AI Heart Attack Predictions Surpass those of Physicians
Self-taught artificial intelligence can predict heart attacks more accurately than doctors.
Artificial intelligence (AI) is progressing rapidly and scientists believe that computers can now be used to predict heart attacks in patients better than doctors can. Vascular surgeon Elsie Ross at Stanford University in California stresses the importance of this work and hopes that doctors will begin to embrace AI. If scientists can implement this breakthrough on a large scale, it could save the lives of untold numbers of people. Each year there are 20 million deaths attributed to cardiovascular diseases including strokes, heart attacks, and coronary artery disease.
The ACC/AHA guidelines doctors use to better predict these diseases weigh the risk factors of each patient which include age, blood pressure, and cholesterol. However, this system does not take into account many other factors like patient medications, diseases, and lifestyle risks. Epidemiologist Stephen Weng from the University of Nottingham says there are many counterintuitive interactions in the human body such as body fat which in some cases can protect against cardiovascular diseases. The hope is that machine-learning AI will quickly compute all factors and better predict future heart disease in patients.
Self Taught AI Predicts Cardiovascular Events
The new study compared four computer algorithms using machine-learning techniques. The computers were given the task of analyzing 378,256 patient medical records from the United Kingdom. The AI algorithms were put to work trying to find patterns from the data that would point to cardiovascular events. The AI was designed to learn by itself, build its own guidelines, and then test themselves for accuracy. The computers went on to predict possible cardiovascular events in patients 10 years into their future. The records analyzed were from 2005, and the researchers compared the AI guesses in predicting cardiovascular events against patient 2015 medical records.
What scientists discovered was that all four AI algorithms performed better than human conclusions based on patient records. Using a scoring system called AUC which uses numbers between 0.0 and 1.0, doctors conclusions equaled about 0.73 (73% accuracy), but the computer AI conclusions equaled about 0.75 (75% accuracy). The neural networks algorithm was the best performer scoring 76% accuracy. Using medical records from 2005 to 2015, 355 patients who died were identified by the computer AI as candidates for possible cardiovascular problems. If these people had preventative care through changes in diet or cholesterol lowering medication, their lives could have been saved.
Will Doctors Soon Adopt Artificial Intelligence?
According to data scientist Evangelos Kontopantelis from the University of Manchester, providing more record data or committing more computers to solve the problem could have obtained greater results. The AI algorithms identified several cardiovascular risk factors not considered by standard ACC/AHA guidelines such as oral corticosteroids or mental illness, but the AI did fail to identify diabetes as a risk factor. Scientists hope that providing more data to the AI algorithms such as genetic or other lifestyle factors will further enhance the prediction results.
There are limitations when working with artificial intelligence as it is hard for humans to understand what is going on inside the AI as it makes conclusions. It also takes time for computer scientists to refine machine-learning algorithms. Also when applying the AI to new scenarios, the AI is unpredictable in its computations. Despite these problems, the hope is that doctors will soon adopt computer AI in assisting their patients and possibly saving more lives in the future.