The dream of replacing doctors with advanced artificial intelligence is unsurprising, as talk of robots replacing human workers in various fields — from eldercare to taxi driving — has become common. But is Vinod Khosla right about medicine? Will we soon walk into clinics and be seen by robot diagnosticians who will cull our health information, evaluate our symptoms, and prescribe a treatment? Whether or not the technology will exist is difficult to predict, but we are certainly on our way there. The IBM supercomputer Watson is already being used in some hospitals to help diagnose cancer and recommend treatment, which it does by sifting through millions of patient records and producing treatment options based on previous outcomes. Analysts at Memorial Sloan Kettering Cancer Center are training Watson “to extract and interpret physician notes, lab results, and clinical research.” All this is awe-inspiring. Let us generously assume, then, for a moment, that the technology for Khosla’s future will be available and that all knowledge about and treatment options for medical problems will be readily analyzable by a computer within the next decade or so. If this is the future, why shouldn’t physicians be replaced?
There are several errors in Khosla’s way of thinking about this issue. First of all, modern health care is not “like witchcraft.” Academic physicians, for example, use evidence-based medicine whenever it is available. And when it isn’t, then they try to reason through a problem using what biologists know about disease presentation, physiology, and pharmacology.
Moreover, Khosla mischaracterizes the doctor-patient interaction. For Khosla, a visit to the doctor involves “friendly banter” and questions about symptoms. The doctor then assesses these symptoms, “hunts around ... for clues as to their source, provides the diagnosis, writes a prescription, and sends you off.” In Khosla’s estimation the entire visit “should take no more than 15 minutes and usually takes probably less than that.” But the kind of visit Khosla writes about is an urgent care visit wherein quick and minor issues are addressed: strep throat or a small laceration requiring a stitch or two. Yes, these visits can take fifteen minutes, but so much of medicine does not involve these brief interactions. Consider the diabetic patient who has poorly controlled blood sugars, putting her at risk for stroke, heart attack, peripheral nerve destruction, and kidney failure, but who hasn’t been taking her medications. Or consider a patient addicted to cigarettes or on the verge of alcoholism. Consider the patient with Parkinson’s disease who wonders how this new diagnosis will affect his life. And what about the worried parents who want antibiotics for their child even though their child has a viral infection and not a bacterial infection? I can go on and on with scenarios like these, which occur hourly, if not daily, in nearly every medical specialty. In fact, fifteen-minute visits are the exception to the kind of medicine most physicians need to practice. One cannot convince an alcoholic to give up alcohol, get a diabetic patient to take her medications, or teach a Spanish-speaking patient to take his pills correctly in fifteen minutes. In addition, all this is impossible without “friendly banter.”
As Dr. Danielle Ofri, an associate professor of medicine at the New York University School of Medicine, wrote in a New York Times blog post, compliance with blood pressure medications or diabetic medications is extremely difficult, involving multiple factors:
Besides obtaining five prescriptions and getting to the pharmacy to fill them (and that’s assuming no hassles with the insurance company, and that the patient actually has insurance), the patient would also be expected to cut down on salt and fat at each meal, exercise three or four times per week, make it to doctors’ appointments, get blood tests before each appointment, check blood sugar, get flu shots — on top of remembering to take the morning pills and then the evening pills each and every day.
Added up, that’s more than 3,000 behaviors to attend to, each year, to be truly adherent to all of the doctor’s recommendations.
Because of the difficulties involved in getting a patient to comply with a complex treatment plan, Dr. John Steiner argues in an article in the Annals of Internal Medicine that in order to be effective we must address individual, social, and environmental factors:
Counseling with a trusted clinician needs to be complemented by outreach interventions and removal of structural and organizational barriers. ...[F]ront-line clinicians, interdisciplinary teams, organizational leaders, and policymakers will need to coordinate efforts in ways that exemplify the underlying principles of health care reform.
Therefore, the interaction between physician and patient cannot be dispensed with in fifteen minutes. No, the relationship involves, at minimum, a negotiation between what the doctor thinks is right and what the patient is capable of and wants. To use the example of the diabetic patient, perhaps the first step is to get the patient to give up soda for water, which will help lower blood sugars, or to start walking instead of driving, or taking the stairs instead of the elevator. We make small suggestions and patients make small compromises in order to change for the better — a negotiation that helps patients improve in a way that is admittedly slow, but necessarily slow. This requires the kind of give-and-take that we naturally have in relationships with other people, but not with computers.
This kind of interaction also necessitates trust — trust regarding illicit drugs, alcohol, tobacco, and sexual activity, all of which can contribute to or cause certain medical problems. And a computer may ask the questions but cannot earn a patient’s confidence. After all, these kinds of secrets can only be exchanged between two human beings. David Eagleman, a neuroscientist at the Baylor College of Medicine, writes in his book Incognito that when we reveal a secret, we almost always feel that “the receiver of the secrets has to be human.” He wonders why, for example, “telling a wall, a lizard or a goat your secrets is much less satisfying.” As patients, we long for that human reception and understanding that a physician can provide and use to our advantage in coming up with a diagnosis.
Khosla neglects other elements of medical care, too. Implicit in his comments is the idea that the patient is a consumer and the doctor a salesman. In this setting, the patient buys health in the same way that he or she buys corn on the cob. One doesn’t need friendly banter or a packet of paperwork to get the best corn, only a short visit to the grocery store.
And yet, issues of health are far more serious than buying produce. Let’s take the example of a mother who brings her child in for ADHD medication, a scenario I’ve seen multiple times. “My child has ADHD,” she says. “He needs Ritalin to help his symptoms.” In a consumer-provider scenario, the doctor gives the mother Ritalin. This is what she wants; she is paying for the visit; the customer is king. But someone must explain to the mother what ADHD is and whether her child actually has this disorder. There must be a conversation about the diagnosis, the medication, and its side effects, because the consequences of these are lifelong. Machines would have to be more than just clerks. In many instances, they would have to convince the parent that, perhaps, her child does not have ADHD; that she should hold off on medications and schedule a follow-up to see how the child is doing. Because the exchange of goods in medicine is so unique, consequential, and rife with emotion, it is not just a consumer-cashier relationship. Thus computers, no matter how efficient, are ill-fitted to this task.
Khosla also misunderstands certain treatments, which are directly based on human interactions. Take psychiatry for example. We know that cognitive behavioral therapy and medication combined are the best treatment for a disease like depression. And cognitive behavioral therapy has at its core the relationship between the psychiatrist or therapist and the patient, who together work through a depressed patient’s illness during therapy sessions. In cognitive behavioral therapy, private aspects of life are discussed and comfort is offered — human expressions and emotions are critical for this mode of treatment.
To be sure, Khosla is right about quite a lot. Yes, technology ought to make certain aspects of the patient visit more efficient. Our vital signs may one day easily be taken with the help of our mobile phones, as he suggests, which would save time checking in to a clinic and could help give physicians constant and accurate measurements of blood pressure in hypertensive patients or EKG recordings in patients with heart disease. Technology of this sort could also indicate when an emergency is happening or how a patient ought to alter medication doses.
Furthermore, Khosla correctly identifies some of the limitations of human physicians: “We cannot expect our doctor to be able to remember everything from medical school twenty years ago or memorize the whole Physicians Desk Reference (PDR) and to know everything from the latest research, and so on and so forth.” True, the amount of information accumulated by modern medical research is beyond the capability of any human being to know, and doctors do make mistakes because they forget or are not up on the latest research. In a 2002 study in the Journal of Neurology, Neurosurgery and Psychiatry, investigators found that 15 percent of patients with a diagnosis of Parkinson’s disease do not necessarily fulfill criteria for the disease and 20 percent of patients with Parkinson’s disease who have already seen medical providers have not been diagnosed. These are large percentages that have profound implications for people’s lives. And this is exactly why physicians must use technologies like Watson to do a better job, not necessarily abdicate the job altogether. Most of us already carry smartphones or tablets on rounds, to look up disease processes or confirm our choice of antibiotic.
Lastly, Khosla wisely points out that physician bias can negatively affect a patient’s treatment. As he writes, “a physician’s bias makes all these personal decisions for patients in a majority of the cases without the patient (or sometimes even the physician) realizing what ‘preferences’ are being incorporated into their recommendations. The situation gets worse the less educated or economically less well-off the patient is, such as in developing countries, in my estimation.” Undoubtedly, this dilemma is real. I have spent many of my posts on this blog writing about the issue of remaining unbiased or level-headed in the face of difficult patient interactions. A study published in Obesity in 2013 found that physicians “demonstrated less emotional rapport with overweight and obese patients ... than for normal weight patients,” which may “weaken the patient-physician relationship, diminish patients’ adherence to recommendations, and decrease the effectiveness of behavior change counseling.” And as Tara Parker-Pope remarks in the New York Times, “studies show that patients are far more likely to follow a doctor’s advice and to have a better health outcome when they believe their doctor empathizes with their plight.” If bias exists in lieu of empathy, it makes sense that patients have worse outcomes. What makes doctors most valuable, their humanity, can have negative consequences.
But people can learn from studies, alter their behavior, and remain human. Computers or robots can learn from studies and alter their behavior, but they will always be robots. They will never earn the trust of the chronically ill drug addict. They will never be able to negotiate with the most difficult patients who demand specific treatments but may not be entirely sure why. An ideal system would not be one built solely on fallible human doctors but one in which new tools significantly augment human physicians’ skill and knowledge. A measured combination of these will put all the information at a doctor’s fingertips while keeping the art of medicine alive.