AI Makes Good Doctors Even Better Doctors

Mar 6, 2017 11:00:00 AM

Highly-paid doctors will not be obsolete and radiologists or anatomical pathologists will not be obsolete in the age of AI in medical practice. In a recent New England Journal of Medicine Paper ( Drs. Obermeyer and Emanuel (1) rightly suggest that big data will transform daily medical practice and computer-based algorithms will guide clinical decision making.

There is good reason for AI to enter the field of medicine:  big data already overwhelm doctors, which includes the outputs of complex companion diagnostic tests, the management of new drugs (including new drug combinations), and the associated administrative activities that ensures the payment for the delivery of care. There is an increasing demand for better prognosis and response prediction from complementary diagnostics to inform patient surveillance, monitoring, and treatment strategies. At the same time doctors are requested to continuously educate themselves to stay on top of medical innovation and apply the most effective treatments available. But which doctor can actually achieve this and read all the latest publications and current guidelines? So the question remains:  How can doctors reclaim “quality time” with their patient to discuss their personal situation, explain what the situation is and still delivering best patient care?

Medical knowledge that directly influences the ways in which diseases are diagnosed and patients are treated must be continuously updated, processed, and understood by doctors. The interactions of different procedures and even more so combination therapies with innovative drugs are almost impossible to understand given the pre-existing complexity of the patient’s body and individual biochemistry based on genetic singularity. Here, AI will certainly contribute to the solution.

Not too long ago, the idea of advanced artificial intelligence was mostly discussed by futurists, speculators, and sci-fi writers. But today, due to swift developments in machine learning and data analysis, AI is on the cusp of surpassing science fiction and becoming scientific fact in medical practice.

The idea is basically this: If you take enormous computing power and feed it tremendous amounts of data (e.g., from millions of patient records including research data sets), you end up getting an artificial intelligence network that can interact with daily medical practice in a useful way. This is comparable to what the IBM Watson computer is trying to achieve.

But what else would be the immediate application for such an approach? The UK-based company Babylon is claiming that doctors only know a fraction of those 10,000 known diseases [2]. According to a 2012 study from researchers at John Hopkins, more than 40,000 patients die annually in the US alone as a result of misdiagnosis, yet an App might help Doctors make the correct diagnosis. While such an App can recommend a certain treatment based on the described symptoms (e.g., headache, coughing, sore throat, and chills for a flu-like condition), it certainly cannot interpret the nature of a skin rash in the patient’s face or at secluded parts of the body. It takes many years in medical school and clinical practice to learn how to describe symptoms in a standardized and agreed way. The same applies to an AI network – these networks also need to continuously learn.

An emerging number of AI-based tools provide doctors with guided solutions, which are mostly web-based repositories of medical information and clinical insights. It’s a bit like recommending purchases based on its massive trove of data about what people have bought in the past. Information technology giants like Google, Amazon, Microsoft, and Apple already have made huge investments in artificial intelligence to deliver tailored search results and build virtual personal assistants. Now, that approach is starting to trickle down into health care, thanks in part to the push under the health reform law to leverage new technologies to improve outcomes and reduce costs–and to the availability of cheaper and more powerful computers. In an effort to better treat their patients, doctors are now exploring the use of everything from IBM’s Watson supercomputer, the machine that won at Jeopardy, to iPhone-like pop-up notifications that appear in your online medical records.

With all potential upsides, AI is still in the early stages of development–in so many ways, it can’t match our own intuitive intelligence (sometimes also called experience) – and computers certainly cannot replace doctors at the patient’s bedside. But today’s machines are capable of crunching vast amounts of data and identifying patterns that humans cannot.

Let’s come back to AI in anatomical pathology and the role it will play in the future of doctors in the field: the reliable recognition of known and learned patterns allows the doctor to identify organs and to diagnose a disease that comes with a certain prognosis. Also, Drs. Obermeyer and Emanuel predict that machine learning will dramatically improve prognosis because it can identify morphological features (= phenes) that have been left ignored so far or simply could not have been detected by simple pathological insights. The availability of big data sets will make it possible to validate such new “signatures” which eventually become clinically meaningful prognostic algorithms in the not-so-distant future. 

Machine learning will also improve diagnostic accuracy. Many studies have already emphasized the high inter- and even intra-observer variability in radiology and anatomical pathology which can lead to an alarming frequency of diagnostic errors, not only putting patients’ lives at risk but also increasing health care costs. Algorithms will soon generate differential diagnoses to guide doctors’ decision making and suggest relevant tests that confirm the working hypothesis while ignoring those which will have no value.

So at this point, we can agree that the future for big data and machine learning in medicine is now. However, while big data is getting even bigger by the minute, we cannot accept algorithms without a medical plausibility check and robust clinical validation. Machine learning can be prone to overfitting which needs to considered as well as the curation of data that are fed into the system.  Therefore, the generation of new hypotheses should still be guided be the framework of human biology.

If both big data and machine learning are to reach the level of maturity for routine application in clinical medicine, doctors will regain much quality time that is still precious to all patients. Doctors will be able to access all the necessary medical information relevant to the patient’s case in addition to recommendations for a diagnostic path that may identify targeted treatment opportunities. Therefore, AI will make good doctors even better doctors because in the end doctors still carry the ultimate responsibility for their patients’ wellbeing. With the use of big data and machine learning, doctors can make better and more informed decisions with higher accuracy and better predict what will happen next.

Dr. Ralf Huss, CMO, Definiens

[1] Obermeyer Z, Emanuel EZ: Predicting the future – big data, machine learning, and clinical medicine. N. Engl. J. Med. 2016 September 29; 375(13):1216-1219