Data Mining Cancers Helps Track Outcomes, Costs

5Feb - by aiuniverse - 0 - In Data Mining


To better study the outcomes and costs of caring for cancer patients and emphasize value, oncologists at Hackensack Meridian Health’s John Theurer Cancer Center, in New Jersey, teamed with data scientists and engineers eight years ago to create their own analytics program to mine patients’ electronic health records for real-world data.

Since then, oncologists have used their cancer outcomes tracking and analytics (COTA) program—which has been spun off into a private company—in numerous research efforts to track metrics from how many patients receive proper workups to how various cancer treatment regimens perform based on disease characteristics.

When building the system, the team embraced the concept of barcodes, said oncologist Stuart L. Goldberg, MD, the chief of the center’s Division of Outcomes and Value Research. The code on a can of tomato soup, for example, contains information on the item’s price, size and manufacturer. This enables the store to accurately know its inventory and what’s selling, and to control quality.

“Although the barcode system revolutionized the retail industry, medicine has lagged behind with a broad billing code system that fails to capture important characteristics needed to distinguish complex diseases, such as cancer,” Dr. Goldberg said.

Linking Disease’s Many Aspects

Dr. Goldberg and his team wanted to use COTA to establish barcodes for cancer that contain prognostic information such as tumor size, genetic characteristics, and whether the cancer metastasized. They set up the system to access doctors’ notes in the electronic health record to “barcode” the disease, allowing the user to determine how many patients had early-stage disease or a particular genetic subtype, for example, with a few clicks on the computer.

“COTA is currently looking at our charts, coding the diseases and giving our oncologists back organized information, so we can know which patients we are treating with a particular regimen, and we can link that with patient outcomes and costs,” Dr. Goldberg said.

One research focus has been looking to streamline therapies so all patients with the same subtypes of cancer receive the same care. They have submitted for publication a report on data from over 4,000 breast, colon and lung cancer patients that shows how unwarranted variations in care drive up costs without noticeable clinical benefits. 

Another recent project that made headlines was a partnership with IBM Watson for Oncology, in which the COTA program fed barcoded information about breast cancer patients to Watson, which, in turn, used that information to generate treatment recommendations. The team separately asked breast cancer specialists what they recommended. About 90% of the time, the experts and Watson for Oncology agreed on the therapies. In contrast, among oncologists who did not specialize in breast cancer, concordance fell to 75%.

“Watson for Oncology may be helpful in the community setting, where you have generalists who could benefit from added decision support,” Dr. Goldberg said. “I’d love to think that every doctor is perfect, but unfortunately the field of oncology changes so fast; and unless you’re really an expert in that one disease, it’s hard to keep up.”

Next on the horizon is looking at methods to better structure physician notes to enable COTA to conduct faster data extraction and generation of codes, he said. This could enable insurers to receive barcoded, accurate, detailed documentation sooner for preauthorizations “and eliminate time-consuming faxing of information back and forth between the insurer and the practice,” he said.

“The days of a handwritten, unintelligible paper chart that gets lost are over,” Dr. Goldberg said. “We have everything in the computer, and it is a rich source of real-world data. Now we have to mine that data to learn from every patient’s experience, so that we know what we are doing, what’s working, and what it costs. That will lead to value.”

Provider Discretion Preserved

Data mining as described here “can more easily help us determine best options for more individualized patient care,” commented Robert Mancini, PharmD, the bone marrow transplant pharmacy program coordinator at St. Luke’s Health System, in Boise, Idaho. “Right now, most guidelines are fairly broad and don’t always take into account things such as comorbidities or genetic characteristics in a detailed manner, which can impact responses to therapy.”

Oncology pathways that allow more patient-specific decision making based on specific disease types and stages are not universally accepted or agreed upon by clinicians, he added.

However, analytics programs need to integrate with—not replace—clinical decision making, Dr. Mancini cautioned. “While it is good to maximize outcomes and minimize cost, we want to be careful about taking away critical thinking and provider discretion.”

Facebook Comments