Missing the Murmur

Emergency medicine physician recalls how AI could have expedited an endocarditis diagnosis.
December 31, 2020
• 
4 min read

Disclaimer - I am an Eko employee but I am also a practicing emergency medicine physician at Northbay Hospital in Fairfield, CA.

What you are about to read is what goes on in my mind after any imperfect patient encounter, similar to the postmortems seen in medical dramas on TV, and which many of us healthcare professionals experience daily. 
Recently, I saw a young patient with a swollen leg, some rib pain, and a low grade fever. As part of my usual emergency room exam, I listened to the patient’s heart with my Eko CORE stethoscope at the right upper sternal and apex positions. COVID-19 had been ruled out. I heard no murmurs, so I began pursuing the diagnosis of a pulmonary embolism (a blood clot) that originated from the leg, traveled through the heart, and lodged itself in the lungs. The patient’s diagnostic work-up included blood tests, lung scans, and a radiology consultation.

After 4.5 hours, the results came in as an unexpected surprise: the patient had endocarditis! But I hadn’t heard a murmur, something almost every patient with endocarditis should have. After conferring with the radiologist, I went straight to the patient’s bedside and listened carefully again at all 4 points of the chest. I still could not hear a murmur! Then, I enabled Eko’s AI analysis and rescanned all 4 points. Sure enough, the artificial intelligence algorithm found the murmur I had missed in the left middle sternal border - likely coming from the tricuspid valve. The AI took 20 seconds to find what I had missed even after multiple attempts.

Endocarditis is rare in young healthy patients with normal hearts, as long as they do not use intravenous drugs. Because all circulating blood must go through the heart valves, an infected valve is like a ticking time bomb. At any moment, the infection could come off the valve and spread to anywhere blood goes. For that reason, endocarditis is also known as a great “masquerader” because its pattern of attack is often first seen far away from the heart, and appears to be random and sometimes incongruous. House M.D. references it probably as often as lupus, which is to say probably half of the time. Left untreated, this shower of infected particles from the infected valve could lead to strokes, kidney failure, loss of fingers, toes and limbs, and damage to literally any other organ you can imagine. It is not only disabling, it is also deadly - 1-year mortality is as high as 40%. Endocarditis literally puts patients’ lives and limbs in serious danger.

Detecting endocarditis is difficult, but it is the most important disease to rule out in any patient with fever and a new murmur. In fact, 80% of patients with fever and a new or changed murmur have endocarditis. Because I did not hear a murmur, I did not suspect endocarditis. Thankfully, the CT scan of the lungs also incidentally scanned the heart, and the diseased tricuspid valve was found.

Could I have done better? Sure. If I had detected the murmur upon my initial auscultation examination, I would have ordered antibiotics immediately and sent the patient for a confirmatory echocardiogram. Instead, I lost precious time in starting therapy, and I was lucky that within those 4.5 hours the patient’s endocarditis did not spread elsewhere

This humbling experience also highlights the fact that I’m an imperfect human, as all physicians are. Like all humans, clinicians could use help to do better. In this case, it was the Eko AI that went beyond my capabilities. AI can and will make us better doctors, enabling us to provide better care for our patients. In cases like this, I am reminded to continue incorporating Eko’s AI analysis into more of my practice.