In handling some kinds of life-or-death medical judgments, computers have already have surpassed the abilities of doctors. We’re looking at something like promise of self-driving cars, according to Zak Kohane, a doctor and researcher at Harvard Medical School. On the roads, replacing drivers with computers could save thousands of lives that would otherwise be lost to human error. In medicine, replacing intuition with machine intelligence might save patients from deadly drug side effects or otherwise incurable cancers.
Consider precision medicine, which involves tailoring drugs to individual patients. And to understand its promise, look to Shirley Pepke, a physicist by training who migrated into computational biology. When she developed a deadly cancer, she responded like a scientist and fought it using big data. And she is winning. She shared her story at a recent conference organized by Kohane.
In 2013, Pepke was diagnosed with advanced ovarian cancer. She was 46, and her kids were 9 and 3. It was just two months after her annual gynecological exam. She had symptoms, which the doctors brushed off, until her bloating got so bad she insisted on an ultrasound. She was carrying six liters of fluid caused by the cancer, which had metastasized. Her doctor, she remembers, said, “I guess you weren’t making this up.”
She did what most people do in her position. She agreed to a course of chemotherapy that doctors thought would extend her life and offered a very slim chance of curing her. It was a harsh mixture pumped directly into her abdomen.
She also did something most people wouldn’t know how to do — she started looking for useful data. After all, tumors are full of data. They carry DNA with various abnormalities, some of which make them malignant or resistant to certain drugs. Armed with that information, doctors design more effective, individualized treatments. Already, breast cancers are treated differently depending on whether they have a mutation in a gene called HER2. So far, scientists have found no such genetic divisions for ovarian cancers.
But there was some data. Years earlier, scientists had started a data bank called the Cancer Genome Atlas. There were genetic sequences on about 400 ovarian tumors. To help her extract useful information from the data, she turned to Greg ver Steeg, a professor at the University of Southern California, who was working on an automated pattern-recognition technique called correlation explanation, or CorEx. It had not been used to evaluate cancer, but she and ver Steeg thought it might work. She also got genetic sequencing done on her tumor.
In the meantime, she found out she was not one of the lucky patients cured by chemotherapy. The cancer came back after a short remission. A doctor told her that she would only feel worse every day for the short remainder of her life.
But CorEx had turned up a clue. Her tumor had something on common with those of the luckier women who responded to the chemotherapy — an off-the-charts signal for an immune system product called cytokines. She reasoned that in those luckier patients, the immune system was helping kill the cancer, but in her case, there was something blocking it.
Eventually she concluded that her one shot at survival would be to take a drug called a checkpoint inhibitor, which is geared to break down cancer cells’ defenses against the immune system.
At the time, checkpoint inhibitors were only approved for melanoma. Doctors could still prescribe such drugs for other uses, though insurance companies wouldn’t necessarily cover them. She ended up paying thousands of dollars out of pocket. At the same time, she went in for another round of chemotherapy. The checkpoint inhibitor destroyed her thyroid gland, she said, and the chemotherapy was damaging her kidneys. She stopped, not knowing whether her cancer was still there or not. To the surprise of her doctors, she started to get better. Her cancer became undetectable. Still healthy today, she works on ways to allow other cancer patients to benefit from big data the way she did.
Kohane, the Harvard Medical School researcher, said similar data-driven efforts might help find side effects of approved drugs. Clinical trials are often not big enough or long-running enough to pick up even deadly side effects that show up when a drug is released to millions of people. Thousands died from heart attacks associated with the painkiller Vioxx before it was taken off the market.
Last month, an analysis by another health site suggested a connection between the rheumatoid arthritis drug Actemra and heart attack deaths, though the drug had been sold to doctors and their patients without warning of any added risk of death. Kohane suspects there could be many other unnecessary deaths from drugs whose side effects didn’t show up in testing.
So what’s holding this technology back? Others are putting big money into big data with the aim of selling us things and influencing our votes. Why not use it to save lives?