Deep Learning Model for Cardiac Amyloidosis

Recruiting in Palo Alto (17 mi)

Overseen by Timothy J. Poterucha, MD

Age: 18+

Sex: Any

Travel: May Be Covered

Time Reimbursement: Varies

Trial Phase: Academic

Recruiting

Sponsor: Timothy Poterucha

Disqualifiers: Primary amyloidosis, Liver transplant, Active malignancy, others

No Placebo Group

Trial Summary

What is the purpose of this trial?

This is a single center, diagnostic clinical trial in which the investigators aim to prospectively validate a deep learning model that identifies patients with features suggestive of cardiac amyloidosis, including transthyretin cardiac amyloidosis (ATTR-CA). Cardiac Amyloidosis is an age-related infiltrative cardiomyopathy that causes heart failure and death that is frequently unrecognized and underdiagnosed. The investigators have developed a deep learning model that identifies patients with features of ATTR-CA and other types of cardiac amyloidosis using echocardiographic, ECG, and clinical factors. By applying this model to the population served by NewYork-Presbyterian Hospital, the investigators will identify a list of patients at highest predicted risk for having undiagnosed cardiac amyloidosis. The investigators will then invite these patients for further testing to diagnose cardiac amyloidosis. The rate of cardiac amyloidosis diagnosis of patients in this study will be compared to rate of cardiac amyloidosis diagnosis in historic controls from the following two groups: (1) patients referred for clinical cardiac amyloidosis testing at NewYork-Prebysterian Hospital and (2) patients enrolled in the Screening for Cardiac Amyloidosis With Nuclear Imaging in Minority Populations (SCAN-MP) study.

Will I have to stop taking my current medications?

The trial information does not specify whether you need to stop taking your current medications. It is best to discuss this with the trial coordinators or your doctor.

What data supports the effectiveness of the treatment Cardiac amyloidosis deep learning model?

Research shows that deep learning models, like convolutional neural networks, can classify different types of cardiac amyloidosis with higher accuracy than human experts, suggesting they could help in diagnosing this condition more effectively. ^{1 2 3 4 5}

Is the deep learning model for cardiac amyloidosis safe for humans?

The research articles focus on the effectiveness of machine learning models in diagnosing cardiac amyloidosis, but they do not provide specific safety data for humans regarding the use of these models. ^{1 2 3 6 7}

How does the deep learning model for cardiac amyloidosis differ from other treatments?

The deep learning model for cardiac amyloidosis is unique because it uses advanced machine learning algorithms to analyze heart function and structure, potentially improving diagnosis and disease progression prediction compared to traditional methods. This approach focuses on using data from cardiac imaging to identify patterns and markers that are not easily detectable by human analysis, offering a novel way to manage the condition. ^{1 2 4 5 8}

Research Team

Timothy J. Poterucha, MD

Principal Investigator

Assistant Professor of Medicine

Eligibility Criteria

This trial is for individuals who may have cardiac amyloidosis, a heart condition that can lead to heart failure. It's aimed at those who haven't been diagnosed yet but are suspected of having the disease based on certain heart tests and clinical factors.

Inclusion Criteria

I am 50 years old or older.

I can understand and sign the informed consent.

Electronically stored ECG and echocardiogram within 5 years of study start date

Exclusion Criteria

I have been tested for cardiac amyloidosis.

I do not have any conditions like a stroke that would stop me from joining the study.

I have a life expectancy of more than 1 year despite my cancer.

Trial Timeline

Screening

Participants are screened for eligibility to participate in the trial

2-4 weeks

Diagnostic Testing

Patients identified by the deep learning model are invited for further testing to diagnose cardiac amyloidosis

Up to 1 year

Multiple visits as needed for diagnostic testing

Follow-up

Participants are monitored for safety and effectiveness after diagnostic testing

4 weeks

Treatment Details

Interventions

• Cardiac amyloidosis deep learning model (Deep Learning Model)

Trial Overview The study is testing a deep learning model designed to spot signs of cardiac amyloidosis using data from echocardiograms, ECGs, and patient history. Participants identified as high-risk by this model will be invited for further tests to confirm diagnosis.

Participant Groups

Group I: Intervention Arm Experimental Treatment 1 Intervention

Patients who are identified by the deep learning model as being at high risk for undiagnosed cardiac amyloidosis who are enrolled in the study.

Who Is Running the Clinical Trial?

Timothy Poterucha

Lead Sponsor

Trials

1

Recruited

100+

Pierre Elias

Lead Sponsor

Trials

1

Recruited

100+

Pfizer

Industry Sponsor

Trials

4,712

Recruited

50,980,000+

Known For

Vaccine Innovations

Top Products

Viagra, Zoloft, Lipitor, Prevnar 13

Albert Bourla

Pfizer

Chief Executive Officer since 2019

PhD in Biotechnology of Reproduction, Aristotle University of Thessaloniki

Patrizia Cavazzoni

Pfizer

Chief Medical Officer

MD from McGill University

Eidos Therapeutics, a BridgeBio company

Industry Sponsor

Trials

12

Recruited

2,400+

American Heart Association

Collaborator

Trials

352

Recruited

6,196,000+

Eduardo Sanchez

American Heart Association

Chief Medical Officer since 2013

MD from University of Texas Southwestern Medical School, MPH from UT Health Science Center at Houston, MS in Biomedical Engineering from Duke University

Katrina McGhee

American Heart Association

Chief Executive Officer since 2020

MBA from the University of Texas at Arlington

Findings from Research

A convolutional neural network (CNN) trained on cine-MR images achieved a classification accuracy of 75% in distinguishing between AL and ATTR amyloidosis, outperforming human readers who had an accuracy range of 61.7% to 67.5%.

The cine-CNN also demonstrated a higher area under the ROC curve (AUC) of 0.839 compared to 0.679 for gado-CNN and 0.714 for the best human reader, indicating its potential as a more effective tool for diagnosis, although it is still not optimal for clinical use.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Deep Learning to Classify AL versus ATTR Cardiac Amyloidosis MR Images. Germain, P., Vardazaryan, A., Labani, A., et al.[2023]

A machine learning algorithm, originally trained on heart failure data from the USA, effectively identified patients with undiagnosed wild type cardiac amyloidosis (ATTRwt) in the UK, achieving an area under the receiver operating curve (AUROC) of 0.84 for definitive cases and 0.86 for possible cases when using combined data from primary and secondary care.

The algorithm's performance was notably lower when using only primary or secondary care data alone (AUROC ranging from 0.68 to 0.78), suggesting that integrating data from both sources enhances its predictive ability, which could lead to earlier diagnosis and better patient outcomes.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Detecting transthyretin amyloid cardiomyopathy (ATTR-CM) using machine learning: an evaluation of the performance of an algorithm in a UK setting. Tsang, C., Huda, A., Norman, M., et al.[2023]

A study involving 138 patients (74 with cardiac amyloidosis and 64 with hypertrophic cardiomyopathy) demonstrated that machine learning models, particularly random forest and gradient boosting, can effectively differentiate between these two conditions with high accuracy (AUC up to 0.98).

The use of machine learning combined with speckle tracking echocardiography shows promise in improving the timely diagnosis of cardiac amyloidosis, which is often misdiagnosed, potentially leading to better patient outcomes.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Machine learning algorithms to automate differentiating cardiac amyloidosis from hypertrophic cardiomyopathy. Wu, ZW., Zheng, JL., Kuang, L., et al.[2023]

This proof of concept study suggests that new biomarkers derived from mathematical models of the left ventricle can effectively predict the progression of cardiac amyloidosis, based on routine clinical MRI scans.

The study found that a combination of mechanical, geometrical, and shape features is necessary for accurate predictions, highlighting the complexity of heart disease progression; however, results should be interpreted cautiously due to the small sample size of only seven cases.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Analysis of Cardiac Amyloidosis Progression Using Model-Based Markers. Li, W., Lazarus, A., Gao, H., et al.[2021]

A convolutional neural network (CNN) demonstrated superior accuracy in diagnosing cardiac amyloidosis from cine-CMR images compared to experienced human operators, achieving an accuracy of 82.5% versus 66.0% for patients.

The CNN also showed a higher area under the curve (AUC) of 0.895 compared to 0.727 for human analysis, indicating its enhanced ability to distinguish cardiac amyloidosis from left ventricular hypertrophy of other origins.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Deep Learning Supplants Visual Analysis by Experienced Operators for the Diagnosis of Cardiac Amyloidosis by Cine-CMR. Germain, P., Vardazaryan, A., Padoy, N., et al.[2022]

The study developed a statistical learning algorithm that can help detect cardiac amyloidosis, a rare and often misdiagnosed condition, by analyzing clinical records from hospitalizations.

Despite the low prevalence of cardiac amyloidosis, the algorithm demonstrated the ability to identify the disease effectively, suggesting it could be a valuable tool for screening patients with heart failure.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Real-World Data and Machine Learning to Predict Cardiac Amyloidosis. García-García, E., González-Romero, GM., Martín-Pérez, EM., et al.[2021]

The study utilized machine learning algorithms to analyze cardiac strain and function data from 43 patients with cardiac amyloidosis, achieving a high diagnostic accuracy of 90.9% using the SVM RBF kernel, indicating its potential for improving CA diagnostics.

Key predictors for cardiac amyloidosis were identified, including bi-atrial longitudinal strain and left and right atrial ejection fraction, suggesting that these parameters could enhance non-invasive diagnostic approaches in clinical settings.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

A Machine Learning Challenge: Detection of Cardiac Amyloidosis Based on Bi-Atrial and Right Ventricular Strain and Cardiac Function. Eckstein, J., Moghadasi, N., Körperich, H., et al.[2022]

References

1. Switzerland pubmed.ncbi.nlm.nih.gov

Deep Learning to Classify AL versus ATTR Cardiac Amyloidosis MR Images. [2023]

2. England pubmed.ncbi.nlm.nih.gov

Detecting transthyretin amyloid cardiomyopathy (ATTR-CM) using machine learning: an evaluation of the performance of an algorithm in a UK setting. [2023]

3. United States pubmed.ncbi.nlm.nih.gov

Machine learning algorithms to automate differentiating cardiac amyloidosis from hypertrophic cardiomyopathy. [2023]

4. Switzerland pubmed.ncbi.nlm.nih.gov

Analysis of Cardiac Amyloidosis Progression Using Model-Based Markers. [2021]

5. Switzerland pubmed.ncbi.nlm.nih.gov

Deep Learning Supplants Visual Analysis by Experienced Operators for the Diagnosis of Cardiac Amyloidosis by Cine-CMR. [2022]

6. Switzerland pubmed.ncbi.nlm.nih.gov

Real-World Data and Machine Learning to Predict Cardiac Amyloidosis. [2021]

7. England pubmed.ncbi.nlm.nih.gov

A machine learning model for identifying patients at risk for wild-type transthyretin amyloid cardiomyopathy. [2022]

8. Switzerland pubmed.ncbi.nlm.nih.gov

A Machine Learning Challenge: Detection of Cardiac Amyloidosis Based on Bi-Atrial and Right Ventricular Strain and Cardiac Function. [2022]