How does IBM Watson augment cancer treatment?

In 2015 we wrote a popular post explaining IBM’s Watson and its implications for the future of healthcare. Just two years later, we’re seeing a clearer picture of how Watson can augment cancer treatment decision-making.

In case you’re not familiar with Watson, we’re referring to IBM’s question answering computer system that famously beat out experts on Jeopardy! back in 2011. Watson uses a process called cognitive computing to understand complex language, which for the first time allows computers to make sense of the 80 percent of medical data considered “unstructured,” like journal articles or the patient narrative in an EHR.

While Watson partnered with healthcare organizations from early in the project's development, in April 2015 Watson Health made cognitive computing capabilities available to a wider array of researchers, doctors, patients, pharmaceutical companies and insurers. Building from an initial collaboration with Memorial Sloan Kettering Cancer Center starting in 2013, IBM launched Watson Oncology to provide evidence-based options that augment cancer treatment decision-making.

New publicly-shared data gives a progress report on Watson's role in healthcare. As MobiHealthNews reported earlier this month, a review of pilots at hospitals in India, Thailand, South Korea and Mexico found that Watson for Oncology achieved high concordance rates in multiple types of cancer.

To understand this finding, it’s helpful to step back and remember that Watson is essentially a tool to augment clinical decision-making. In most healthcare applications, Watson is fed a patient’s medical data and cognitive computing provides the medical team with a ranked set of treatment options along with rationale and supporting evidence for each approach. Just as many practitioners will get a second or third opinion to determine a diagnosis or treatment option, Watson is another source of information for practitioners.

In the context of Watson for Oncology, a concordance rate is basically a number that answers the question: “How often do Watson and practitioners at a healthcare institution come to the same conclusion?” So just how high are Watson for Oncology's concordance rates?

At a cancer center in Bangalore, India, Watson for Oncology matched the organization’s tumor board’s treatment recommendations in 96 percent of its lung cancer cases. Concordance with rectal cancer cases was 93 percent and colon cancer 81 percent. At a hospital in Bangkok, Thailand, Watson reached the same treatment decision as practitioners in 83 percent of cases across all cancer types.

These numbers are what IBM Watson was hoping for when it launched the Oncology program, according to deputy chief health officer Dr. Andrew Norden. As Norden told MobiHealthNews, “This is really exciting for us, of course, because what it means is this is really the beginning of showing really robust evidence of our solutions are working in practice.”

The evidence Norden is referring to is simply that Watson for Oncology is really good at providing treatment options that are consistent with those of experienced practitioners. The difference is that Watson can process more data than a medical team ever could. As IBM puts it, “The amount of research and data available to help inform cancer treatments is growing exponentially. Yet the time care teams have to consume this information—locating insights specific to each patient’s unique needs to potentially improve treatment outcomes—is more limited than ever.”

The new numbers also give even more teeth to Watson Health, which now has partnerships with 55 hospitals and health organizations around the world, and counting. The success of Watson for Oncology shows how cognitive computing can help physicians quickly identify key information in a patient’s medical record, surface relevant articles, and explore treatment options to reduce unwanted variations in care. Ideally, the time they save can be given back to their patients. Watson for Oncology is expected to support at least 12 cancer types by the end of this year.