Heart disease has been the number one killer in the U.S. every year since the 1920s, and accounted for almost one in three deaths in 2019 (874,613 lives lost). Frustratingly, at least 200,000 deaths from heart disease and stroke could be avoided each year. The solution starts with correctly identifying patients with heart disease and then ensuring they are treated appropriately, in a timely manner. Assisted by the latest data science tools, such as artificial intelligence (AI), we can leverage the massive amount of healthcare data being generated to help solve this problem. However, before we talk about solutions, let's dive deeper into the problem.
First, we need to know who has heart disease. Millions of Americans, and many more people globally, suffer from treatable heart disease, yet many remain unaware of their condition and live "undiagnosed" for years. Without knowing who actually has heart disease, patients cannot be treated to avoid potentially devastating consequences. For example, atrial fibrillation (Afib) is an abnormal heart rhythm that impacts 5.3 million Americans. It substantially increases the risk of stroke; in fact, nearly one-quarter of all strokes of unclear origin result in a new diagnosis of Afib, and these strokes usually result in permanent debilitation. Two-thirds of strokes related to Afib are preventable if doctors and patients are aware of the diagnosis; unfortunately an estimated 700,000 Americans suffer from Afib, yet remain unaware that they have it.
Second, even when we know patients have heart disease, a large proportion do not receive guideline-directed medical therapies that are proven to prolong life and reduce suffering. Among thousands of patients with heart failure in a recent study, three out of four patients were missing at least one of eight indicated, guideline-recommended therapies. Undertreatment is a problem we have faced in the U.S. for decades, and it is not unique to cardiology. A 2003 study showed that only half of guideline-recommended medical care was being delivered to U.S. adults, across a variety of acute and chronic conditions, including preventative care. The combination of both undertreated and undiagnosed heart disease leads to a large burden of likely avoidable, debilitating outcomes such as stroke and premature death.
How can such a large burden of heart disease go undiagnosed or untreated when the U.S. has access to some of the most cutting-edge treatments and healthcare technology? The problem -- and therefore the solution -- is multifaceted and highly complex. A recent study sheds some light on one of the most significant obstacles: the massive amounts of data and guidelines clinicians must sift through to effectively treat patients. Researchers have estimated that primary care physicians need 26.7 hours per 24 hour day to keep up with and provide essential guideline-directed medical care to patients. This impossible task leads to immense frustration and burnout for physicians, while patients are left carrying the burden of undiagnosed or undertreated disease.
More specifically, a cardiologist has under 10 minutes on average to interpret an echocardiogram (an ultrasound or imaging study of the heart). An echocardiogram consists of roughly 100 separate videos, thus the cardiologist has only a few seconds to scan through each video in pursuit of disease diagnosis. This leaves little to no time for reading through a patient's medical history and record, which often consists of hundreds of clinical notes, labs, and complicated test reports. Important clinical information that may help with accurate interpretation of the echocardiogram, such as family history or recent abnormal lab results, is often buried.
Data science and AI, which have already shown promise in cardiology, can help solve this problem. A study published in Nature Biomedical Engineering showed that a neural network, a type of AI, can help improve doctors' ability to predict clinical outcomes from echocardiograms. Neural networks are the same type of AI method that allows facial or object recognition in photos taken by our smartphones, and these approaches are now showing promise to help find undiagnosed heart disease. Multiple studies, such as research in The Lancet and a study we co-authored in Circulation, have demonstrated the ability to identify patients with a high risk of undiagnosed Afib by using AI to analyze their 12-lead electrocardiogram (ECG). In the Circulation study, over half of the patients who presented with a stroke as their initial manifestation of Afib would have had a "high risk" result prior to the stroke had their ECG been analyzed. This gives us hope that integrating these algorithms into clinical practice could help us treat patients sooner to avoid stroke. Afib is just the tip of the iceberg -- new studies are being published regularly that use data science and AI to target various heart and cardiovascular diseases including structural heart diseases, abnormal electrical activity in the heart, pulmonary hypertension, and even approaches to reduce the undertreatment of atrial fibrillation and heart failure.
With mounting evidence highlighting the promise of new technologies, how do we ensure they are implemented clinically to minimize the devastating consequences of undiagnosed and undertreated heart disease? While the steps are complex, we can learn from the playbook of other medical specialists, like oncologists treating cancer, the second leading cause of death for Americans. Oncologists have broadly adopted data-driven precision medicine; they expect a combination of multimodal molecular, clinical, and imaging data to inform not only which therapies they offer to patients with cancer, but also to help match their patients to clinical research trials to bring the next generation of novel cancer therapies to market.
Cardiology is now in a position to leverage multimodal clinical data to its fullest potential -- clinicians can soon expect a patient's ECG or echocardiogram to be interpreted with assistance from AI algorithms, within the context of all available imaging, laboratory, and genomic data. This will ensure not only the most accurate diagnoses but also optimize the important next steps to discuss with patients, including potential therapies, additional studies to evaluate for risk of undiagnosed heart disease, and relevant clinical trials. It is time to push beyond the status quo of the last several decades and step forward into the data-driven precision medicine of the future. If we don't, the consequences of undiagnosed and undertreated heart disease will continue to devastate millions of people.
Brandon Fornwalt, MD, PhD, is SVP of Cardiology at Tempus, an artificial intelligence and precision medicine company. John Pfeifer, MD, MPH, is VP of Clinical Cardiology at Tempus.