Constantine Tarabanis

Constantine Tarabanis

Auto-generated profile, based on publicly available information

Constantine Tarabanis

Cardiology Division, Heart and Vascular Institute, Mass General Brigham, Boston, MA, USA

Research Focus & Metrics

Academic Career Length

8Years (2019-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
16 June 2023 in Cardiovascular Digital Health Journal
Verified
Journal Article
Open access
CTConstantine Tarabanis
Constantine Tarabanis
EKEvangelos Kalampokis
Evangelos Kalampokis
MKMahmoud Khalil
Mahmoud Khalil
CACarlos L. Alviar
Carlos L. Alviar
LCLarry A. Chinitz
Larry A. Chinitz
LJLior Jankelson
Lior Jankelson
A lack of explainability in published machine learning (ML) models limits clinicians' understanding of how predictions are made, in turn undermining uptake of the models into clinical practice. The purpose of this study was to develop explainable ML models to predict in-hospital mortality in patients hospitalized for myocardial infarction (MI). Adult patients hospitalized for an MI were identified in the National Inpatient Sample between January 1, 2012, and September 30, 2015. The resulting cohort comprised 457,096 patients described by 64 predictor variables relating to demographic/comorbidity characteristics and in-hospital complications. The gradient boosting algorithm eXtreme Gradient Boosting (XGBoost) was used to develop explainable models for in-hospital mortality prediction in the overall cohort and patient subgroups based on MI type and/or sex. The resulting models exhibited an area under the receiver operating characteristic curve (AUC) ranging from 0.876 to 0.942, specificity 82% to 87%, and sensitivity 75% to 87%. All models exhibited high negative predictive value ≥0.974. The SHapley Additive exPlanation (SHAP) framework was applied to explain the models. The top predictor variables of increasing and decreasing mortality were age and undergoing percutaneous coronary intervention, respectively. Other notable findings included a decreased mortality risk associated with certain patient subpopulations with hyperlipidemia and a comparatively greater risk of death among women below age 55 years. The literature lacks explainable ML models predicting in-hospital mortality after an MI. In a national registry, explainable ML models performed best in ruling out in-hospital death post-MI, and their explanation illustrated their potential for guiding hypothesis generation and future study design.

Verified
Journal Article
Open access
PDF file
Alexandros LaiosAlexandros Laios
Alexandros Laios
EKEvangelos Kalampokis
Evangelos Kalampokis
RJRACHEAL LOUISE JOHNSON
RACHEAL LOUISE JOHNSON
ATAmudha Thangavelu
Amudha Thangavelu
CTConstantine Tarabanis
Constantine Tarabanis
DJDiederick de Jong
Diederick de Jong
Complete surgical cytoreduction (R0 resection) is the single most important prognosticator in epithelial ovarian cancer (EOC). Explainable Artificial Intelligence (XAI) could clarify the influence of static and real-time features in the R0 resection prediction. We aimed to develop an AI-based predictive model for the R0 resection outcome, apply a methodology to explain the prediction, and evaluate the interpretability by analysing feature interactions. The retrospective cohort finally assessed 571 consecutive advanced-stage EOC patients who underwent cytoreductive surgery. An eXtreme Gradient Boosting (XGBoost) algorithm was employed to develop the predictive model including mostly patient- and surgery-specific variables. The Shapley Additive explanations (SHAP) framework was used to provide global and local explainability for the predictive model. The XGBoost accurately predicted R0 resection (area under curve [AUC] = 0.866; 95% confidence interval [CI] = 0.8–0.93). We identified “turning points” that increased the probability of complete cytoreduction including Intraoperative Mapping of Ovarian Cancer Score and Peritoneal Carcinomatosis Index < 4 and 4, patient’s age < 60 years, and largest tumour bulk < 5 cm in a surgical environment of optimized infrastructural support. We demonstrated high model accuracy for the R0 resection prediction in EOC patients and provided novel global and local feature explainability that can be used for quality control and internal audit.

17 September 2024 in PLOS Digital Health
Verified
Journal Article
Open access
PDF file
CTConstantine Tarabanis
Constantine Tarabanis
SZSohail Zahid
Sohail Zahid
MMMarios Mamalis
Marios Mamalis
KZKevin Zhang
Kevin Zhang
EKEvangelos Kalampokis
Evangelos Kalampokis
LJLior Jankelson
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.

7 September 2022 in Journal of Cardiovascular Electrophysiology
Verified
Journal Article
Closed access
CTConstantine Tarabanis
Constantine Tarabanis
JKJonah Klapholz
Jonah Klapholz
SZSohail Zahid
Sohail Zahid
LJLior Jankelson
Lior Jankelson

26 August 2023 in Nutrition, Metabolism and Cardiovascular Diseases
Verified
Journal Article
Closed access
CTConstantine Tarabanis
Constantine Tarabanis
CLClarine Long
Clarine Long
BSBianca Scolaro
Bianca Scolaro
SHSean P. Heffron
Sean P. Heffron

12 March 2026 in PLOS Digital Health
Verified
Journal Article
Open access
PDF file
CTConstantine Tarabanis
Constantine Tarabanis
SKShaan Khurshid
Shaan Khurshid
AKAreti Karamanou
Areti Karamanou
RPRodo Piperaki
Rodo Piperaki
LMLucas A. Mavromatis
Lucas A. Mavromatis
EKEvangelos Kalampokis
Evangelos Kalampokis
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.