AI analysis may help doctors spot hidden signs of ATTR-CM
New tool analyzing echocardiogram data outperforms other methods
Written by |
An artificial intelligence (AI) model that analyzes the results of echocardiograms may help predict the likelihood of transthyretin amyloid cardiomyopathy (ATTR-CM) in people at high risk for the condition, a study suggested.
The model, AI-Echo, showed superior diagnostic potential than an AI model based on data from electrocardiograms (ECGs) and a clinical ATTR-CM score, which accounts for six demographic, clinical, and echo-based factors.
If validated, AI-Echo could help with ATTR-CM diagnosis, potentially allowing doctors to avoid unnecessary diagnostic procedures in some cases.
Analyses suggested that, in clinical practice, “a combination of AI-Echo and the ATTR-CM score may be best suited to maximize case detection,” the researchers wrote.
The study, “Diagnostic Performance Evaluation of Clinical and AI Risk Models in Patients Referred for Cardiac Amyloidosis Testing,” was published in the Journal of the American Society of Echocardiography.
Therapies more effective when started early
ATTR-CM is characterized by the toxic buildup of a faulty transthyretin (TTR) protein in the heart. It can be caused by genetic mutations or aging-related processes. Without treatment, the disease can ultimately lead to heart failure.
Disease-modifying therapies targeting the underlying causes of ATTR-CM have become available in recent years. These are typically most effective when patients begin the therapies early in the disease course.
“Given the introduction of novel disease modifying treatment regimens, early accurate diagnosis of [ATTR-CM] is critical for improving outcomes and survival,” the researchers wrote.
Diagnosing ATTR-CM poses several challenges. The process typically involves considering symptoms and examining standard heart tests, including echocardiograms, which assess the heart’s structure and function, and ECGs, which measure the heart’s electrical impulses.
“However, patients may present with nonspecific heart failure symptoms, and findings on electrocardiography (ECG) and echocardiography often reveal only subtle, nonspecific evidence of [ATTR-CM] disease,” the researchers wrote.
People with suspected ATTR-CM are referred for PYP scans or other noninvasive tests that visualize TTR clumps in the heart.
A team of researchers at Mayo Clinic compared the performance of three validated ATTR-CM prediction models by analyzing the medical records of 598 people with heart problems who were referred for PYP or related scans to detect ATTR-CM.
Participants’ median age was 76, 72.6% were men, and 92.9% were white. An ATTR-CM diagnosis was confirmed in 30%.
In addition to AI-Echo, the team assessed AI-ECG, which uses standard electrocardiography, and the Mayo ATTR-CM Score, which incorporates age, sex, high blood pressure, and certain heart disease markers on echocardiogram.
The team used these tools to discriminate between people with ATTR-CM and those without. AI-Echo had a sensitivity of 86%, meaning it correctly identified 86% of the participants who had ATTR-CM. Its specificity was 85%, meaning it correctly ruled out participants who didn’t have ATTR-CM. Overall, the model misclassified about 15% of individuals, either as false positives or false negatives.
False positives identified by AI-Echo tended to have thicker heart wall muscles, an echo feature often associated with ATTR-CM. False negatives also tended to have heart wall muscle thickness that more closely resembled people without ATTR-CM.
AI-ECG and ATTR-CM scores had similar sensitivity to AI-Echo (80% and 86%, respectively), but they had much lower specificity (64% and 69%, respectively), meaning they resulted in more false positives.
“Within a high-risk population for [ATTR-CM], the AI-Echo model demonstrated superior diagnostic discrimination and clinical utility for identification of [ATTR-CM] compared with the AI-ECG model and the ATTR-CM clinical score,” the researchers wrote.
In practice, combining multiple risk assessment strategies could improve ATTR-CM diagnosis. Determining the best combination requires identifying thresholds for which cases require follow-up testing, such as PYP scans.
“When selecting threshold probabilities to guide clinical decision making, the risk of missing a case of [ATTR-CM] must be compared to the benefit of avoiding unnecessary [PYP scans],” the team wrote.
They calculated thresholds they believed would allow doctors to refer most people with actual ATTR-CM for PYP scans while minimizing procedures in people without the condition.
Lower thresholds would mean more people without ATTR-CM would get PYP scans, but fewer people with ATTR-CM would be skipped. At these more conservative thresholds, combining information from AI-Echo and the ATTR-CM clinical score yielded the best results.
Higher thresholds would allow for a higher risk of missing cases but a lower risk of unnecessary PYP scans. The team found that with slightly more lenient thresholds, AI-Echo risk prediction alone was most effective at guiding decision-making, avoiding more unnecessary PYP scans than the other models.
Before incorporating AI-Echo analysis into routine clinical practice, further studies should validate the metric and help refine the thresholds for PYP referrals, the researchers wrote.
“As with any predictive model, translation to [groups] with distinctly different [initial] risk may require model calibration to ensure that predicted probabilities reflect observed risk,” the team concluded.
Leave a comment
Fill in the required fields to post. Your email address will not be published.