In 2011, a supercomputer won $1 million on Jeopardy! In 2016, that same supercomputer is tackling a challenge quantified not in millions of dollars but in millions of cancer patients.
The goal is to use Watson’s natural language processing to mine the medical literature and a patient’s records to provide treatment advice. And this month the Watson computer system is drastically expanding its reach—from one hospital in Thailand to six in India and a planned 21 more in China.
This instantiation of Watson, dubbed Watson for Oncology, is an artificial intelligence system that has access to millions of pages of medical textbooks and journal articles. Oncologists at Memorial Sloan Kettering Cancer Center trained the system to provide appropriate treatment recommendations by giving it descriptions of patients and then telling Watson the correct treatment.
“[It’s] the same way we would train a young doctor by showing how our specialists would treat a certain cancer in a certain person,” said Dr. Mark Kris, a Memorial Sloan Kettering lung cancer specialist who’s leading the hospital’s collaboration with IBM.
Watson looks at records from a patient, like doctors’ notes and lab test results, and gives opinions about what treatment to pursue. Those opinions are backed up by easily accessible evidence—doctors can click on the recommended treatment and see which medical studies support it.
Preliminary studies show Watson is fairly good at recommending treatment. When examining different patients, Watson agreed with Memorial Sloan Kettering doctors over 90 percent of the time in one study, and 50 percent of the time in another. These studies were presented at a 2015 oncology meeting, and the abstracts were published, but the studies themselves are not. Authors include IBM-affiliated scientists.
Doctors at Bumrungrad International Hospital in Thailand have been using the system for about a year now, and oncologists at six hospitals in India started using Watson a few weeks back.
One of those oncologists is Dr. Amit Rauthan, who practices at Manipal Hospital in the southern Indian city of Bangalore. Watson is fully integrated into the computer system at the hospital, so soliciting Watson’s opinion takes a few seconds and a few clicks.
In an interview with STAT, he said he has used Watson for about two dozen patients. Most of the time, he said, Watson affirms that the decisions he had already come to were correct, such as one breast cancer patient who needed aggressive chemotherapy.
Other times, Watson disagrees, and the doctors change their mind. One of Rauthan’s patients had stage 2 colon cancer, and Rauthan thought that it would need chemotherapy. Watson recommended skipping the radiation and observing the patient—and its opinion won the day.
Rauthan emphasized that Watson itself didn’t make the decision. In the case of the colon cancer patient, Watson provided Rauthan with evidence about why chemotherapy might not be so beneficial, which he was able to share with the patient.
However the information about what happened to Rauthan’s patients—whether the chemotherapy helped the breast cancer patient, for instance—isn’t presently used to help Watson improve going forward, Kris said.
“That’s the ultimate goal—that it will learn from that experience,” Kris said. But that goal is years away, and one of the hurdles to achieving it is making sure that Watson can access and understand all of the information it needs about the patient—the complete medical record.
IBM announced last week that it was expanding the program to China, though the exact rollout date is some years in the future. Part of the utility of Watson is to “democratize access to cancer care,” said Dr. Kyu Rhee, chief health officer for IBM Watson Health, reaching areas that don’t have easy access to information. Cancer is the leading cause of death in China, killing an estimated 2.8 million people in 2015.
Dr. Qunwei Chen, an oncologist in eastern China whose hospital is adopting the system, said he anticipated that Watson for Oncology will be useful, especially for doctors in small hospitals, where oncologists have less access to conferences at which they might learn about new treatment guidelines and procedures.
IBM declined to say how much money who was paying whom for the use of Watson in China.
The importance of data
Because Watson for Oncology is a machine learning system, its outputs are only as good as its inputs. If it can’t read the doctor’s handwriting, it doesn’t know what the doctor tried to do. So a key next step is making sure that the data feeding into Watson is really good, which may prove a challenge.
“The vast majority of patient data gathered and stored in electronic health records is totally unstructured,” said Zach Weinberg, president of Flatiron Health, a medical software company. One of the things his company is doing is trying to organize this data in a more useful way.
Weinberg said that IBM is tackling “one interesting piece of the problem”—helping physicians make treatment decisions—but that machine learning algorithms are never going to be good enough to know everything a physician would know. (Weinberg’s company has been in touch with IBM, but has no formal collaboration.)
“Machine learning is a piece of the puzzle that doesn’t solve everything,” Weinberg said. “Sometimes it can be presented as a cure-all solution to all of our problems. That’s not exactly true.”
IBM’s Rhee concurs that doctors aren’t going anywhere. Watson, he said, serves as “a trusted advisor.”
There’s no published research yet on how Watson’s advice impacts patients’ health, and oncologists in India say that it’s too early to tell how it is even impacting their decision-making process.
That hasn’t stopped hospitals from promoting Watson. Manipal Hospital has even taken the system one step further, allowing patients to seek a Watson consultation remotely. The person can upload their medical records, pay 9,500 rupees (about $140) and then receive a PDF report of Watson’s analysis of their case, verified by Manipal physicians, along with an e-consultation with a Manipal doctor. The report does not come directly from Watson, an IBM spokesperson said.
Rauthan said that Watson opinions are free for patients who come into the hospital.