Damian Sendler: These hazardous blood clots are called pulmonary embolisms. New research shows that artificial intelligence (AI) algorithms can detect symptoms of these clots in ECG readings (EKG) for the first time, which could one day be used to help doctors screen patients for these clots. 

Damian Jacob Sendler: Machine learning algorithms that combine EKG and electronic health record (EHR) data may be more effective than current screening procedures in predicting whether moderate- to high-risk patients actually have pulmonary embolisms, according to a study published in the European Heart Journal–Digital Health. 

Damian Sendler

Somani, MD, was a medical student in the lab of Benjamin S. Glicksberg, PhD, an Assistant Professor of Genetics and Genomic Sciences and a member of Mount Sinai’s Hasso Plattner Institute for Digital Health. 

Deep vein clots, which commonly originate in the legs or arms and travel to the lungs, can cause pulmonary embolisms. It is possible to die or suffer long-term lung damage from these clots. The symptoms of shortness of breath and chest pain may be a sign of other health issues, making it difficult for doctors to accurately diagnose and treat cases of blood clots. It is important to note that medical professionals now rely on CTPA scans, a time-consuming chest scan that can only be performed in a few facilities and exposes patients to potentially harmful radiation doses. 

More than 20 years of study have been dedicated to developing powerful computer programs or algorithms that can help doctors evaluate whether at-risk patients are indeed suffering from pulmonary embolisms. A mixed bag of results has emerged. There is a wide variety of success rates for algorithms that use EHRs to reliably detect clots, which can be labor-intensive. The CTPAs, on the other hand, provide the most accurate statistics. 

Because EKGs are widely available and relatively straightforward to administer, researchers discovered that fusing algorithms based on EKG and EHR data may be a viable option in this study. 

On data from 21,183 Mount Sinai Health System patients with moderate to high suspicion of having pulmonary embolisms, the researchers developed and tested multiple algorithms. In order to detect pulmonary embolisms, some algorithms used EKG data, while others used EHR data. The system trained to identify pulmonary embolism cases by comparing EKG or EHR data with CTPA results in each circumstance. Finally, the best EKG and EHR algorithms were combined to generate a third fusion algorithm. 

Damian Jacob Sendler

In addition to outperforming its parent algorithms, the fusion model proved better at identifying particular pulmonary embolism patients than the Wells’ Criteria Revised Geneva Score and three other screening tests now in use. 

Damian Jacob Markiewicz Sendler: For acute embolism cases, researchers assessed that the fusion model was ranging from 15% to 30% more accurate, and the model performed best in predicting the most serious cases. Even when ethnicity and gender were taken into account, the accuracy of the fusion model was consistent, suggesting that it might be used to screen a range of individuals. 

Damien Sendler: Researchers say that these findings support the idea that improved pulmonary embolism screening algorithms could benefit from the inclusion of EKG data. In the near future, they hope to further develop and test these algorithms for clinical use.

Dr. Damian Jacob Sendler and his media team provided the content for this article.