Share this article on:

Case Study: How Health IT is Changing the Practice of Oncology—Using Data in New Ways

Butcher, Lola

doi: 10.1097/01.COT.0000425687.18243.ff

Memorial Sloan Kettering Cancer Center was one of the first cancer centers to adopt electronic medical records (EMR), and now, more than a decade later, it is pioneering new uses of technology to improve cancer care.

Earlier this year, lymphoma specialist Andrew Zelenetz, MD, PhD, was named Vice Chair of Medical Informatics in the Department of Medicine. In that role, he is harnessing data about cancer patients to improve treatment decision-making and find ways to make cancer care more efficient.

Back to Top | Article Outline

How can big data sets improve cancer care?

“There are different ways that you can use large datasets. If you have a specific hypothesis—for instance, that a new drug increases the duration of febrile neutropenia—then you can extract the patients getting the drug, identify an appropriate matched control, and ask if there is an association. Alternatively, with very large datasets you can use analytic tools to look for patterns in datasets and identify significant trends.

“At Memorial, we have very large data repositories with hundreds of thousands of encounters collected over many years. From these data, we can look for patterns that emerge from the way we treat patients. One example is trying to understand why there has been an increase in the hospital's average length of stay. A lot of it has to do with shifts to patients with higher acuity of care, but we also want to make sure there are not other changes in patterns of practice that are easily reversible.

“Some of these analytics are going to be used in a collaboration that Memorial has with IBM to re-educate IBM's Watson of Jeopardy fame to aid in medical decision-making. This is a large ongoing effort that started with breast and lung cancer and will be moving to other diseases in the coming year.”

Back to Top | Article Outline

This data comes from Memorial's electronic medical records?

“We are using not just EMR data, but admission and discharge data and a database that tracks diagnosis type. We have so many different computer systems—the situation is not unique to Memorial but is common in large institutions—like the financial systems, the institutional database, the cancer registry. So we are pulling in data from these disparate sources to try to get new insights by bringing them together. For example, we might see unique patterns if we bring together the cancer registry data with the admission and discharge data.

“The ability to look across databases is actually quite important. For several years, Memorial has had a tool called Darwin, which can access data from about 1.2 million inpatients and outpatients from the last two decades stored in disparate systems.

“There is a web-based interface that allows us to build queries and draw information about patients meeting the query parameters from multiple sources such as the pathology computer and the laboratory computer.

“For example, we used the Darwin system to ask questions about patients who receive rituximab: Do they have an increased risk of developing low immunoglobulin levels, and do they have an increased risk of developing infection? This search required us to look at all patients in the pathology computer who had the diagnosis of lymphoma, all patients in the pharmacy computer who had the administration of the drug rituximab.

“We had to look in the clinical laboratory system for all patients who had immunoglobulin levels tested, and we had to look at the electronic medical record system to see how many patients were admitted for infection or treated for infection. Darwin enabled us to combine data from these disparate sources for individual patients.

“We were able to demonstrate that, in fact, patients who are exposed to rituximab have a higher risk of developing low immunoglobulin levels and that those patients have a higher risk of developing recurrent infection.

“Without Darwin, there was no simple way to do this because it required us to pull data from so many different sources and then integrate them.”

Back to Top | Article Outline

How do Memorial's own datasets compare with the brain of Watson?

“The physician in charge of the Watson effort is Mark Kris, a lung cancer specialist here at Memorial. Watson is a massively parallel computer that can take multiple streams of information and try to understand them. The heart and soul of Watson is that it can read and interpret natural language. So if you give it a book, it can read the book and actually understand the book, to some extent.

“But Watson does not know how to synthesize information. That is what the collaboration with Memorial is all about. Watson actually has already read more of the world's literature about cancer than any human would ever be able to read in multiple lifetimes. As a result, Watson has a lot of book knowledge, but book knowledge is not necessarily clinical acumen. The Memorial input is to take experts in the field and basically develop algorithms that say, ‘This is actually how we use the data; this is how we value data.’

“But this is more than just embedding the expert. We actually learn from Watson as well because Watson has read more than we have. So it might sometimes come up with a recommendation that we have to say, ‘Is that the right thing it's recommending based on what it knows? Or do we have to tweak the algorithm because it actually made the wrong suggestion?’ Sometimes it will make a suggestion that we will say, ‘Oh, that is the right suggestion’ even though it is a surprise to us.”

Back to Top | Article Outline

When will Watson's expertise be available to oncologists outside Memorial?

“The hope is that within a couple of years, there will be a workable version. What might happen is that the Watson advisor—or whatever it is called by then—would be initially restricted to address metastatic lung cancer or adjuvant breast cancer and a few other diagnoses. But if it's a different clinical situation, you have to wait until the next iteration where that clinical situation is appropriate.

“Obviously the approach is to take the most important clinical situations and implement them first. As we go forward, we will add the less common clinical situations or those clinical situations in which Watson will have less of an impact on outcome.”

Back to Top | Article Outline

Part 2 of a Series

Part of a continuing look at how information technology is changing the practice of oncology.

Back to Top | Article Outline

iPad Exclusive!

PODCAST: Listen on the iPad edition of this article as Andrew Zelenetz, MD, PhD, Chief of the Lymphoma Service and Vice Chair of Medical Informatics at Memorial Sloan-Kettering Cancer Center, describes how oncologists are training the Watson computer—and how that work will help cancer patients beyond Memorial.

To receive our iPad issues, download the free Oncology Times app from the App Store today! Visit, search in the App Store, or follow the link on

© 2012 Lippincott Williams & Wilkins, Inc.
Home  Clinical Resource Center
Current Issue       Search OT
Archives Get OT Enews
Blogs Email us!