Healthcare Delivery Science Research Projects

Principal Investigator, Steve Horng

Heart failure represents a significant and leading health problem in the United States and is one of the leading causes of hospitalization in the developed world, accounting for nearly 1 million hospitalizations annually and nearly $70 billion in annual costs. Despite its prevalence and importance, heart failure hospitalization is difficult to predict using existing metrics. Use of echocardiographic data may improve prediction of heart failure, but current prediction model performance is insufficient for clinical use. Current advances in machine learning, specifically in deep learning, have great potential to revolutionize heart failure prediction, but is limited by the lack of large publicly available data sets.

This project will create an echocardiogram (MIMIC-ECHO) and electrocardiogram (MIMIC-ECG) extension to the MIMIC-IV database. The Medical Information Mart for Intensive Care (MIMIC) database was launched in 2003 as a freely-available database is now a widely used resource with over 12,000 credentialed users in academia and industry. The MIMIC-IV database includes newer data and features a modular structure that facilitates the linking of multiple data types. This project will add approximately ~145,000 transthoracic echocardiograms (50 terabytes) and ~980,000 ECGs (70 gigabytes) along with their associated reports, fully integrated with all of the clinical, waveform, laboratory data already found in the MIMIC-IV database.

This work builds on the long-standing collaboration between BIDMC and MIT responsible for the MIMIC database, and in partnership with Philips Healthcare, a leading manufacturer of both echocardiogram and electrocardiogram equipment. This project is being funded by the Mass Life Sciences Center. The MIMIC-extensions generated from this work will be piloted in early release through the MIT Critical Data Consortium, then publicly release through the PhysioNet, similar to the release of previous versions and extensions of MIMIC. This project is funded by the Massachusetts Life Sciences Center through its 2020 Bits to Bytes Program.

Site Co-Principal Investigator, Jennifer P. Stevens

Studies have shown that low tidal volume ventilation (LTVV) is one of the most effective therapies for treating acute respiratory distress syndrome (ARDS), but its use remains low. This multi-site project, led by Dr. Curtis Weiss of NorthShore University HealthSystem, sought to identify the facilitators and barriers to LTVV adoption and the differences that exist between academic and community settings. The goal of the study was to advance implementation science by providing a model for how data science and network science can be applied to understand the adoption of complex interventions. This project was conducted across an international consortium of medical centers and was funded by the National Institutes of Health.

Principal Investigator, Jennifer P. Stevens

Hospitalized patients experience a wide range of preventable harms despite significant efforts to provide safe and effective healthcare. This project seeks to identify and predict stressed clinical environments -- or "risky states" that predispose patients to preventable harms - initially in the intensive care unit, and then in the emergency department. This approach goes beyond the clinical specifics of any individual patient and any category of harm. It represents a departure from the historical "silo" approach of hospitals addressing specific harm categories, such as falls or infections. This project was funded as a Career Development Award by the Doris Duke Charitable Foundation. Learn more.

Principal Investigator, Jennifer P. Stevens

Inpatient consultation is the primary mechanism by which specialists provide care to hospitalized patients. For example, more than 90% of Medicare patients have at least one consultation during an inpatient admission. Previous research has demonstrated that the scope of variation in inpatient consultation use across U.S. hospitals is substantial. This project sought to advance knowledge of inpatient consultation in three ways: by defining the characteristics of beneficial consultations from multiple perspectives, including those of patients and families; by investigating novel non-clinical factors that drive consultation for reasons other than patient or family need; and by quantifying the relationship between inpatient consultations and patient outcomes and costs. This project was funded by the Agency for Healthcare Research and Quality. This work was preceded by research focused on consult quality in the critical care setting, funded by the Gordon and Betty Moore Foundation. Learn more.

COVID-19 has created catastrophic direct health harms, but a method known as excess death analysis has also shown tens of thousands of increased deaths during the pandemic not directly associated with the virus. These deaths are believed to be the result of delays in care for other conditions, which may worsen over time as delays in preventive care may not manifest health harms for many months or even years. Deferral of routine care may also be profoundly inequitable since more-advantaged patients are more likely to have access to certain interventions. This project sought to quantify the impact of COVID-19 on non-COVID-associated healthcare by analyzing national databases using econometric methods, assessing the health equity impact of delays in non-COVID-associated healthcare, and exploring the relationship of regional patterns with subsequent delays in non-COVID-associated healthcare. This project was led by Dr. Jennifer P. Stevens, Director of the Center for Healthcare Delivery Science at Beth Israel Deaconess Medical Center, and was funded by Google.org, Google’s Foundation.

Learn More

• Hospitalizations for Emergent Medical, Surgical, and Obstetric Conditions in Boston During the COVID-19 Pandemic J Gen Intern Med (July 22, 2020)
• Preventability of 30-day Hospital Revisits Following Admission with COVID-19 at an Academic Medical Center Jt Comm J Qual Patient Saf (August 24, 2021)

2025

• SAFEMED2 - Optimizing Opioid Prescribing Through Multitask Modeling of Patient Outcomes at Hospital Discharge
  Principal Investigator, Shani Herzig
• Large Language Model (LLM)-generated Plain Language Addendum to Notes (LLM-PLANs)
  Principal Investigators, Timmy Ho and Kristyn Beam 

2024

• Improving Language-Concordant Care for Limited English Proficiency (LEP) Patients in the Emergency Department Using Large Language Models (LLMs)
  Principal Investigators, Adam Haimovich and Jossie Carreras-Tartak 

2023

• Assessing the Implementation of the Healthcare Racial Justice Assessment Tool (HC-RJAT): A Novel Tool to Measure Structural and Institutional Racism in the Healthcare Setting
  Principal Investigator, Yarden Fraiman

2022

• Impact of Cultural Competency in Outpatient Surgical Care: Implementation of a Sustainable Latinx Colorectal Surgery Clinic
  Principal Investigators, Ana Sofia Ore and Evangelos Messaris

2021

• Improving Care Disparities in Dementia Diagnosis and Treatment
  Principal Investigator, Alexandra Stillman
• Tele-Brain Interprofessional Consultation (TBIC)
  Principal Investigator, Martina Stippler

2024

• Deep Learning to Identify Echocardiographic Abnormalities in Using Electrocardiograms
  Principal Investigator, Jordan Strom
• Machine Learning to Refine Heart Failure Endpoints
  Principal Investigator, Jennifer Ho