Looking forward to Dr. Mac-Dreamy

Of course, as my pediatrician mother and my brother, a second-year medical student, routinely remind me, real-life doctors usually devote a little more time to medicine and a little less time to frenzied trysts in the hospital on-call rooms. But their patients don’t seem to fare much better than those of their television counterparts.

Preventable medication errors cause more than 7,000 deaths each year, according to the Institute of Medicine (IOM), often because doctors mix up medications with similar names. Patients with a nasty case of foot fungus sometimes get a dose of the antiepileptic drug Lamictal instead of the antifungal drug Lamisil; patients with seizures may be prescribed the arthritis medication Celebrex or the antidepressant Celexa instead of the appropriate anticonvulsant, Cerebyx.

Every year in the United States, about 1,500 surgical instruments and objects are left inside patients during medical procedures. Overall, the IOM reports, between 44,000 and 98,000 Americans die every year because of preventable medical errors that cost the healthcare system nearly $38 billion annually.

The IOM described these statistics, as well as some suggestions for improving them, in a report forgivingly entitled “To Err is Human: Building a Safer Health System.” And perhaps therein lies the problem: humans.

Last week’s artificial intelligence (AI) column explored how people and computers learn and excel in different areas. Medicine, in many ways, is a field that is not ideally suited to the human mind. Would we really be surprised if Meredith Grey of “Grey’s Anatomy” left her scalpel in a patient’s abdomen given all the more interesting things she has to think about? Humans — both on TV and in real life — get distracted, tired, bored, drunk, nervous and careless. And sometimes we make mistakes just because, well, we’re human.

Computers, by comparison, are consistent, vigilant and focused. They can do the same mind-numbingly boring task over and over, 24 hours a day, without pausing to flirt with a cute neurosurgeon with perfectly coiffed hair or sneak in a quick nap. Due in large part to these abilities, computers already play a central role in monitoring patients’ vital signs and alerting doctors to critical changes in patient status.

Today, hospitals employ computers to a much greater extent than they did 20, or even 10 years ago, but many researchers say we have barely even scratched the surface of machines’ potential to contribute to healthcare. “The use of artificial intelligence in medicine is still in its infancy, which is sort of interesting because it was one of the very first fields where AI tried to help out 30 or 40 years ago,” said C. William Hanson, a professor at the Penn School of Medicine who taught a computer science course on medical informatics at Princeton from 2002 to 2005.

In the early days of AI, medical “expert system” computers were designed to provide diagnostic assistance to doctors. The Health Evaluation through Logical Processing (HELP) system, developed in the 1960s and ’70s, allowed physicians to tap into a vast medical database by entering patient data into the computer that would then provide guidance for clinical decisions and treatment plans. Similarly, the DXplain software of the 1980s generated a ranked list of diagnoses based on a patient’s symptoms and lab results. DXplain stored information for more than 2,000 different diseases: more facts than even the smartest, most experienced doctors are likely to be able to recall instantly.

Support nonprofit student journalism. Donate to the ‘Prince’. Donate now »

This isn’t to say that humans don’t have a place in hospitals, or even that computers make better doctors. Even the most optimistic views of what medical AI technology will be capable of in the coming decades usually concede that human physicians will still be in charge.

“Medicine is a very human-intensive industry. That’s one of the reasons medical costs are so high,” Hanson said. “I think that in the next five to 10 years we’ll begin to see networks evolve around the patients that will act relatively autonomously in monitoring their health. These machines are not going to control patient care, but they’re going to be intelligent assistants.”

As AI technology continues to improve, however, it’s less and less clear which domains of healthcare should belong to humans, both inside and outside the hospital.

In the future, intelligent machines will be able to assist the elderly in living independently for longer, former president of the Association for the Advancement of Artificial Intelligence Ron Brachman ’71 said.

Get the best of the ‘Prince’ delivered straight to your inbox. Subscribe now »

“With healthcare being as good as it is, people are living longer, and many of us live in a generation where we feel vigorously independent,” he said. Computers can help compensate for failing memory by keeping an eye on the elderly, observing their behavior and routines and helping them remember to take medications or turn off the stove.

“We’re really primed for this kind of assistance, I think,” Brachman said. “The baby boom generation is used to technology, and we absolutely, at all costs, want to stay out of nursing homes.”

Inside today’s hospitals, the importance of intelligent medical technology is immediately apparent in the variety of specialized machines whose functions have expanded dramatically in the past decade from suggesting diagnoses and monitoring patient heart rates to include providing therapy, analyzing medical images and performing surgery.

In 2002, a London clinic began offering depressed patients therapy sessions with an interactive software program called “Beating the Blues.” The system shows patients video clips of a set of case studies and then asks them questions about the characters in the clips to help them recognize and adjust negative modes of thinking that may be causing their depression.

Computers are also revolutionizing the field of medical image interpretation, which has traditionally been particularly susceptible to human error because skillful image analysis requires years of experience and practice. Programs such as FocalPoint — a system that examines pap smears for signs of cervical cancer — can learn to interpret medical images faster, more accurately and more reliably than a human doctor. Currently, FocalPoint screens about 5 million pap smear slides each year, or 10 percent of all slides in the United States. Still, this may be a technology best used in conjunction with human experts who can notice things that computers may not pick up on, Hanson said.

“Machines may be better at some kinds of pattern recognition than humans. They can learn to recognize trends pretty quickly,” Hanson said. “But humans think in very nonlinear ways, computers don’t have our intuitive ability to recognize patterns. They tend to perform best by churning through all the possibilities.”

In recent years, as the field of robotics advances, machines have also been turning up in the operating room to assist surgeons. The one-armed robot Penelope, developed in 2005, performs many of the tasks of a surgical scrub nurse. Trained by Columbia professor Michael Treat, Penelope can identify and hand surgeons the instruments they ask for and even keep track of how many tools are used and returned by the doctor to ensure that none are left inside the patient. Similarly, the robotic da Vinci Surgical System has limbs that can be programmed to cut patient tissues and even operate in synchrony with tissues moving in a predictable fashion, like a beating heart.

Robotic innovation may also pave the way to less invasive forms of surgery and clinical testing, Hanson said.

“The instruments of five or 10 years from now may operate from within the body,” he said. “You could put a robot in through the mouth, and it could perform surgery without even opening up the body. Computerized pacemakers already serve as little doctors inside people, constantly, vigilantly analyzing what’s going on in their patient’s heart, making diagnoses and treating all by [themselves].”

Hanson’s own research focuses on the development of an electronic nose, a robot that can analyze gases emanating from a patient’s skin or spit samples and detect pneumonia much faster and more easily than traditional tests.

So if computers are faster, more reliable, better focused and harder-working than, say, Meredith Grey, why are our hospitals still predominantly staffed by humans? Perhaps part of the explanation is that machines cannot yet reproduce all of a doctor’s many skills — though several people have suggested to me that they think robot surgeons might actually have better bedside manner than their human counterparts — but much of the reason lies in our own innate distrust of machines.

“There’s a huge conservatism in medicine that we shouldn’t be turning processes over to machines because humans are better at them,” Hanson said. “The hardest thing about high-level AI in medicine is figuring out how to combine what humans do well with what machines do well. We need both.”

The medical TV shows of the future may feature robots and computers in important supporting roles, but I’d like to think that Mac won’t completely edge out McDreamy. With computer assistance in the operating room, who know? He may have even more time to devote to his curls, which many of us feel are his real subspecialty.

This is the third in a series of articles examining current and emerging artificial intelligence technologies and their impact on today’s world.