Welcoming Medicine To The Machine

Abstract

Book Review Health AffairsVol. 38, No. 9: Neighborhoods & Health, Medicaid & More Welcoming Medicine To The MachineDhruv Khullar AffiliationsDhruv Khullar ([email protected]) is a physician at NewYork-Presbyterian Hospital and an assistant professor in the Departments of Medicine and Healthcare Policy and Research, Weill Cornell Medical Center, in New York City.PUBLISHED:September 2019Free Accesshttps://doi.org/10.1377/hlthaff.2019.01001AboutSectionsView PDFPermissions ShareShare onFacebookTwitterLinked InRedditEmail ToolsAdd to favoritesDownload CitationsTrack CitationsPermissionsDownload Exhibits TOPICSDiseasesSystems of careComputed tomographyPharmaceuticalsPatient safetyEric Topol sees a future in which doctors use artificial intelligence (AI) to analyze billions of pieces of medical, social, genetic, and environmental information—continuously and automatically collected—to produce diagnoses and treatments individually tailored to the patient in front of them.It’s a future in which algorithms digest the peculiarities of your gut’s microbiome (advising, say, strawberry Danishes rather than banana nut muffins) and in which the medication prescribed for your particular flavor of diabetes was discovered by a machine that sifted through trillions of molecules to identify the most promising chemical compounds. It’s a future with AI-powered electronic medical assistants—Alexas with medical school degrees—who listen to your visit, place orders, schedule follow-up, and generate the required documentation. All without your doctor ever breaking eye contact.Topol, a cardiologist, geneticist, and director of the Scripps Research Translational Institute, is the author of three books about the digitalization of health care. His most recent, Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again, is an exhaustive tour-de-force review of the past, present, and future of AI in medicine,brought to life with compelling personal anecdotes about his life as a patient, physician, husband, and son. Topol makes a passionate—and compassionate—case that while AI can improve the safety and efficiency of health care, where it will have a truly transformative impact is in freeing clinicians from mundane tasks and allowing them to spend more time with patients. AI, he argues, could make medicine more human.There’s certainly room for improvement. The US health care system is the most expensive in the world but fails to deliver outcomes commensurate with its cost. By some estimates, there are twelve million significant diagnostic errors each year, and up to 50 percent of the eighty million computed tomography (CT) scans performed annually in the US are unnecessary. Fax machines still transfer medical records, physicians struggle with epidemic levels of burnout, and patients—even those who can access care—often don’t feel cared for.AI could help. If Atul Gawande argues that medicine has grown too complex for individuals, so we need teams (physician-cowboys must now join pit crews), Topol thinks that medicine has grown too complex for humans, so we need machines. He explores both the promise and the limits of AI and argues that we need humans and machines working together to address the health care challenges of the twenty-first century.Topol describes the case of a newborn boy who, initially healthy and breast-feeding well, developed unrelenting seizures on his eighth day of life. The usual workup (blood tests, CT scans, electroencephalograms) was unrevealing. A blood sample, containing 125 gigabytes of data, was then sent out for analysis. An AI system found five million variants in his genome and identified the thousand most likely to cause disease. After linking his genetic information with phenotypic information from the electronic health record, the system settled on a single mutation most likely to be causing his seizures—the treatment for which was dietary supplementation with vitamin B6 and arginine. The boy was discharged home within days and has been healthy since.But without human judgment, AI is not enough, and will never be enough, according to Topol. He shares another story: this one of an older man with fibrotic lung disease who became increasingly fatigued when walking, despite stable pulmonary function tests. Topol thought that the problem might be related to heart disease, but a CT scan showed narrowing of only the right coronary artery—an unlikely culprit, by itself, for the patient’s profound weakness. An algorithm, based solely on the literature, would have advised against a cardiac stent, but Topol reasoned that the combination of the patient’s lung disease and the blockage in the artery might be causing fatigue in his case. The patient got the procedure, and by the end of the week he was swimming laps again. “What’s remarkable about this story,” Topol writes, “is that a computer algorithm would have missed it.”Topol acknowledges that AI in health care is still very much in its infancy. He admits that the “field is long on…promises but very short on real-world, clinical proof of effectiveness.” He describes the many risks of widespread use of AI, including data privacy and security issues, the “black-box” nature of many algorithms, the ease with which outputs can be manipulated, and the potential for algorithms to incorporate human biases and worsen health disparities.Still, Topol takes a generally optimistic view, especially when it comes to his central thesis: Machines can not only make medicine safer and cheaper and more convenient, but they can also make it more humane. “The greatest opportunity offered by AI is not reducing errors or workloads,” he contends. “It is the opportunity to restore the precious and time-honored connection and trust—the human touch—between patients and doctors.”Topol invokes the “gift of time” that AI could offer by assuming many of the unfulfilling clerical tasks currently performed by clinicians. Here there is reason for hope, but also for doubt. The reasons why patients and physicians are often unsatisfied with the length and quality of clinical interactions are as much cultural, economic, and administrative as they are technological. Even if AI ultimately succeeds in making care more efficient, it’s far from clear that clinicians will get to spend more time with patients than they do today. What is to prevent health systems from using the “extra” time to, say, schedule more patients per day? Or policy makers from demanding that the efficiency gains created by AI be used to reduce the number of clinicians—the largest expense for most health systems—and divert resources to other sectors?A second reason for caution lies in the gap between knowledge and behavior. AI excels at identifying patterns and making predictions. But that’s only half the battle. We know that cigarettes cause cancer and obesity is linked to heart disease. That doesn’t make it any easier to stop smoking or lose weight. AI may predict the chance that a patient will forget to take their medication to the nearest tenth of a percent, but improving adherence remains a stubbornly knotty problem.In Deep Medicine Topol offers perhaps the most comprehensive overview of AI in health care to date and articulates a compelling vision for its future. Whether that future is the stuff of science or science fiction remains to be seen. Loading Comments... Please enable JavaScript to view the comments powered by Disqus. DetailsExhibitsReferencesRelated Health Affairs may receive a commission for purchases through links. We appreciate your support of Health Affairs! Article Metrics History Published online 3 September 2019 Information© 2019 Project HOPE—The People-to-People Health Foundation, Inc.PDF download