Watson, the IBM supercomputer that grabbed headlines after beating human contestants on Jeopardy!, has commercial applications in healthcare. And its developers have chosen to debut it in oncology.
Watson's first application in cancer is highly focused: recommending the best drug treatment options for patients with stage IV adenocarcinomas, said Mark Kris, MD, from the Memorial Sloan-Kettering Cancer Center in New York City, which is developing the tool with IBM and WellPoint, the managed care company.
"It's the biggest decision in lung cancer treatment," Dr. Kris toldMedscape Medical News. He explained that systemic therapies are the most common form of treatment, and adenocarcinomas are the most common form of lung cancer.
The supercomputer has synthesized an array of data gleaned from thousands of sources, including journal articles, national guidelines, individual-hospital best practices, clinical trials, and even textbooks.
Programming Watson for clinical decision-making in lung cancer "is not an easy task," said Dr. Kris. He pointed out that it took 4 years to program the computer so that it could participate in the game show.
Although Watson is a work in progress, it is currently being used at the Maine Center for Cancer Medicine and Blood Disorders, according to an IBM press statement.
Watson's hardware is physically located in Raleigh, North Carolina, and it uses cloud computing to interface with electronic medical records, Dr. Kris explained.
A clinician enters case notes into an electronic medical record and Watson immediately provides evidence-based and individualized treatment recommendations.
Individual treatment recommendations are updated as case notes become more detailed over time. The "natural language processing," which is the capability of the computer to read and comprehend case notes (in all their variety), is "the heart of Watson," he said.
IBM has produced an 8-minute video, entitled IBM Watson Demo: Oncology Diagnosis and Treatment, that illustrates a clinician interacting with Watson to receive treatment recommendations about a patient with lung cancer.
Watson can help human beings to do the increasingly complex job of being an oncologist, said H. Jack West, MD, from the Swedish Cancer Center in Seattle, Washington, and the author of the Blowing Smoke Medscape blog. He has been following the development of Watson and has commented onlineabout its potential usefulness.
"A computer can incorporate a nearly infinite amount of new data coming out, whereas the human brain can't attend to as much and integrate it," he told Medscape Medical News in an email.
Replacing Doctors?
Dr. West, who is not involved with Watson, sees 2 problems with its use in oncology: first, cancer is "defined by its variability"; and second, clinical decisions are often, in the end, matters of "judgement."
Even the most complex algorithms rely, to some degree, on "classic presentation," he said. Medical practice is less orderly, he noted. "The truth is that just about every cancer case is special and has some unique aspects to it, or at least there are so many extenuating factors that it's hard to be rigidly rule-based," he said.
Ideally, Dr. West would like to see a randomized controlled trial comparing clinical outcomes for patients treated with and without Watson. At the very least, he would like to see Watson's recommendations compared with those from several cancer experts to clarify its value over currently available second opinion.
Dr. West tips his hat to Watson's developers who have admitted that "there is no substitute for the most expert input when it is available." In other words, computers will never replace doctors, both Dr. West and the IBM team assert.
But a high-profile technology pioneer disagrees.
At the 2012 Health Innovation Summit in San Francisco California, Vinod Khosla, a cofounder of Sun Microsystems, reportedly said that computers could replace 80% of doctors. "Healthcare is like witchcraft and just based on tradition," he said. The solution? Data-based, computer-generated decision making.
Watson is "not a replacement" for physicians, according to Dr. Kris. In fact, it will take a "long, long time" to fully develop clinically, he said. However, he added Watson can improve on current oncology resources, such as the National Comprehensive Cancer Network guideline on nonsmall-cell lung cancer.
Dr. Kris, who is one of the authors of this guideline, pointed out that, for stage IV adenocarcinomas, bevacizumab and/or chemotherapy are recommended for patients with a performance status of 0 or 1. The advice is fairly nonspecific for chemotherapy. Two-drug chemotherapy regimens are "preferred," according to the guideline, but it does not specify which of the many of potential chemotherapies are best, nor does it address dosing or length of treatment, Dr. Kris notes.
"Guidelines are a scaffolding...but you've got to finish the building," he explained, referring to the relatively increased "granularity" of detail that Watson will provide and then improve upon as it is further developed.
Dr. Kris, commented on Watson in his Kris on Oncology Medscape blog last year when Memorial-Sloan Kettering entered into its partnership with IBM to develop the technology.
Watson is not the only electronic medical record-mediated computing initiative aimed at facilitating healthcare delivery. Major healthcare players, including McKesson, the American Society of Clinical Oncology, and Aetna, are developing clinical decision-making tools, said Dr. Kris.
In this study, they aimed to test that hypothesis in the much larger IMS-III trial.
IMS-III was a phase 3, randomized, open-label international trial comparing a combined intravenous and intraarterial approach to stroke treatment to standard intravenous tPA alone. The planned target for patient enrollment was 900 patients, randomly assigned in a 2:1 ratio to combined vs standard therapy within 3 hours of stroke onset.
Endovascular therapy included a choice of catheters and devices or intra-arterial tPA based on the lesion characteristics, the experience and training of the investigator, and the specified use of devices.
The trial was halted early for futility after enrollment of 656 patients. Full results were presented here by principal investigator Joseph P. Broderick, MD, from the University of Cincinnati at a session dedicated just to this trial. They were published online simultaneously in the New England Journal of Medicine, and reported by Medscape Medical News.
In the overall analysis, there was no statistically significant difference between the interventions on the primary endpoint, a modified Rankin Scale score of 2 or less at 90 days, indicating functional independence. The absolute difference between groups was 1.5% (95% confidence interval [CI], -6.1% to 9.1%), adjusted for stroke severity by National Institutes of Health Stroke Scale (NIHSS) score. A score of 8 to 19 indicates moderate to severe strokes, and a score of 20 or more indicates the most severe strokes.
In this new time analysis, Dr. Khatri and colleagues used the same methods as in their previous report, excluding patients with very large clots to make a more homogeneous population. Included were patients with M1, M2, or internal carotid artery terminus occlusions who were treated with endovascular therapy and in whom reperfusion was technically successful, defined as a Thrombolysis in Cerebral Infarction (TICI) score of 2 to 3.
Time to successful reperfusion was considered the time from symptom onset to the time the procedure ended, Dr. Khatri noted. Good clinical outcome was defined as a modified Rankin Scale score of 0 to 2, also the primary endpoint of IMS-III.
Of the 656 patients in the overall study, 240 had a qualifying occlusion for this analysis, among whom operators achieved TICI 2/3 reperfusion in 182 (76%). The mean time from symptom onset to reperfusion was 325 minutes (range, 180 to 418 minutes). The longest period was time from symptom onset to start of intravenous tPA, at 121 ± 34 minutes.
After adjustment for baseline variables (such as a baseline Alberta Stroke Program Early CT Score [ASPECTS] of 5 to 10, indicating a favorable computed tomographic scan; lack of disability prior to stroke; and an NIHSS score of 8 to 19 vs 20 or higher, indicating a less severe stroke among these moderately severe strokes), Dr. Khatri said, "we still had a relationship between time to reperfusion and outcome. Every 30-minute delay in reperfusion was associated with a 10% relative reduction in good outcome."
Table. IMS III: Predictors of Favorable Outcome (Modified Rankin Scale Score, 0 to 2)
| Variable | Risk Ratio (95% Confidence Interval) | P Value |
| Time to reperfusion (every 30-min delay) | 0.90 (0.82 - 0.99) | .02 |
| Baseline ASPECTS of 5 to 10 | 3.70 (1.25 - 11.00) | .01 |
| Lack of premorbid disability | 2.61 (1.05 - 6.50) | .01 |
| NIHSS score ≤19 vs ≥20 | 1.64 (1.07 - 2.51) | .01 |
They found the time relationship was maintained across ASPECTS groups, NIHSS strata, TICI score achieved, and type of occlusion.
One limitation is that by using the time to the end of the procedure as a surrogate for time to reperfusion, they may have overestimated the actual time to reperfusion, she noted.
Going forward, Dr. Khatri said, they plan to carry out more statistical analyses to help define any "point of no clinical return," where no benefit will be seen even if reperfusion is achieved, probably around the 7-hour mark after symptom onset.
Finally, she showed another analysis plotting the IMS-III results on their previous graph showing probability of good outcome against time to reperfusion, "just raising the possibility that if we had opened up clots faster, transitioned patients faster to IA [intraarterial] therapy, perhaps we would have had a positive trial."
IMS-III was funded by the National Institute of Neurological Disorders and Stroke (NINDS). Dr. Khatri reports significant support from NINDS as part of the IMS-III trial executive committee, significant support from Penumbra as neurology principal investigator for the THERAPY trial and from Genentech as the overall principal investigator for the PRISMS trial in planning stages, and modest support from Travek for Genentech as an unpaid consultant.
EmoticonEmoticon