On a recent overnight shift in the emergency room, a woman who was having vague abdominal pain and chest discomfort for several days was referred to me. When her symptoms began, after searching google, she came up with a diagnosis list that included everything from influenza, to Zika, to lupus. She came to the hospital several days later when it became hard to breath and it turned out that she had a massive heart attack.
Cardiac pain originates from the heart muscle, most typically when blood flow to the heart (through vessels called coronary arteries) become blocked. In the heart muscle, there are nerve endings which transmit signals to the brain which get interpreted as chest pain. Unfortunately, just like other pain arising in other organs in the body, cardiac pain is poorly localized. Think of when you had stomach cramps after eating some food you probably shouldn't have--the pain is often vague and generalized, as opposed to being isolated to one specific location. Furthermore sensations arising from other organs in the chest, such as the esophagus, can produce pain indistinguishable from cardiac pain. Adding to this ambiguity is depending on your sex, race, and other factors, your description of pain may be significantly different than someone else. Topping it off, different healthcare professionals may also interpret your description of pain very differently. To help cut through the mountain of linguistic ambiguity, studies are now being done using artificial intelligence (AI) to help decode our description of symptoms to provide more accurate diagnoses.
At the recent American College of Cardiology meeting, I had a chance to discuss some of the developments in medical AI with Dr. Catherine Kreatsoulas, a Fulbright Scholar working at Harvard University. Dr. Kreatsoulas's research focuses on gender differences in cardiovascular disease and symptom differences between men and women. "There is a perception that women's symptoms are different than men" said Dr. Kreatsoulas. Her research suggested that while the symptoms between women and men may not be that different, "the way [women] describe symptoms are different." She also points out that "There may be differences in the way physicians interpret the symptoms described by men and women".
Her research, dubbed the HERMES study, is testing how AI algorithms can help predict the likelihood of cardiovascular disease based on what patients are saying and how they describe symptoms. This study will be evaluating outcomes based on patients who are going for a coronary angiogram, the gold standard in identifying blockages in the coronary arteries. These AI algorithms will develop over time and can be further refined with more data. The hope is that these algorithms will eventually lead to developing a lexicon that can help identify cardiovascular disease in women and men based on how they describe their symptoms.
From a medical standpoint, the implications of using linguistic AI algorithms are substantial. When calling 911 for help, based on how and what someone says, these AI algorithms may identify someone at high risk for a heart attack and can make appropriate triaging recommendation. Or conceivably, you can talk into your phone, and based on the lexicon of words being used, combined with your risk factor profile, your risk of cardiovascular disease can be accurately identified. These AI algorithms can be extended to other acute emergencies that need urgent intervention such as strokes.
While still in its infancy, AI can significantly change how we identify symptoms and diagnose patients. Thinking back to the patient I saw, I can't help but think she would have benefited from this type of technology, which could have picked up that she was having a heart attack much earlier on. As we move forward, linguistic AI algorithms can be a very valuable tool to help doctors make accurate diagnoses so that patients can receive timely and appropriate treatment.