It was an amazing feat when Watson, IBM’s “cognitive” — listening, “thinking” and learning — computing platform handily beat two Jeopardy grand champions in February 2011, but it also left us wanting to know if, when and how the technology would be used in the real world. The world got its first glimpse of Watson at work in a commercial setting on Friday when IBM announced the release of three new health care decision-support applications.
The three new applications include one for recommending cancer treatment options and two for reviewing and authorizing treatments and related health insurance claims. They are the first examples of what IBM describes as a next-generation cognitive computing that has the potential to change healthcare, and IBM promises it’s just the beginning, as IBM and several partners are planning many more applications. IBM is also moving to roll out Watson in other information-intensive industries including banking, insurance and telecommunications.
The company started with healthcare and, within that field, chronic care and cancer because of the challenge. “We didn’t choose it because it was easy,” said Manoj Saxena, general manager of IBM Watson Solutions, at last week’s event. “We chose it because it has the most meaning and impact on society.”
The topics that IBM is addressing these days — using analytics to improve public services such as policing, run electrical utilities more reliably, and, most particularly, treat cancer more effectively with Watson — are a stark contrast to the preoccupations of many of its biggest rivals. IBM presciently sold its PC business to Lenovo back in 2004. Today, server rivals Hewlett-Packard and Dell are both mired in the question of how to move on either with or without the low-margin, commoditized PC business.
Oracle, meanwhile, is trying to stabilize the Sun server business — which saw its heyday in the dot-com era — by tying it closely to its successful and profitable software business with its Exa systems. That’s a sound business strategy, but it’s not exactly a grand vision for transforming industries or benefiting society. CEO Larry Ellison seems mostly interested in chest thumping about these engineered systems and taking pot shots at rivals IBM, SAP and Salesforce.com.; the closest thing to real vision is Oracle’s fairly new customer-experience push.
IBM teamed with Memorial Sloan-Kettering (MSK) in New York, one of the world’s top institutions for cancer treatment, to develop Interactive Care Insights for Oncology, the clinical decision-support application announced last week. A key difference between Watson and prior-generation decision-support tools, said executives from IBM and Sloan-Kettering, is that it keeps learning. It incorporates the latest symptoms and test results on individual patients and also the latest medical research and clinical trial outcomes. Over the last year, Watson has been trained on more than 600,000 pieces of medical evidence and two million pages of text from 42 medical journals and clinical trials in the field of oncology. Sloan-Kettering has added details on 1,500 lung-cancer cases, training the technology to interpret physicians’ notes, lab results and clinical research on specialized treatments based on the genetics of tumors.
“There has been an explosion in medical research, and doctors can’t possibly keep up,” said Dr. Mark Kris, chief of Sloan-Kettering’s Thoracic Oncology Service, who demonstrated the oncology application on an iPad at last week’s event.
The iPad app is akin to the patient’s chart, with light note taking and entry of new symptoms possible, but the heavy-duty entry of patient records and lab test results happens behind the scenes so it doesn’t interfere with a patient consultation. The doctor uses touch navigation to browse the latest symptoms and test results. The decision support is delivered as a prioritized list of recommended tests and treatment regimens together with confidence scores and links to supporting research.
In Kris’ demo, a lung cancer patient’s tumor is revealed to have a genetic marker that recent research says is not responsive to therapies usually prescribed in such circumstances. It was a fictitious demo example, but Kris said it highlights the very real opportunity to take advantage of the very latest advances in areas such as personalized care based on genetic research.
“We always rely on our colleagues to back us up on tough cases, and now we have an opportunity to consult with this new [digital] colleague to bring us additional information,” Kris said.
Sloan-Kettering plans to move beyond lung cancer into other forms of the disease in the months and years ahead. Meanwhile, IBM has other partners, such as Cleveland Clinic and Cedars-Sinai, and plans to move into medical training and other chronic-care areas, such as diabetes, heart disease and mental health.
On the payer side of health care, IBM partnered with WellPoint, one of the largest U.S. health insurance networks, operating in 14 states and covering one in nine Americans. This collaboration yielded the other two apps, WellPoint Interactive Care Guide and Interactive Care Reviewer, which are designed to help payers and providers, respectively, to quickly review claims and pre-authorizations and either approve or flag treatments as either appropriate or out-of-step with the latest, evidence-based treatment options.
WellPoint has trained the apps on 25,000 test-case scenarios and 1,500 live cases as part of a pilot project, and they are now in a learning mode. The Interactive Care Reviewer is being used by a handful of selected providers in the Midwest, and WellPoint expects 1,600 to be up and running on the application by year end. The key benefit is expected to be faster review times for pre-authorizations, moving away from faxes and phone calls to evidence-based decisions.
Of course other vendors, including business rules and analytics vendors, have been working on better, faster and more reliable healthcare utilization and adjudication apps, but Watson stands apart from systems that rely on rules and decision trees, according to IBM. “Those systems tend to be brittle and static, so it’s hard to constantly learn from new clinical trials and research,” said Stephen Gold, worldwide director of marketing for IBM Watson, in an interview with InformationWeek. “Those tools also tend to be deterministic, which means they try to get to ‘yes’ or ‘no,’ whereas Watson is probabilistic, so it gives you a degree of confidence in various choices based on the specific circumstances of the patient.”
It remains to be seen whether the Watson-based apps have an edge over analytics-based applications, from vendors such as SAS, or blended rules-and-analytics systems, such as those from Pegasystems. Healthcare is a $2 trillion industry worldwide, so it has attracted plenty of vendors to work on tough problems.
Watson stands on the shoulders of earlier achievements, including search technologies like Google and language-processing technologies such as Apple’s Siri. “But you can’t ask Google, ‘What is the appropriate initial dosage of antibiotic for a 7-year-old child with a history of asthma?'” said Saxena. At least you can’t and expect something more concise than a jumble of thousands of hits. The top search results might be on the right track, but the Watson applications are designed to deliver concise, evidence-based advice, such as, “two teaspoons, 80% confidence; one-and-half teaspoons, 45% confidence” accompanied by citations and links to supporting research. “It’s a deep, semantic question that has to be supported not just by keywords, but by actual medical evidence,” Saxena said.
To apply Watson to an industry-specific problem, IBM has over the last two years streamlined the code and focused the body of knowledge the technology has to work with. To play Jeopardy, Watson had to train on a smorgasbord of information ranging from history, literature and science to politics, the arts and pop culture. Instead of learning a broad range of idioms, slang and regional dialects, the technology had to target domain-specific jargon such as that used in the medical profession. As a result, the conference-room sized Watson that played Jeopardy has been shrunk down to a single server that can be scaled out on racks, according to IBM. The applications can be delivered as a cloud-based service, which is how The Maine Center for Cancer Medicine & Blood Disorders plans to use the Interactive Care Insights for Oncology app.
“Our state has real problems with access to care, particularly in its rural areas,” said Tracey Weisberg, medical oncology president at the center. “This will enable us to provide comprehensive, evidence-based treatment we could have only dreamed of in the past, yet our patients won’t have to leave the state and travel to a big-city hospital.”
The release of three Watson applications doesn’t guarantee success, but WellPoint executives said its nurses reviewing patient-care cases are already seeing 90% accuracy rates on the treatments advised by the Interactive Care Guide and Interactive Care Reviewer apps. Given the increasing number of new medical treatments and technologies, the complexity of managing multiple chronic diseases and the growing personalization of treatments and diagnostics, the challenge is to soak all that information in, retain it and “deliver practical evidence that clinicians and patients can follow,” said Dr. Samuel Nussbaum, WellPoint’s Chief Medical Officer. “That’s something Watson can address, and it will revolutionize the quality of care.”
IBM has its problems like many tech companies, including high-profile legal battles, low-margin businesses needing restructuring, and mature businesses facing long-term decline. But when it comes to efforts such as Watson, which is now the IBM Software Group’s single largest investment, you have to acknowledge that the company is operating on a higher level, setting more audacious goals than any company in IT.