios Mamalis, Kevin Zhang, Evangelos Kalampokis, Lior Jankelson

    We investigate the performance of four large language models (LLMs) on internal medicine board-style examination questions from the Medical Knowledge Self-Assessment Program released by the American College of Physicians. GPT4 outperformed GPT3.5, human users, LaMDA and LLaMA 2 in that order. A drop in GPT4 performance when accessing its API versus its publicly available chatbot (ChatGPT) was recovered through fine tuning in the form of Harrison’s Principles of Internal Medicine.