Aaron Eisman

I am an Internal Medicine Resident at Yale-New Haven Hospital as part of the Physician-Scientist Training Pathway in Cardiovascular Medicine. I earned my MD/PhD in Computational Biology and Biomedical Informatics at Brown University in 2024 under the direction of Neil Sarkar at the Center for Biomedical Informatics

My research interests focus on the development of informatics methods that leverage population-level data to improve the accuracy of cardiovascular disease risk estimation towards a learning healthcare system. My work aims to improve adherence to clinical practice guidelines, enhance the precision of preventative medical therapies, better explain observed health outcome disparities in racial and ethnic minority populations, and develop mechanisms for translating omics research into clinical medical practice. I received an Sc.B. from Brown University in Applied Mathematics.

Before returning to Brown, I spent two years as a clinical research coordinator for the Cardiopulmonary Exercise Laboratory at Massachusetts General Hospital. As a member of that research team, we worked to understand cardiovascular pathophysiology including heart failure and pulmonary hypertension using exercise as a physiologic probe. Among other projects, we demonstrated that increased pulmonary capillary wedge pressure relative to cardiac output during exercise predicts exercise capacity and heart failure outcomes. These findings have informed the defintions of both heart failure with preserved ejection fraction (2021) and pulmonary hypertension (2022) in the European Society of Cardiology / European Respiratory Society guidelines.

selected publications

Protein-metabolite association studies identify novel proteomic determinants of metabolite levels in human plasma

Mark D Benson*, Aaron S Eisman*, Usman A Tahir, and 38 more authors

Cell Metabolism Sep 2023

Abs

Although many novel gene-metabolite and gene-protein associations have been identified using high-throughput biochemical profiling, systematic studies that leverage human genetics to illuminate causal relationships between circulating proteins and metabolites are lacking. Here, we performed protein-metabolite association studies in 3,626 plasma samples from three human cohorts. We detected 171,800 significant protein-metabolite pairwise correlations between 1,265 proteins and 365 metabolites, including established relationships in metabolic and signaling pathways such as the protein thyroxine-binding globulin and the metabolite thyroxine, as well as thousands of new findings. In Mendelian randomization (MR) analyses, we identified putative causal protein-to-metabolite associations. We experimentally validated top MR associations in proof-of-concept plasma metabolomics studies in three murine knockout strains of key protein regulators. These analyses identified previously unrecognized associations between bioactive proteins and metabolites in human plasma. We provide publicly available data to be leveraged for studies in human metabolism and disease.
Clinical Note Section Detection Using a Hidden Markov Model of Unified Medical Language System Semantic Types

Aaron S Eisman, Katherine A Brown, Elizabeth S Chen, and 1 more author

AMIA Annual Symposium Proceedings Sep 2021

Abs HTML

Clinical notes are a rich source of biomedical data for natural language processing (NLP). The identification of note sections represents a first step in creating portable NLP tools. Here, a system that used a heterogeneous hidden Markov model (HMM) was designed to identify seven note sections: (1) Medical History, (2) Medications, (3) Family and Social History, (4) Physical Exam, (5) Labs and Imaging, (6) Assessment and Plan, and (7) Review of Systems. Unified Medical Language System (UMLS) concepts were identified using MetaMap, and UMLS semantic type distributions for each section type were empirically determined. The UMLS semantic type distributions were used to train the HMM for identifying clinical note sections. The system was evaluated relative to a template boundary model using manually annotated notes from the Medical Information Mart for Intensive Care III. The results show promise for an approach to segment clinical notes into sections for subsequent NLP tasks.
Extracting Angina Symptoms from Clinical Notes Using Pre-Trained Transformer Architectures

Aaron S Eisman, Nishant R Shah, Carsten Eickhoff, and 4 more authors

AMIA Annual Symposium Proceedings Sep 2020

Abs HTML

Anginal symptoms can connote increased cardiac risk and a need for change in cardiovascular management. In this study, a pre-trained transformer architecture was used to automatically detect and characterize anginal symptoms from within the history of present illness sections of 459 primary care physician notes. Consecutive patients referred for cardiac testing were included. Notes were annotated for positive and negative mentions of chest pain and shortness of breath characterization. The results demonstrate high sensitivity and specificity for the detection of chest pain or discomfort, substernal chest pain, shortness of breath, and dyspnea on exertion. Model performance extracting factors related to provocation and palliation of chest pain were limited by small sample size. Overall, this study shows that pre-trained transformer architectures have promise in automating the extraction of anginal symptoms from clinical texts.
Pulmonary Capillary Wedge Pressure Patterns During Exercise Predict Exercise Capacity and Incident Heart Failure

Aaron S Eisman, Ravi V Shah, Bishnu P Dhakal, and 9 more authors

Circulation. Heart failure May 2018

Abs HTML

BACKGROUND: Single measurements of left ventricular filling pressure at rest lack sensitivity for identifying heart failure with preserved ejection fraction (HFpEF) in patients with dyspnea on exertion. We hypothesized that exercise hemodynamic measurements (ie, changes in pulmonary capillary wedge pressure [PCWP] indexed to cardiac output [CO]) may more sensitively differentiate HFpEF and non-HFpEF disease states, reflect aerobic capacity, and forecast heart failure outcomes in individuals with normal PCWP at rest. METHODS AND RESULTS: We studied 175 patients referred for cardiopulmonary exercise testing with hemodynamic monitoring: controls (n=33), HFpEF with resting PCWP\geq15 mm Hg (n=32), and patients with dyspnea on exertion with normal resting PCWP and left ventricular ejection fraction (DOE-nlrW; n=110). Across 1835 paired PCWP-CO measurements throughout exercise, we used regression techniques to define normative bounds of “PCWP/CO slope” in controls and tested the association of PCWP/CO slope with exercise capacity and composite cardiac outcomes (defined as cardiac death, incident resting PCWP elevation, or heart failure hospitalization) in the DOE-nlrW group. Relative to controls (PCWP/CO slope, 1.2\pm0.4 mm Hg/L/min), patients with HFpEF had a PCWP/CO slope of 3.4\pm1.9 mm Hg/L/min. We used a threshold (2 SD above the mean in controls) of 2 mm Hg/L/min to define abnormal. PCWP/CO slope >2 in DOE-nlrW patients was common (n=45/110) and was associated with reduced peak Vo2 (P2 mm Hg/L/min) is common in DOE-nlrW and predicts exercise capacity and heart failure outcomes. These findings suggest that current definitions of HFpEF based on single measures during rest are insufficient and that assessment of exercise PCWP/CO slope may refine early HFpEF diagnosis.
An Automated System for Categorizing Transthoracic Echocardiography Indications According to the Echocardiography Appropriate Use Criteria

Aaron S Eisman, Rory B Weiner, Elizabeth S Chen, and 4 more authors

AMIA Annual Symposium Proceedings May 2017

Abs HTML

The Echocardiography Appropriate Use Criteria (EAUC) are a set of indications for transthoracic echocardiography (TTE) developed to guide physician decision making around ordering of TTE. In this study, an automated rule-based method for processing “indications” listed within TTE reports and classification into one of the major EAUC categories was developed and validated against a clinician-annotated reference standard. The system performed at a comparable level to trained physicians allowing for the automated classification of more than 30,000 TTE indications from a public database in less than ten minutes. The most common indication for TTE was Valvular assessment closely followed by General. Hypertension/Heart Failure/Cardiomyopathy, Acute, and Cardiac Structure assessment each contributed more than ten percent within this patient population. These results suggest potential for automated approaches for tracking appropriate use of TTE, as well as guide the development of systems for prospectively identifying when TTE use is recommended.