Visionary pundits have already painted pictures of futuristic healthcare scenarios with robots strolling through hospital corridors in place of doctors – examining patients, making diagnoses, and prescribing medications. A lot of hype surrounds artificial intelligence (AI) and its application in healthcare resulting in questions about its usability, safety, and practical value– and for good reason. We’re still a long way off from completely replacing human medical practitioners, and perhaps we never will. The objective for healthcare today is to move past the hype and chart a course for AI that balances its opportunities and liabilities, while understanding its current limitations and finding areas where it can create real value and solutions.
AI is not a singular technology but an umbrella term that includes using deep learning and machine learning (ML), among other methods, to perform intelligent, human tasks such as learning and reasoning. Deep learning occurs when an AI technology learns from repeated actions and becomes more efficient at adjusting for errors to modify its approach when faced with new data. Machine learning uses advanced statistical techniques to identity patterns in data and then make predictions. In both cases, deep learning and machine learning have real, practical, and valuable applications in healthcare, several of which will be discussed in depth below.
Physician shortages and burnout are a glaring problem in healthcare today. AI can help narrow the gap between supply and demand, making it easier for fewer physicians to see more patients while reducing stress, burnout, and overwork. At a basic level, AI can be used to automate administrative tasks, add features to electronic health records (EHRs), and augment workflows. Allscripts, for instance, has developed an EHR platform called Avenel which has the ability to adapt to an individual user’s preferences. Using ML, the platform recognizes patterns in physician workflows and eventually offers suggestions for orders, finishing an encounter, or certain treatment plans. Ultimately, the platform aims to create a system that is more intuitive and adaptive to practice patterns – which can save physicians two or three extra hours every evening catching up on documentation.
Streamlining and optimizing a physician workflow also depends heavily on the ability to gather all of a patient’s important data in one place, creating a seamless, intuitive, and comprehensive health profile. If a physician can see all of a patient’s data in one place, including gaps in care and complications with chronic and complex conditions, they can make better clinical decisions relating to the patient’s care. GE Healthcare, a leading imaging and monitoring provider, recently partnered with Roche Diagnostics, an in-vitro diagnostics leader, to create the industry’s first data-driven software that combines patients’ in-vivo and in-vitro diagnostics. By marrying a patient’s in-vitro diagnostic data including genomics, tissue pathology, and biomarkers with their imaging and monitoring data – and then adding data analytics and machine learning – clinicians will be able to access a comprehensive portfolio of patient information to make faster diagnoses earlier and develop individualized treatment. Artificial intelligence stands to play a pivotal role in this new technology, providing actionable insights from multiple datasets which could not be visible to a human eye nor synthesized by human cognitive processes.
AI can provide additional guidance by augmenting the clinical expertise and judgement that a physician exercises each day. Researchers from Oxford University are using AI to improve diagnostic accuracy for heart disease. Currently, about 1 in 5 echocardiograms, which are ultrasonic scans of the heart, are misdiagnosed each year – the equivalent to 12,000 patients. Inadvertently, patients can be sent home and then go on to have a heart attack. The researchers are training a machine learning system to recognize signs of heart disease by studying scans from previous patients, along with data about whether or not the patients went on to have a heart attack in the future. Already, this new system has improved diagnostic accuracy. Likewise, if a radiologist has to look through thousands of images each day and make very careful, nuanced decisions about each one – an AI engine that could parse through all the images quickly and identify patterns and points of interest for the physician to pay attention to would be very useful. In this way, AI operates as a “safety net” for physicians, giving them the benefit of a second eye and making significant headway into early disease identification.
Patient Adherence and Engagement
A lack of patient adherence to prescribed medications and low patient engagement levels are significant barriers to better chronic disease health outcomes. Adherence to chronic medications is currently estimated at a mere 50 percent, and nonadherence causes around 125,000 deaths and up to 25 percent of hospitalizations each year in the U.S. Medical wearable devices that incorporate AI can help encourage greater patient adherence and engagement. AI has the ability to make suggestions and tailor feedback based on learning. As AI collects individual patient data and begins to learn how patients react differently to feedback, it can begin tailoring feedback that is personalized and predictive.
For instance, an AI-enabled medical wearable device can be prescribed to a patient as part of a hypertension care plan, wherein the patient can use the wearable to record blood pressure readings, food intake, and food purchases. Through deep learning, the embedded AI engine might be trained to evaluate food options in terms of sodium content and then measure how the patient’s blood pressure reacts after the consumption of certain high sodium foods. When the patient next logs the consumption of a high sodium food, the wearable may send an alert to the patient, reminding the patient that this food causes a spike in their blood pressure reading. An added ancillary benefit may be that the patient will start reading nutrition labels on food packaging for sodium content. Such feedback is the foundation upon which a preventive healthcare system is built, and critical for improving the long-term management process. The experience of getting reminders, inputting data, and receiving feedback can guide patients to actualize better health through long-term medication adherence and positive lifestyle choices.
The opioid crisis has been declared a public health emergency, with an estimated 115 deaths each day in the U.S. due to overdoses. One particular challenge is managing opioid prescriptions after surgery. Although pain management is important, excess opioid prescribing is a gateway to overuse. This was the problem targeted by the winning team of the Nokia Medicine X Digital Health Challenge, an international research challenge using wearables to investigate clinical problems. The researchers used smart watches to track the physical activity and sleep patterns of 36 patients for 30 days following knee arthroplasty. They correlated this data with opioid prescribing patterns and actual opioids used (patient self-reporting through a smartphone app). The key finding was that the average amount of opioids used per day was much less than the total amount prescribed. In particular, early sleep disturbances were associated with high opioid consumption at a later time.
This study highlights the potential for wearable devices to help track opioid consumption and possible means of identifying patients at high risk for persistent post-surgical pain and opioid use. Through machine learning, for example, a medical wearable device with AI capabilities can identify patterns in a patient’s metrics and data to make an intelligent prediction as to whether the patient is at greater risk for overusing and/or becoming dependent on opioids. The wearable might also evaluate other relevant risk factors for opioid overuse and dependency. These insights would in turn assist physicians in better customizing opioid prescriptions for individual patients based on risk metrics and use patterns.
One Small Step for AI, One Giant Leap for Healthcare
AI’s transformative power is already reverberating across the healthcare industry, helping it become both more effective and inexpensive. Today we are seeing AI used mainly in data intensive applications, from supporting physicians in disease diagnosis to helping them sift through patient records. We are also seeing an impeccable shift in the way patients are diagnosed and in the way treatment is chosen and managed because physicians and patients now have actionable and medically relevant data that can be put to good use. AI’s next leap forward will likely see it land in the realm of practical applications as it begins to tackle supply and demand issues. It might truly begin to augment human activity, displacing the current burden on human physicians to review, understand, and interpret clinical data. We don’t know what the future will bring, but one thing is certain already – AI is truly a game changer for healthcare.