Clinicians someday be able to predict more accurately how long patients with fatal diseases live. Medical systems learn how to save money without going through the expensive and unnecessary tests. Radiologists will be replaced by computer algorithms.
These are just some of the realities patients and doctors should prepare to as “machine learning” enters the world of medicine, according to Dr. Ziad Obermeyer, an assistant professor at Harvard Medical School, and Dr. Ezekiel Emanuel, of the University of Pennsylvania, who recently co-authored a article in the New England Journal of Medicine on the subject.
But what is “machine learning”? And how medical systems make use of it?
The article continues after the announcement
Obermeyer, who is also an emergency physician at Boston’s Brigham and Women’s Hospital, spoke with STAT to give some answers. This discussion has been edited and condensed.
How is the different learning, for example, artificial intelligence machine?
The traditional approach to solving problems with technology is to give the team some rules and apply brute force computing. With machine learning, which actually does not give machines rules. They are given the data and asked to learn the rules.
we can bring this very powerful tool in a medical problem and say, “.. I’ll show a lot of people who had heart attacks, and a group that they did Go learn to distinguish” Then, once the algorithm has seen a million patients and what happened to them, you can display that information on a new patient and let predict if you might be at imminent risk of a heart attack.
These algorithms are extraordinarily good at telling the difference. What we need to know more is, what are the rules of the machine are being learned, and how to reach those rules? In a way it is the next frontier of this.
Can not see the logic of the computer you used to reach a conclusion?
That’s really strange. With machine learning, either in medical or other centers, there are really only predictions.
Do doctors will be willing to accept the conclusions of an algorithm without understanding how those conclusions achieved?
That kind of thinking is no stranger to medicine. With something like the development of the hip prosthesis, it was kind of product engineering that left a deep understanding of the mechanics of the hip. But in the history of medicine, there are also plenty of things that make less sense. Think about the discovery of steroids for immune suppression, where the medicine begins with a very pragmatic observation “Oh, this works”, and then goes to work in the attempt to fill our understanding of why it works. That will be the model for many applications of machine learning.
Your article suggests that all medical jobs, radiology, in particular, may face the most profound changes as a result of machine learning. Is it still an option intelligent race to medical students?
In 20 years, radiologists do not exist anywhere near its current form. It might look more like cyborgs: monitoring algorithms reading thousands of studies per minute and zoom to inspect and judge ambiguous cases; or they could become “diagnosticadores” as Dr. House, to go out and have more contact with patients and integrate into their diagnostic judgments.
Think of construction workers: They remain essential for building – but they are doing very different than they were before mechanization work 100 years ago. Bank tellers do not provide more cash, but they are able to handle much more complex than they used transactions.
Technology is not always eliminated jobs; sometimes they are changed, and is sometimes beyond recognition. So those who adapt, as I imagine highly educated doctors could do, it can become big winners.
What will be the first place or how the patients feel the impact of machine learning, and when you can predict that?
One of the areas where we will see the fastest transition is in the manufacture of very personalized predictions for patients depending on their history and únicas career. In areas like end-of-life care or optimizing the use of diagnostic tests for diagnostic purposes. We will see many of these applications will be operational very quickly in the coming years.
currently we do not yet have studies to show that this directly improves care. Once they start to leave, however, I think it’s going to be an explosion in this type of technology where information is taken available – which occurs as the escape of clinical care we have given – and recycle that predictions very adapted to the risks and future patient outcomes.
What is the best example of that?
better prognostic information about the end of life. “How much longer do I have?” It is a very common question is that doctors are quite poorly equipped to answer. We find that such estimates are for most measures far from what patients end up experiencing or doctors could not give any information, partly because they do not want, but also partly because we do not know how.
Predicting remaining life of the people is actually one of the easiest applications of machine learning. a single set of data that have information related to electronic records about people died when required. But once we have that for enough people, you can reach a factor of very accurate prediction of the probability of a person being alive a month out, for example, or a year away.
Such information is a very important value for patients and physicians. Patients have a lot of things they need to plan around the end of life, whether advance directives or medical proxies, but also for doctors who need to know how to think about a plan of treatment and diagnostic tests for the next months.
What is the greatest obstacle to achieving significant benefits of machine learning? And how do we get beyond that?
Two. First, these data are extremely messy when they leave the electronic medical record or database of the insurer. And it is also very difficult to link to other data. Say you want to know when someone died. If they go home and die, or die in another hospital that is not part of its health system, you will not know who died. So that fragmentation not only within our health care system, but through the layers of our health care system is a huge challenge. It is solvable, but it takes time.
The biggest – obstacle not so much, but requirement – has not been fulfilled is that, in medicine, actually have a playbook for change of medicine, where we study and measure a very rigorous way whether or not something works.
I think that for all the enthusiasm for machine learning in clinical medicine to date, has not been accompanied by a flowering of activity in the rigorous testing of ideas and interventions. It’s all very well and good to say that you have an algorithm that is good at predicting. Now we actually harbor them for the real world in a safe and responsible and ethical way and see what happens.