Expertise

Developing multimodal AI models to improve clinical care, healthcare delivery, and operational efficiency.
Partnering with health technology companies to validate algorithms and support regulatory submissions, including FDA clearances.
Implementing and evaluating AI and digital health tools through quality improvement initiatives and pragmatic randomized controlled trials with rigorous, mixed-methods approaches.

Highlighted Current Projects

AIDA-AI: Amyloidosis Diagnosis Acceleration with AI

Opportunity
Cardiac amyloidosis remains underdiagnosed, delaying treatment and worsening patient outcomes. Digitally-enabled strategies, including flagging echocardiogram reports with natural language processing, automated calculation of amyloidosis risk scores, and the use of FDA-cleared AI echo algorithms, offer an opportunity to enhance early detection and streamline diagnosis.

Approach
As part of a quality improvement project, we will use a mixed methods approach and implementation science methodology to evaluate an AI-powered care pathway program to improve the timely diagnosis of cardiac amyloidosis.

Impact
The anticipated impact of the quality improvement program will include:

Decreased time from suspicion to diagnosis and treatment.
A better understanding of the utility of AI models for the detection of cardiac amyloidosis.
A streamlined clinical trial enrollment screening.

Improving Cardiogenic Shock Outcomes: Using AI for Timely Diagnosis and Management 

Opportunity
Cardiogenic shock is a rapidly evolving and life‑threatening condition where early recognition is essential yet often difficult in busy clinical settings. Clinicians may lack timely tools to identify patients whose conditions are worsening, despite rich clinical data available across the health system. This gap creates missed opportunities for earlier intervention and more consistent, equitable care.

Approach
This project will develop and test AI models to identify patterns of deterioration and risk patterns among hospitalized patients. We will also design and evaluate a clinician‑facing decision support tool through a user-centered process with cardiovascular clinicians and operational partners. The goal is to ensure the AI insights are delivered in a clear, actionable way that fits naturally into clinical workflows.

Impact
This work will generate evidence on how AI-driven decision support can enhance early detection and management of cardiogenic shock. The project aims to improve patient outcomes, support timely clinical decision-making, and establish scalable, equitable approaches for integrating AI into cardiovascular care across the health system.

I-I-CAPTAIN Trial: Implementation and Interaction of Clinician And Patient-facing Tools Aiming to Intensify Neurohormonal medicines for Heart Failure with reduced ejection fraction

Opportunity
Guideline-directed medical therapy (GDMT) for patients with reduced ejection fraction (HFrEF) improves quality of life, reduces hospitalizations, and extends survival. However, busy clinicians treating stable patients with chronic diseases often “leave well enough alone”. This clinical inertia results in underuse of GDMT and frequently does not align with patient health preferences.

Approach
The I-I-CAPTAIN trial will implement and test patient- and clinician-facing nudges for HFrEF medication intensification at five health systems around the country through a randomized, implementation-effectiveness trial.

Impact
The results of this pragmatic trial will answer broad questions related to decision support for evidence-based care, including whether patient-facing or clinician-facing decision support tools are more effective or if the two approaches are additive or synergistic. Ultimately, this trial aims to identify scalable strategies that promote the equitable uptake of evidence-based therapies.

The HeartShare EHR Study

Opportunity
The goal of the HeartShare/AMP Program is to conduct large-scale analysis of phenotypic data, images, and omics from patients with heart failure with preserved ejection fraction and controls to characterize mechanisms of disease and identify therapeutic targets. The Data Portal Core is a component of HeartShare focused on the integration of EHR data from participating clinical centers into a common data model.

Approach
Informaticists from the Data Translation Center and participating clinical centers have curated, mapped, and loaded data from their EHR systems into the Observational Medical Outcomes Partnership (OMOP) Common Data Model. Participating sites will share data with the Data Translation Center to create an individual-level, pooled, limited dataset for analysis.

Impact
The HeartShare EHR Study will generate novel epidemiological insights and supplement the prospective cohort data collected for the HeartShare Deep Phenotyping and HeartShare Registry study. Loading selected clinical notes will also enable multi-institution natural language processing projects.

Using AI to Improve Timely Evaluation for Advanced Heart Failure

Opportunity

Detecting the progression of heart failure (HF) to advanced HF is challenging. Too often, patients are not identified until they have end-organ damage and are too ill to consider life-saving therapies like orthotopic heart transplant or left ventricular assist device placement, missing the optimal window for referral.

Approach

Identify patients at high risk for worsening heart failure and facilitate expedited evaluation by HF specialists across the health system.

Impact

We developed and deployed a multimodal, longitudinal AI model and workflow to detect those with or approaching advanced HF and facilitate evaluation by the HF team. By using a customized MLOps approach, we learned from outside healthcare to improve model development, deployment, optimization of run times, bias and fairness tracking, and real-time monitoring and alerts for model shifts and performance degradation. A key next step is ongoing real-time model evaluation and impact assessment on clinical operations and patient outcomes.

High-Throughput Precision Identification of Cardiac Amyloidosis

Using AI to Improve the Timely Diagnosis and Management of HCM

Opportunity

Despite significant advances in therapeutics for patients with various types of heart failure, substantial gaps remain in evidence-based care, including timely diagnosis, treatment, and monitoring. The delay in implementing evidence-based therapies affects patients with hypertrophic cardiomyopathy (HCM), for whom new treatments are available, such as cardiac myosin inhibitors like mavacamten, along with other guideline-recommended therapies that enhance quality of life and lower the risk. of sudden cardiac death.

Impact

Our objective is to develop and implement solutions that accurately identify care gaps within the total population of patients with HCM and facilitate referrals to the BCVI HCM program.

Approach

Using AI and a digital workflow, we will identify patients across Northwestern Medicine and facilitate their expedited evaluation by an HCM specialist.

Highlighted Past Projects

Barriers and Facilitators to Heart Failure GDMT in an Integrated Health System and FQHCs

Opportunity
Patients with heart failure with reduced ejection fraction do not consistently receive evidence-based therapies that could improve their quality of life and longevity. The purpose of this study was to evaluate multi-level barriers to the provision of optimal guideline-directed medical therapy (GDMT) for patients treated in diverse care settings.

Approach
Clinicians who treat patients with heart failure with reduced ejection fraction were interviewed to identify patient- and clinician-level barriers to achieving target dosing of GDMT. Thematic qualitative analysis was conducted on the interview transcripts.

Impact
This study indicated that determinants of GDMT intensification may vary by clinician specialty and care setting. Future research should explore implementation strategies that address these determinants by specialty and setting.

Read our GDMT barriers outcomes paper.

PROMIS+HF

Opportunity
Patient-reported outcome measures (PROMs) are used in clinical studies and in routine clinical care to capture health status directly from patients. Creating research and clinical PROMs for patients with heart failure based on the extensively tested Patient-Reported Outcomes Measurement Information System (PROMIS®), a set of largely person-centered measures that can be used in the general population and in those with chronic conditions, may facilitate patient-centered care and research and enable comparisons across populations with different chronic conditions.

Impact
The PROMIS+Heart Failure-27 Profile v1.0 (PROMIS+HF-27) and PROMIS+Heart Failure-10 Profile v1.0 (PROMIS+HF-10) instruments were created from the PROMIS+HF long-form Profile measure (PROMIS+HF Profile) are validated measures for patients with HF available at healtmeasures.net and on REDCap for patient-centered research and for clinical care.

Approach
Using state-of-the-art methods from measurement science, we developed and validated research and clinical profiles to capture physical, mental, and social health in patients with HF.

Read our PROMIS+HF outcome paper.

Pharmacist Nudges for Heart Failure Medication Adherence

Opportunity
Heart failure clinics with medication titration protocols led by pharmacists lead to higher quality of care. However, there are no mechanisms in place at NM to track patient adherence to guideline-directed medical therapy (GDMT) following discharge from the medication titration clinic. This pilot will explore the value of an EHR-generated proportion of days covered metric in tracking and notifying pharmacists of adherence drops among their patients with heart failure with reduced ejection fraction.

Approach
We are conducting a pilot at Northwestern Medicine to explore how pharmacists respond to information about patients who have had an adherence drop following discharge from the MAT Clinic.

Impact
This pilot will provide information about the utility of using an EHR-generated measure to track adherence to GDMT following optimization in the MAT Clinic. Additionally, this pilot will generate understanding about how pharmacists react to information about patients who have experienced adherence drops and how to integrate this information into pharmacist workflows.