Large Language Models for Cardiomyopathy

Overseen By Jack W O'Sullivan, MBBS, DPhil

Age: 18+

Sex: Any

Trial Phase: Academic

Sponsor: Stanford University

Disqualifiers: Not practicing clinically

No Placebo Group All trial participants will receive the active study treatment (no placebo)

Compensation: Varies

Number of Visits: Unknown

12 Participants Needed

What You Need to Know Before You Apply

What is the purpose of this trial?

This trial aims to compare large language models (LLMs), a type of AI, with traditional tools in assisting cardiologists with decisions about treating heart conditions. Cardiologists will review real patient cases from a clinic specializing in genetic heart diseases and will use either AI support or their usual resources to make decisions. Heart specialists will evaluate these decisions to determine which method is more effective. Board-certified or board-eligible cardiologists who are currently practicing are the right fit for this study. As an unphased trial, this study offers a unique opportunity to contribute to innovative research that could transform decision-making in cardiology.

Will I have to stop taking my current medications?

The trial information does not specify whether participants must stop taking their current medications.

What prior data suggests that this method is safe for clinical decision-making in cardiology?

Research has shown that large language models (LLMs) can assist doctors in making heart health decisions. One study found that LLMs, such as GPT-4, achieved safety scores ranging from 39% to 86%, indicating that they often provided safe advice to doctors.

However, LLMs are not perfect. Their accuracy and safety can vary, so doctors should use them as a helpful tool but not rely solely on them. In summary, LLMs hold promise for supporting medical decisions, but they remain a developing technology. ^{1 2 3 4 5}

Why are researchers excited about this trial?

Researchers are excited about using Large Language Models (LLMs) for cardiomyopathy because these AI systems can analyze vast amounts of medical data quickly and provide personalized insights. Unlike traditional treatments that focus on medications or lifestyle changes, LLMs can help doctors by offering real-time recommendations and deeper understanding based on the latest research. This approach could potentially enhance decision-making and improve patient outcomes by integrating cutting-edge technology with existing healthcare practices.

What evidence suggests that this trial's treatments could be effective for cardiomyopathy?

Research has shown that large language models (LLMs) could enhance heart health care. In this trial, one group will access a Large Language Model to explore its potential benefits. Studies have found that these models accurately identify causes of heart disease and recognize signs of heart failure. They also offer a cost-effective method for selecting participants for heart failure trials. Additionally, LLMs can predict heart failure risk from ECG data, which is crucial for early treatment. While this technology is still developing, early results suggest that LLMs could make diagnosing and managing heart conditions more precise and efficient. ^{3 6 7 8 9}

Who Is on the Research Team?

Euan A Ashley, BSc, MB ChB, DPhil

Principal Investigator

Stanford University

Are You a Good Fit for This Trial?

This trial is for board-certified or board-eligible cardiologists who are currently practicing clinically. It's not open to those who aren't actively seeing patients.

Inclusion Criteria

Board certified or board eligible Cardiologist

Exclusion Criteria

Not currently practicing clinically

Timeline for a Trial Participant

Screening

Participants are screened for eligibility to participate in the trial

2-4 weeks

Assessment

General cardiologists are randomized to manage real patient cases with or without AI assistance

1 month

Multiple assessments per participant

Evaluation

Subspecialty experts evaluate the cardiologists' responses using a standardized rubric

1 month

Follow-up

Participants are monitored for feedback on the use of the Large Language Model

1 hour

What Are the Treatments Tested in This Trial?

Interventions

• Large Language Model

Trial Overview The study is testing whether large language models (AI tools) can help general cardiologists make better decisions in managing real patient cases of genetic heart diseases, compared to using traditional decision support tools.

How Is the Trial Designed?

Group I: Large Language Model Active Control 1 Intervention

This group will be given access to a Large Language Model

Group II: Usual resources Active Control 1 Intervention

Group will not be given access to a Large Language Model but will be encouraged to use any resources they usually use in their practice besides large language models (UpToDate, Dynamed etc).

Who Is Running the Clinical Trial?

Stanford University

Lead Sponsor

Trials

2,527

Recruited

17,430,000+

Google LLC.

Industry Sponsor

Trials

12

Recruited

31,700+

Published Research Related to This Trial

Dilated Cardiomyopathy (DCM) patients often face significant risks despite treatment, indicating that traditional measures like left ventricular ejection fraction are insufficient for predicting outcomes, highlighting the need for better disease characterization.

Cardiac MRI (CMR) combined with advanced artificial intelligence techniques, particularly deep learning, can enhance the assessment of DCM by providing detailed functional and tissue characterization, which may improve diagnosis and risk stratification for patients.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

The Role of AI in Characterizing the DCM Phenotype. Asher, C., Puyol-Antón, E., Rizvi, M., et al.[2023]

A comprehensive analysis of 3,290 approved drugs and over 1.6 million reported adverse events revealed that animal studies can effectively predict human safety outcomes, particularly for conditions like QT prolongation and arrhythmias observed in dogs.

Despite the overall predictivity of animal safety data for humans, the study highlighted challenges in data curation and changes in methodologies over time that can affect the reliability of these analyses.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

A big data approach to the concordance of the toxicity of pharmaceuticals in animals and humans. Clark, M., Steger-Hartmann, T.[2018]

Citations

1. pmc.ncbi.nlm.nih.gov pmc.ncbi.nlm.nih.gov/articles/PMC12282349/

Applications of large language models in cardiovascular ... Large language models were primarily evaluated on common CV conditions, such as hypertension, diabetes, heart failure, and atrial fibrillation, whereas their ...

2. sciencedirect.com sciencedirect.com/science/article/pii/S0010482525009606

Assessing large language models for acute heart failure ... In the extraction of heart failure characteristics, DrLongformer demonstrated superior performance in identifying the cause of heart disease, with an average F1 ...

3. 2minutemedicine.com 2minutemedicine.com/a-large-language-model-demonstrates-superior-cost-effectiveness-in-screening-heart-failure-candidates-for-clinical-trials/

A large language model demonstrates superior cost ... A large language model demonstrates superior cost-effectiveness in screening heart failure candidates for clinical trials. byCheng En Xiand ...

4. ahajournals.org ahajournals.org/doi/10.1161/CIRCHEARTFAILURE.124.012514

Natural Language Processing to Adjudicate Heart Failure ... We developed a novel model for automated AI-based heart failure adjudication (Heart Failure Natural Language Processing) using hospitalizations ...

5. nature.com nature.com/articles/s44387-025-00011-z

Large language models for disease diagnosis: a scoping ... Large Language Model-Informed ECG Dual Attention Network for Heart Failure Risk Prediction. IEEE Transactions on Big Data 11, 948–960 (2024).

6. sciencedirect.com sciencedirect.com/science/article/pii/S0735675725002311

Automated computation of the HEART score with the GPT ... In this study, we evaluated the performance of the GPT-4 large language model in calculating HEART scores and predicting major adverse cardiac events (MACE) at ...

7. thelancet.com thelancet.com/journals/landig/article/PIIS2589-7500(24)00151-1/fulltext

The potential for large language models to transform ... This Series paper explores opportunities and limitations of AI models for cardiovascular medicine, and aims to identify specific barriers to and solutions.

8. nature.com nature.com/articles/s41591-024-03445-1

Medical large language models are vulnerable to data- ... We perform a threat assessment that simulates a data-poisoning attack against The Pile, a popular dataset used for LLM development.

9. ahajournals.org ahajournals.org/doi/10.1161/HYPERTENSIONAHA.125.25492

Performance of Large Language Models in Analyzing ... The rate of accurate and safe response varied between 35% to 83% and 39% to 86%, respectively. GPT-4 had the highest accuracy and safety scores ...