June 2018 Edition Vol.11, Issue 6

‘The Smartest Doctor’: Real-World Data Drive Decision Support Pilot Study

By John McCleery

An integration of data from two major clinical decision-support systems has the ambitious goal of bringing individualized clinical recommendations to the bed side for cancer patients.

A year into the pilot venture between Cota and IBM Watson for Oncology, development remains in a proof-of-concept stage, but both parties say the emerging data and clinical insights are helping win acceptance among oncologists and health care administrators.

“The combination of us together is very illuminating and bringing to light far more valuable information at the point of care than, I think, what has ever been done before,” Lisa Rometty, general manager of oncology, genomics and life sciences for IBM Watson Health, said in an interview during the American Society of Clinical Oncology (ASCO) meeting.


Integration of Real-World Data with Watson for Oncology Data

The venture brings to bear the growing decision support system of Watson for Oncology—encompassing the content of 300 medical journals, 200 textbooks, and 15 million journal articles and other texts. Knowledge gained from the literature has been augmented by training experience with oncologists at Memorial Sloan Kettering Cancer Center in New York.

“The smartest doctor there is,” Stuart Goldberg, MD, chief scientific officer of Cota, said of Watson for Oncology.

Goldberg’s company brings to the venture real-world, individual-patient data, extracted directly from electronic health record (EHR) systems. Cota processes fragmented and unstructured data into organized information that can be classified by multiple parameters to facilitate identification of associations across cohorts of patients. The data can be used to gain insights into practice patterns and evaluate the performance of treatments in clinical oncology practice as compared with clinical trials.

Integration of the Watson for Oncology and Cota data has the goal of bring an unprecedented level of personalization to clinical decision support for oncologists and their patients.

“The Cota real-world evidence gives you the ability to see how you’ve done with your patients,” said Goldberg. “What’s worked, what hasn’t worked, what doesn’t match what the experts are doing? Are there areas where you need knowledge? Are there areas where, even though you’re giving expert care, you’re not performing well because you might not have the nurses, might not have the support system?

“It can dissect that out so that you not only know whether you’re giving the right therapy, but you’re giving the right therapy effectively. Those are two different things.”

A Real-World Insights Influence Treatment

As an example of how the integrated data can provide insights and influence treatment decisions, breast cancer specialists evaluated records of actual patients and made treatment recommendations, which were compared with the care recommended by Watson for Oncology. The two sets of recommendations were in agreement for the vast majority of patients, said Goldberg.

The same patients’ data were then processed by the Cota system to see how patients with the same characteristics were being treated in clinical practice. The analysis showed a 20% variation from recommended care.

“They were getting bad therapy, therapy that was not recommended, therapy for which there was evidence that this was a bad thing to do,” said Goldberg. “That was in simple breast cancer cases, not complex cases. So, there is a definite need. A 20% difference means that 20% of women are not getting the best therapy, 20% wasted care, and hospital administrators will say that’s money thrown down the drain.”

How the Integrated Data Recommends Care

For the past year, the integrated decision support system has undergone testing and evaluation at Hackensack Meridian Health in New Jersey. The health system has the largest cancer center in the state, with a patient volume of approximately 20,000 per year, and the fifth largest bone marrow transplant program in the nation. The pilot program was limited to breast cancer but eventually will be expanded to include the four most common types of cancer.

The decision support system has been integrated into the Hackensack EHR, and the patients of the breast cancer specialists involved with the pilot program have been loaded into the system.

During the ASCO meeting, Cota and Watson for Oncology representatives gave OBR an exclusive demonstration of the integrated data system, exactly as oncologists at Hackensack see it on the EHR display. The case involved an actual patient included in the pilot program at Hackensack Meridian Health. Michael Hansen, offering manager for IBM Watson Health, described the background of the case.

“Watson has already ingested the patient record,” said Hansen. “Looking at surgical notes, radiology notes, clinical notes. Looking at the age of the notes and prioritizing certain notes over others. Looking over information in the unstructured part of the notes and prioritizing them by medical logic and looking for key attributes to get to a recommendation. That’s the foundational component.”

The basic characteristics of the patient:

  • 55-year-old female
  • Performance status = 1
  • Stage 2b breast cancer
  • ER/PR positive, HER2 negative
  • No prior treatment

If a key piece of information is missing, the presentation will pause, the missing information will be highlighted, and the clinician will be prompted to provide the information.

The attributes of the case are compared with those of patients who have similar characteristics and at the same stage of care. The display shows 95% concordance between the patient being evaluated and the historical cases. The system offers a link to clinical trials, so the treating physician can determine whether the patient might be a good fit for an ongoing clinical trial, either locally, regionally, or nationally.

Continued progression through the system brings the user to the standard-of-care treatment options for patients with the same characteristics as the current case. Various options are ranked in order. The top-ranked option consists of neoadjuvant chemotherapy, followed by surgery, followed by radiation and adjuvant endocrine therapy. Within each component of the recommended treatment strategy, the user can view additional options and recommendations, such as types of chemotherapy and surgery.



As a second option, the decision support system displays upfront surgery followed by evaluation for adjuvant therapy.

Hansen emphasized that the information displayed represented “recommended” care. The decision support system also tracks how patients are actually treated in clinical practice. Switching to a different screen shows that 99 of 150 patients—two thirds—with attributes identical to the current case had upfront surgery. One third received the recommended care, starting with neoadjuvant chemotherapy.

Patient information is updated continually during follow-up, which is reflected in each patient’s record and added to the cumulative database. Oncologist-users can see how each therapeutic option performed over time on a month-by-month basis, including progression-free survival and overall survival at 1, 3, and 5 years.

The integrated decision support system shows historical results with the recommended care for patients with the same characteristics as the clinical case. The system also shows how actual patients with the same characteristics fared in clinical practice with the care they actually received (recommended or otherwise). System users can see how their institution performed in comparison the overall dataset. Hansen said the system will soon provide performance data for individual physicians.

“It brings together the ‘here’s what you should do,’ based on everything known in the medical literature, with the ‘here’s what you’ve actually done,’ or ‘here’s how you practice medicine,’” said Cota CEO John Hervey. “It bridges the gap between the two and leads to ‘how do we make sure you pick the best treatment.’”

Proof of Acceptance?

Goldberg emphasized that the joint venture is not an attempt “to practice medicine.”

“This is giving physicians the information, so they can make the best decisions for their patients,” he said. “But then they have to take into consideration things such as the insurance situation at a particular hospital, the institutional philosophy on a number of issues, social support. Things that aren’t in the dataset.”

“It’s not ‘push a button and that’s what you’re going to give the patient.’ That’s not practicing medicine,” he added.

The system probably will be most valuable to general oncologists, Goldberg continued. Breast cancer specialists, for example, already know the best approach to care for patients with breast cancer. The situation is quite different for general oncologists, many of whom work in the community setting. They see patients with all types of cancer.

The growing complexity of cancer and the rapid evolution of knowledge and information make it virtually impossible for a general oncologist to remain abreast of the latest developments in care for multiple types of cancer, said Goldberg.

The cumulative dataset of the system continues to grow geometrically, said Hervey. The pilot program is continuing at Hackensack Meridian Health, but Cota and Watson for Oncology have plans to expand the program to additional centers.

Both organizations see signs of that the merger of the two types of data are gaining acceptance, among clinicians and health care systems.

“How do you convince doctors? You show them data, and we’re starting to get that,” said Hervey.

From the Watson for Oncology perspective, proof of acceptance has come from expansion of the Watson business.

“Today we’re in over 200 hospital organizations and 11 countries around the world,” said Rometty. “We still have a long way to go. We’re still early in our journey, but in our mind, we’ve been able to validate a viable value proposition.”

Post a Comment

OBR Archives

To view previous issues of OBR green you can visit our archives. The entire library of OBR green articles is searchable.