Myocardial infarction is one of the leading causes of death around the world and its early diagnosis is essential to ensure timely and effective treatment. However, existing diagnostic pathways have several limitations that could benefit from recent technological advancements in machine learning. A new study published in Nature Medicine proposed the use of machine learning-based models that integrate cardiac troponin concentrations and clinical features, in order to estimate the probability of a myocardial infarction for patients. This could mean a reduction in inequalities in care, the prevention of unnecessary admissions as well as a faster diagnosis and treatment of affected patients.
Clinical decision support featuring these novel models could potentially address the current challenges in diagnosis. For instance, the fixed troponin thresholds that are commonly used do not take into account age, sex, or comorbidities, and thus do not ensure an equitable diagnosis. Secondly, the current pathways are not efficient, as most patients require serial troponin measurements. Lastly, electrocardiogram findings or the time of symptom onset are not systematically taken into account.
The study was led by Professor Elizabeth Varker, the Director of Cardiology Research at the George Institute for Global Health and Professor of Medicine at the University of Sydney, Australia. Her research has focused on designing and evaluating clinical trials to assess and improve the management of acute coronary syndrome. Professor Varker has worked extensively to develop and implement evidence-based strategies such as risk stratified clinical decision support using machine learning approaches that optimize the diagnosis of myocardial infarction.
These findings not only present an opportunity to enhance the diagnosis of myocardial infarction, but could also potentially reduce healthcare disparities by allowing for more precise and more equitable patient assessment. Furthermore, implementation of decision support tools through machine learning models could potentially improve patient care by reducing diagnostic time and preventing unnecessary admissions.
