The term “artificial intelligence”, defined in the Oxford Dictionary as “the capacity of a machine to simulate or surpass intelligent human behaviour”, was first used in the 1950s, but the idea that the human brain, or at least parts of it, could somehow be ‘copied’ or ‘imitated’ using a series of mechanical networks has been part of myth and literature for centuries, as far back as, for example, the automatons of Hephaestus and Daedalus.

by Prof Jean-Louis Vincent

Artificial intelligence is an umbrella term that covers multiple tools and technologies. In recent years, huge advances in machine learning algorithms, neural networks, deep learning, facial recognition, computing power and statistical techniques have increased the potential applications of artificial intelligence across multiple aspects of everyday life. In terms of medicine, artificial intelligence is already beginning to find a place in disease prediction and diagnosis, clinical decision support and therapeutic guidance, prognostication, and remote follow-up and monitoring.

So, what do these new developments mean for the medical profession – is this ever advancing technology going to decrease the need for doctors? To some extent, the answer is of course yes, perhaps particularly in the field of diagnostics. Using machine learning, computerized systems are able to study and develop pattern recognition from millions of test results over a very short period of time, whereas a doctor can only see and process a limited number of events throughout a whole professional career. The vast amounts of data becoming available as increasing numbers of patients have electronic health records and are managed with electronic monitoring systems are providing the input needed for artificial intelligence to develop, while also reducing the administrative load on doctors and other healthcare staff, a common source of complaint. Machine-based interpretation of visual data is now more accurate, effective, and rapid than the human eye and associated with less risk of error. Interpretation of X-rays and other medical images will become increasingly automated, reducing the need for radiologists to perform this role. Similarly, electrocardiograms are already read automatically; many of the reports are still validated by a cardiologist, but is this really necessary? In ophthalmology, examination of the fundus for many diagnoses, including diabetic retinopathy and macular degeneration, can be performed more accurately by computerized algorithms than by most experienced ophthalmologists. Application of deep learning to a database of eye CT-scans at a London hospital has enabled more than 50 ocular diseases to be recognized by the system. Non-image-based diagnosis is also possible. A pilot system developed in California that applied machine learning to electronic health record data has been used to diagnose common pediatric problems and shown to perform better than junior doctors.

Although artificial intelligence may therefore indeed reduce the need for doctors in some areas, this is not necessarily a bad thing; it should not be seen as the enemy, but rather as a means of improving medical practice and patient care. More rapid, accurate diagnosis can only be positive. Another important example of how artificial intelligence can be beneficial is in predictive medicine. Natural language processing of data from electronic medical records was able to more accurately predict postoperative complications in patients undergoing major surgery than were patient safety indicators. Machine learning-based sepsis prediction has been associated with reduced mortality rates. Indeed, early identification of patients with an increased risk of sepsis raises awareness and facilitates early investigation and management by sepsis-trained physicians in a condition where rapid diagnosis and treatment are crucial to maximize chances of survival. More broadly, identification of patients at risk of deterioration on the general floor or after surgery will enable appropriate monitoring systems to be offered to a selection of patients in an environment where it is not feasible, financially or practically, to monitor every hospitalized patient.

Another example is the field of therapeutics. Machine-learning can identity the most effective therapeutic algorithms for individual patients, with the most appropriate evaluation and follow-up. These algorithms are being incorporated into closed-loop systems enabling appropriate treatments to be administered and doses adjusted in real-time for individual patients according to continual personal data provision. Just some of many examples include implantable cardioversion defibrillators, diabetes management with implantable glucose monitors and insulin pumps, and use of vasopressors to prevent hypotension during surgery. 

Artificial intelligence can thus be used to provide continually adjusted, up-to-date, patient-relevant information on the best therapeutic options for a specific patient much more rapidly and accurately than doctors can. Doctors can currently assimilate only a small portion of the information available on which to base such decisions. By configuring the huge amounts of patient data now available into predictive models, machine learning can provide physicians with recommendations regarding the optimum treatment schedule for a specific patient with a specific condition(s). Doctors are still needed to interpret the recommendation, explain them to the patient and their family, and initiate the chosen treatment course. Using artificial intelligence, doctors will be more enabled to select the optimum treatment regimens for individual patients, rather than the one-treatment-fits-all packages that are still used for many conditions today. Moreover, doctors will have greater access to accurate statistics to inform discussions with patients and their families, both about the likely outcomes from different treatment options, but also about likely prognosis if treatment is refused or withdrawn. The role of doctors in this situation will increasingly be one of advisor as patients become ever more involved in their own healthcare decisions.

Artificial intelligence is going to continue to invade every aspect of daily life, including medicine, and needs to be carefully developed and validated. Although currently very much associated with rich economies of the developed world, artificial intelligence could play a large role in improving healthcare in countries where qualified doctors are less widely available. These new technologies could therefore help to decrease inequalities in the health sector. Even in low income countries, the combination of artificial intelligence with telemedicine can help increase access to high quality medicine. Wherever it is employed, it will not entirely replace doctors but rather make them more efficient, provide more time for one-to-one patient contact and help provide care best adapted to individual patients, with reduced errors. People often raise the question of associated costs. But costs may not be very high, especially as the technology becomes more widely used. Indeed, artificial intelligence may even help reduce global healthcare costs by increasing efficiency.

Doctors should not fear this technology, but learn to accept and embrace it, determining and directing how it can best be incorporated into medical practice to improve patient care.
The author
Jean-Louis Vincent, MD, PhD
Dept of Intensive Care, Erasme Hospital, Université libre de Bruxelles, Brussels, Belgium

Suggested reading:
Darcy AM, Louie AK, Roberts LW. Machine learning and the profession of medicine. JAMA 2016;315:551-2.
De Fauw J, Ledsam JR, Romera-Paredes B, et al. Clinically applicable deep learning for diagnosis and referral in retinal disease. Nat Med 2018;24:1342-1350.
Joosten A, Alexander B, Duranteau J, et al. Feasibility of closed-loop titration of norepinephrine infusion in patients undergoing moderate- and high-risk surgery. Br J Anaesth 2019 Jun 26. doi: 10.1016/j.bja.2019.04.064129
Kilbride MK, Joffe S. The new age of patient autonomy implications for the patient-physician relationship. JAMA 2018;320:1973-1974
Liang H, Tsui BY, Ni H, et al. Evaluation and accurate diagnoses of pediatric diseases using artificial intelligence. Nat Med 2019;25:433-438.
McCoy A, Das R. Reducing patient mortality, length of stay and readmissions through machine learning-based sepsis prediction in the emergency department, intensive care unit and hospital floor units. BMJ Open Qual 2017;6:e000158
Murff HJ, FitzHenry F, Matheny ME, et al. Automated identification of postoperative complications within an electronic medical record using natural language processing. JAMA 2011;306:848–855. 
Topol EJ. High-performance medicine: the convergence of human and artificial intelligence. Nat Med 2019;25:44-56
Vincent JL, Creteur J. Big data are here to stay! Anaesth Crit Care Pain Med 2019;38:339-340.
Vincent JL, Einav S, Pearse R, et al. Improving detection of patient deterioration in the general hospital ward environment. Eur J Anaesthesiol 2018;35:325–333.