Software Assistants for Doctors Are Making Progress

NY Times Bits story published 2/17/2013 by Steve Lohr

Doctors have long been in the high-stakes information management business. They must quickly sort through a patient’s symptoms, comments, test results, records and history to come up with a diagnosis. The physician brings to each diagnostic encounter a storehouse of knowledge and experience, all that he or she has read and learned over years.

The information overload for doctors is only growing worse. Medical information is estimated to be doubling every five years, and surveys show most doctors can find only a few hours a month to read medical journals. So it is not surprising that automated assistance for doctors has been pursued by researchers and companies for many years. Decision-making in medicine, after all, involves not just time and money, but also human lives.

A worthy goal, but a frustratingly difficult one. Yet in the last few years, real progress is being made in what is called “clinical decision support” technology. And the story in medicine is the same as in so many applications of modern computing: advances in sensors for measuring, calculating power and artificial-intelligence software are opening the door to a new generation of smarter tools.

The marquee example is I.B.M.’s Watson. After defeating human “Jeopardy” champions, the clever question-answering computer has moved on to medicine. It is working with oncologists at Cedars Sinai Cancer Institute in Los Angeles, and Watson is being trained as a medical student at the Cleveland Clinic Lerner College of Medicine of Case Western Reserve University.

But there are a flurry of other, smaller-scale examples from teams in universities and start-ups. The promising work of one such company, PeriGen, was published recently in the American Journal of Obstetrics and Gynecology and is being presented on Monday in San Francisco at the annual clinical meeting of the Society of Maternal Fetal Medicine.

The company, based in Princeton, N.J., specializes in fetal monitoring technology. The research being presented is an assessment by the federal National Institutes of Health, and the company’s software grew out of years of work by physicians and scientists at McGill University in Montreal. The team includes computer scientists, engineers, mathematicians and statisticians, and has been led by Emily F. Hamilton, an obstetrician and gynecologist who is senior vice president for clinical research at PeriGen, and an adjunct professor at McGill.

The software assessed in the N.I.H. study, called Pericalm Patterns, collects and analyzes the data from fetal heart monitors. Traditionally, doctors see the output of fetal heart monitors as wavy lines — called tracings — printed on paper that scrolls off the machine (roughly similar to the scribblings of lie-detector tests). Physicians are looking for patterns that might suggest the baby is in distress, perhaps suffering from oxygen deprivation. If so, they typically order a Caesarean section, to get the baby out quickly.

The trouble for obstetricians, Dr. Hamilton said, is that most babies, as they are about to be born, exhibit some unusual heart rate patterns. “You’re looking at all those tracings, and you’re applying rules of thumb and a lot of judgment,” she said in an interview over the weekend. “The challenge is to distinguish what is critical from what is just distracting.”

The N.I.H. assessment concluded that the analysis of three human experts agreed with that of the company’s software 97 percent of the time. The data used for this phase of the study was 100 tracings in the final hour before a baby was born. The technology can be used for research to identify characteristic tracing patterns, and in real time to give physicians alerts.

A second phase of the N.I.H. study, which is already under way, will involve 5,000 tracings — or about 20 miles of paper. “You can see why that cannot be done by humans. But we’ve reached the point now where we can use modern techniques of statistics and computing to really advance the science.”

Better science, Dr. Hamilton said, should pave the way for better, more consistent decision-making in obstetrics. On the one hand, she said, that should mean faster, clearer signals of a baby in distress. On the other hand, Dr. Hamilton said, it should also reduce unnecessary Caesarean-section deliveries. Faced with uncertainty, physicians tend to order a Caesarean to remove any risk to the baby, as well as the risk of malpractice suits.

While leveling off, the rate of Caesareans — at about 32 percent of all births in the United States — is twice as high as it was two decades ago.

Full Article

Download PDF

Written by