I am in an aeroplane crossing the Atlantic Ocean as I write this. We took off from Heathrow Airport more than three hours ago. By now, it’s likely the plane’s captain and crew are not physically in control of the aircraft. Something as complex as flying a metal tube packed with more than 300 living souls at 12,000 meters and 900kph is left to a computer and a set of algorithms. The autopilot.
Such a device is badly needed in our hospital wards. Critical patients needing 24/7 intensive care could certainly benefit from data-based approaches that could leverage on state-of-the-art analytics and AI.
For instance, a wise intensive care unit (ICU) nurse once told me: “Don’t get sick, but if you do get sick, don’t do it at night.” Data suggests evenings and weekends are not a good time to fall ill due to an increase of the patient’s risk of death. If our healthcare professionals had their capabilities augmented like the pilot in charge of my plane (who is able to rest now, so he can be 100 per cent focused during the approach and landing), we could not only get sick anytime without an increased risk of dying, we would also improve patient outcomes and decrease overall costs for the healthcare system. Just consider the upcoming shortage of medical professionals in the NHS and in the US, and the fact that medical errors are already the third-highest cause of death in the US (after heart disease and cancer) with 251,000 deaths in 2013.
Many healthcare organisations are working on potential AI applications. Research groups such as the Stanford Vision Lab are devoting efforts to the general use of AI in healthcare, and startups such as Etiometry in Boston and Better Care in Barcelona are focusing on critical care hospital units. Etiometry’s goal is to develop a predictive analytics platform to improve the quality of care in the ICU. Better Care is focusing on a software platform to capture biomedical data around the ICU patient – incorporating medical knowledge and algorithms. This is also an area of focus for companies such as Google, IBM and Qualcomm.
In the ICU, data from a patient is extensive and complex. But AI deals well with complexity. Based on a patient’s data, an AI platform could ensure the most basic mission of the ICU team (“keep the patient alive”): provide descriptive analytics for “what is going on”, predictive analytics for “what’s going to happen” and prescriptive analytics for “what shall you do”.
The first layer with descriptive analytics would help them understand “what is going on” with a specific patient within the context of thousands of other patients with that same condition. Crunching all that data in real time is an example of a skill set that is not yet available to human beings. The second layer would allow them to allocate resources according to “what’s going to happen” and the progressing complications of patients who are fighting for their lives. Finally, as the presence of AI in the ICU becomes the norm, the availability (and quality) of data would allow for the use of prescriptive analytics as a complement to trial and error that is still predominant when managing critical patients.
Of course, hospitals are not ready for this yet. Just consider that, in 2016, most world-leading hospitals still had no internet access in their operating rooms. Furthermore, doctors have historically been reluctant and conservative when it comes to the introduction of new technologies. To some extent, technology companies have also made the mistake of suggesting that AI will replace doctors – and no trade group likes to feel threatened. We cannot expect healthcare professionals to be free from error. It has been their creative thinking and diligent care that has driven our healthcare systems to greater heights. It has also been their human problem-solving that has allowed us to develop novel medical technologies, contributing to increased life expectancy.
The first step should be in complementing our doctors, not replacing them. Just imagine an ICU room with three screens reporting 20 essential parameters in real time – both invasive and non-invasive monitoring – along with data coming in from the labs, imaging tests and the discrete measurements and clinical observations made by healthcare professionals. The potential in this scenario is not just to mimic the doctors, but to perform tasks that no doctor can manage. If we are able to develop systems that enhance their capabilities and allow them to provide their patients better care, we will be in a win-win situation for healthcare professionals, patients and taxpayers.