The Promise of Cognitive Computing in Cancer Care

By Lynda Chin and Hagop Kantarjian

Computers have enhanced workplace productivity by automating calculations and programming processes based on preconfigured rules. Each rule dictates one right answer. However, computers are evolving and so are the ways we are using them.

For example, consider IBM Watson. The latest generation of Watson beat several top champions on the game show Jeopardy! by learning data and then analyzing the information to arrive at the “best” answer, rather than the “right” answer. In other words, Watson is not just a more powerful “Deep Blue” that beat the world chess champ Gary Kasparov. It is the world’s first cognitive computer.

Cognitive computers learn in a similar way as humans do. With proper “training,” they can address human-like situations that are characterized by ambiguity and uncertainty, and deal with pieces of data that change frequently, and which are often conflicting. Cognitive computers are designed to answer questions posed in conversational language with a range of possible “accurate answers” based on the available information. This capability is useful as a decision support system to help people extend their expertise across any domain of knowledge to make complex decisions.

This is also the case in the field of medicine. In medicine, there are often no black-and-white answers. The best answers are based on evolving and often ambiguous or even conflicting literature, colored by individual experiences or intuition. This process is often referred to as “The Art of Medicine.” Thus, a cognitive computer can become a valuable decision support system for physicians, particularly given the rapidly and ever expanding body of knowledge – so much so that even the most devoted physicians cannot possibly keep up to date with the amount of new information, much less be able to assimilate and apply it consistently in real time to the next patient they care for.

This is particularly true in the understanding and treatment of cancer. Major advances in cancer genomics revealed the complexity of the molecular causes and progressions of cancer. The same data also helped us define cancer as a heterogeneous collection of hundreds of diseases. The explosion in knowledge and the acceleration in its translation into new therapies and care paradigms mean a widening disparity between what is possible versus what is practiced.

Cancer-oriented cognitive computer systems can help in this regard. They are developed the same way new cancer specialists are trained. The computers are designed to read and remember the evolving and ever expanding body of medical literature. Then, they are trained by expert cancer specialists to weigh a patient’s case against this knowledge, and to suggest appropriate treatment options, including clinical trials, tailored to the individual patient.  Patient management advisories can be developed within cognitive computers that advise physicians of the potential side-effects of particular therapies, so they can proactively intervene to deliver the best possible outcome. Importantly, evidence supporting these suggestions is available for review by physicians, allowing them to judge its relevance. In other words, cognitive computers do not make decisions; rather, they offer physicians the tools to help them tailor the best treatments to an individual patient.

Cancer-oriented cognitive computers can increase penetration of the most current cancer treatment knowledge into worldwide cancer communities. General oncologists or non-cancer physicians may utilize such cognitive computers to gain immediate access to state-of-the-art management and treatment guidelines and care. These computers can potentially deliver the best evidence-based care to patients through sharing knowledge and expertise, ensuring that patients have access to equitable care, no matter who and where they are.

Believing in the potential of this new way of sharing and collaborating, here and around the world, several institutions — including Memorial Sloan Kettering Cancer Center and MD Anderson Cancer Center — have partnered with IBM Watson to develop cancer-oriented cognitive clinical decision support systems.

At MD Anderson, the Oncology Expert Advisor system (OEA) has been introduced into clinical validation. Because cancer-oriented cognitive computers have never been utilized in a clinical setting, part of the challenge is determining how a cognitive system should be evaluated. In addition, since it is designed to “read, remember, recommend and remind,” it could also drive continuous improvement in quality and safety of the care delivered to patients. The lessons learned in this endeavor may create a benchmark for developing and evaluating future cognitive systems for clinical applications.

To harness the full potential of cognitive computing, accurate and complete clinical data on a patient is crucial. Several organizations are investing in platforms to aggregate clinical and genomic data. These include CanSAR, CancerLinQ (created by the American Society of Clinical Oncology), and Flatiron, among others. The ability to extract valuable insights from these complex data using advanced analytics like cognitive computing will drive more complete and improved data collection in care settings.

Guest author Lynda Chin, M.D., is chair of the Department of Genomic Medicine, Division of Cancer Medicine, and scientific director of the MD Anderson Institute for Applied Cancer Science at The University of Texas MD Anderson Cancer Center.

Hagop Kantarjian, M.D., is chair of the Leukemia Department at The University of Texas MD Anderson Cancer Center and a nonresident fellow in health policy at Rice University’s Baker Institute for Public Policy.

This material may be quoted or reproduced without prior permission, provided appropriate credit is given to the author and Rice University’s Baker Institute for Public Policy. The views expressed herein are those of the individual author(s), and do not necessarily represent the views of Rice University’s Baker Institute for Public Policy.