Project Description Narrative:
Cardiologists and emergency physicians are tasked with deciding whether to activate the catheterization lab for cardiac angiography and possible percutaneous coronary intervention (PCI). They rely on a myriad of factors including cardiac troponin measurements, patient symptoms, serial ECG changes, etc. for making that decision. An important consideration is the automatic computerized interpretation (ACI) as provided by most ECG devices on the market today. Physicians are more likely to provide a correct ECG interpretation if given a correct ACI with a very impressive odds ratio (OR) of 10.87. However, incorrect ACIs are frequently not corrected by overreading cardiologists. In one study, primary cardiologists (those ordering the ECG that provided the initial interpretation) corrected incorrect atrial fibrillation ACI for 94% while overreading cardiologists (those without direct patient contact) corrected only 78%.
AMI criteria used in Hill-Rom’s VERITAS (and most modern commercial programs) were developed decades ago based on clinically accepted criteria, with additional rules to increase sensitivity/specificity, yet have shown relatively little evolution over so many years. Current ACI algorithms cannot replace physician reading of ECGs, and do not claim such an ability. Indeed, a recent study of patients with normal ACI found that delaying over-reader review should not be recommended. Therefore, an overreading process is typically used for ECGs; even so, it still takes 24-48 hours for many over-reader interpretations in the routine workflow. In emergency situations, consultation of a cardiologist/electrophysiologist is often required; such interruptions take physicians away from their patients, increasing clinical errors by 12%. Meanwhile, cardiologists and computerized interpretation programs miss many occlusion/Type I AMIs (T1AMI as opposed to non-occluded/Type II AMI, T2AMI). In a study of STEMI, cardiologists were found to be only 49% sensitive to T1AMI whereas a computer program was only 21% sensitive. A recent review of physician interpretation accuracy studies found highly variable results with a median post-training accuracy of 64% overall that improves to 78% for cardiologists.
For AMI, it is a trying situation: the most common mortal disease, where mere minutes can be the difference between life and death, as physicians and algorithms apply decades-old/questionable criteria fraught with perilous, potentially fatal, errors of misdiagnosis due to improper training and/or the influence of inaccurate algorithms. This project’s research team aims to improve the accuracy of ACI for an AMI. Their goal is to improve the classical STEMI criteria used by VERITAS. As a result of their work, they aim to develop a high-quality curated database using multiple sources to adjudicate the ECG annotations. Such a database will be beneficial for incrementally improving AMI algorithms in the future and will enable new research beyond the scope of this project. Additionally, clinical practice may benefit from application of revised criteria for ACI algorithms.