Health Care in 2030: Will Artificial Intelligence Replace Physicians?

Abstract

In 2016, the world chess champion was a computer program named Stockfish 8. That a computer was the reigning champion is no surpriseStockfish is programmed with centuries of accumulated human knowledge of chess and can examine 70 million positions per second. This changed in 2017 with the introduction of Google s AlphaZero (1). Coders gave AlphaZero just 1 input: the rules of the game. After only 9 hours of training, AlphaZero scored 28 wins and 72 draws in a 100-game match against Stockfish. It did not lose a game. Instead of building on centuries of human knowledge, it started with a clean slate. It then used machine learning to play 44 million games against itself to generate new insights and become the best in the world (2). Grandmasters described AlphaZero s playing style as combining the finesse of a virtuoso with the power of a machine. This was humankind s first glimpse of an awesome new kind of intelligence, showing creativity, beauty, and perhaps even playfulness. What might this tell us about what medical practice will look like in the not-so-distant future? Will Google algorithms beat out physicians? I don t think so. Algorithms, such as AlphaZero, cannot articulate what they are thinking. We do not know why they work and therefore do not know whether they can be trusted. Humans want more than answers. They want insight, particularly in medicine. This is a source of tension in physicians interactions with computers. Algorithms perform well when there are rules; they do not do as well when there is imperfect information or less structure, as in medicine. Self-driving cars may do well in a small city, but they could not move an inch in the chaos of New Delhi traffic. The reality of medical practiceparticularly primary care practiceresembles a New Delhi traffic jam. How can we make things better as we look ahead? Adeptus Health, a company of approximately 100 freestanding emergency departments (EDs) in Texas and Arizona that are located close to residential communities, offers some instructive examples (3). Adeptus reimagined freestanding EDs as the front porch of the hospital rather than the front door. At a given time, a physician, nurse, radiology technician, and receptionist staff each ED. By seeing approximately 10 patients over 24 hours, an Adeptus ED can earn a profit. They use standardized workflows for common presenting problems and efficient team-based care where the radiology technician also performs phlebotomy and electrocardiography. Flexible staffing enables each ED to scale from 8 to 80 patients in 1 day. During flu season, a single Adeptus ED saw up to 120 patients in 24 hours. Adeptus EDs collectively break many traditional rules to function well. Do EDs have to be part of a hospital? Should we hand off patients among multiple teams and providers, in the process matching each organ to a specialist? Should ED physicians multitask across many patients simultaneously? In this model, the answer to all of these questions is no. The outcomes speak for themselves. The average door-to-electrocardiography time is 4 minutes compared with 30 minutes in most hospital-based EDs. Without handoffs in care, fewer errors occur. Not diluting expensive physician time across dozens of patients enables these EDs to best align patients needs with their preferences. Standardized care pathways facilitate evidence-based care. Physicians and nurses triage and discharge patients jointly, and 2 clinicians can better catch potential mistakes, reinforce patient learning, and develop trusting relationships with patients. Physicians or nurses also telephone many patients within 24 hours of ED discharge to check that they are safe. Feedback loops enable continuous quality and service improvement. The commitment to optimizing each component of care telescopes: Just as the ED is outsourced from the hospital, radiology is outsourced through teleradiology. Further, the job satisfaction of the physicians involvedsome of whom fly in from other parts of the country for a week of service each monthis off the charts. Patient satisfaction is also high. Although this experiment is still relatively new, it and other novel models around the country show promise. CityMD has thriving urgent care clinics in New York City. Patients looking for care can visit the CityMD Web site to see actual wait times at each location (4). Millennials, who value convenience and access over nearly everything else, are not willing to wait the average 29 days needed to obtain a primary care appointment. CityMD sees patients in minutes and controls where they go afterward. Owning that downstream channel is big business. Based in San Francisco, Forward is another practice targeting Millennials (5). For $149 each month, members receive unlimited visits, genetic testing, cardiac screening, and more. There is a cool app, an iPad that shows patients their playlist of tests and visits for the day at check in, and more gadgets and monitors than an electronics store. Although Forward is unlikely to replace most practices, it has elements that physician practices must understand and incorporate to remain viable. The sensors, platforms, and screens are noise behind what really matters to patients: convenience, access, and no hidden fees. ChenMed in Miami believes that workflow should be based on the complexity of the patient (6). Typical patients receive a 15-minute appointment, and atypical patients receive a 1-hour appointment. As such, ChenMed physicians spend 189 minutes per year per patient compared with an average of 21 minutes for Medicare patients in traditional primary care practices. Studies suggest that more time spent with primary care physicians is associated with lower costs. Oak Street Health was started in Chicago and aims to deliver the world s best primary care to the poorest, sickest elderly patients (7) by being evidence-based, equitable, and accountable while focusing on social determinants of health. For Oak Street, this focus means covering patients transportation to and from the clinic, among other things. Their model is a full-risk, globally capitated one that primarily includes dual-eligible patients. Their quality metrics are impressive: a 92% Net Promotor Score, Healthcare Effectiveness Data and Information Set 5-star ratings, and a 40% reduction in hospitalizations. Knowing your patients and redefining health care to match their needs make a big difference. These practices are financially viable, whereas many other practices struggle. Common elements of these and other promising care models include hiring the right clinicians, using data analytics to better target patient needs, and being willing to try new things. However, none of these examples offers the complete answer. So, is there a secret sauce for high-value primary care? Arnold Milstein, Director of Stanford University s Clinical Excellence Research Center, has identified characteristics that high-value primary care practices share. After ranking more than 50000 practices on quality and cost, his research team identified consistent highest performers and compared them with average practices (8). They found that the best performers create deeper patient relationships through extended hours and thoughtful use of tools, such as e-mail. They practice conscientious conservation of resources, adhere to guidelines, leverage decision support, aggressively close care gaps, and use morning team huddles to match patient needs to services. They practice informed shared decision making and good advance care planning. They use patient complaints to guide improvement. They perform more of the basic services, such as stress tests, in-house. When referral is necessary, they coordinate care and choose specialists who also embrace conscientious conservation. They keep overhead low with modest offices. Finally, regardless of how they are reimbursed, they are thoughtful about how they pay themselves (for example, they share bonuses with the frontline staff). I have not really said much about technology, which you might have thought would be the focus of a commentary on the future of health care. The reality is that I cannot say where we will be. Twenty years ago, they said that electronic health records would be the silver bullet; look where we are today. Five years ago, artificial intelligence was predicted to render every radiologist, dermatologist, and pathologist obsolete; that has not happened. However, by 2030 physicians may have digital assistants that listen in on health care encounters and simultaneously write notes for clinical care, the patient, and billing purposes. These digital assistants will create referrals on the fly and arrange preventive and other evidence-based care. However, the physician will remain important, especially to manage the complex tasks left behind after technology addresses the easy stuff. This complexity will demand a higher-level performer who can understand what is right for the patient in ways that no algorithm can. General internists are such high-level performers. Educators suggest that the next generation of students should focus on critical thinking, communication, collaboration, and creativity (9). These are already the superpowers of internists. Google s AlphaZero may have helped create a new philosophy of chess. It is up to us to imagine a new practice of medicine.