Meeting News

Cognitive computing platform may help oncologists deliver subspecialist-level care

NEW YORK — A cognitive computing platform produced ranked treatment options that were largely concordant with expert opinions, according to study results presented at Chemotherapy Foundation Symposium.

However, real-world patients with similar characteristics received nonrecommended therapies in approximately 20% of cases, suggesting a need for decision support, researchers wrote.

IBM Watson for Oncology — a cognitive computing point-of-care system — provides evidence-based, confidence-ranked treatment options for individuals with cancer.

The platform uses structured and unstructured data from medical literature, medical records, treatment guidelines, and imaging, laboratory and pathology reports.

The system draws on data from the Cota Observational Database, which contains clinical, treatment, outcome and demographic data from nearly 15,000 patients with breast cancer treated at 30 cancer centers in the United States.

Stanley E Waintraub, MD, FACP, breast cancer division co-chief at John Theurer Cancer Center at Hackensack University Medical Center, and colleagues aimed to assess whether Watson for Oncology with Cota RWE — a point-of-care decision support tool that ingests patient attributes from the electronic health record, facilitates comparisons among similar patients and ranks treatment options as “recommended,” “for consideration” or “not recommended” — could help oncologists deliver subspecialist-level care.

Investigators initially presented EHR-derived data from 88 postmenopausal women with breast cancer to three breast oncologists at John Theurer Cancer Center. Without guidance from Watson for Oncology, clinicians were asked to record their recommended treatment strategies for each case, including choice of chemotherapy or hormonal agent.

The breast cancer specialists reviewed a combined 223 cases.

Their treatment recommendations matched acceptable Watson for Oncology strategies in 87.9% of cases. These included 175 cases (78.5%) in which specialists chose Watson for Oncology’s recommended option, and 21 cases (9.4%) in which the specialists chose one of Watson’s “for consideration” options.

The specialists chose “not recommended” treatment options in 27 cases (12.1%).

Seven (8%) of the 88 cases accounted for 59% of nonconcordant responses, with at least two specialists disagreeing with Watson for Oncology in each case.

The characteristics of each of the 88 test cases then were matched to generate a comparison cohort of similar patients from the real-world Cota observational database.

Researchers determined 69.3% of matched historical controls were most commonly treated with Watson for Oncology’s recommended treatment strategy. Patients received “for consideration” treatment in 10 cases (11.4%) and “not recommended” treatment in 17 cases (19.3%).

“Thus, the historical patients — treated in both academic and nonacademic centers — had a trend toward more nonrecommended therapies compared to breast cancer experts at John Theurer Cancer Center,” Waintraub and colleagues wrote.

Researchers then assessed the influence of Watson for Oncology with Cota real-world experience data support on clinical decision-making among clinicians who do not specialize in breast cancer.

A selection of cases were presented to solid tumor oncologists or hematologic malignancy oncologists at John Theurer Cancer Center, with or without the Watson for Oncology/Cota RWE decision-support tool.

Solid tumor oncologists reviewed a combined 105 cases. They selected a recommended treatment option 66.7% of the time without Watson for Oncology/Cota RWE support and 86.7% of the time with the support tools. The percentage of cases in which they chose a nonrecommended treatment declined from 24.8% to 8.6% with use of the support tools.

Hematologic malignancy oncologists reviewed a combined 234 cases. They selected a recommended treatment option 60.7% of the time without Watson for Oncology/Cota RWE support and 88.5% of the time with the support tools. The percentage of cases in which they chose a nonrecommended treatment declined from 24.4% to 3% with use of the support tools.

“When presented with [Watson for Oncology and Cota real-world experience] support, generalist oncologists might be able to improve their treatment to match that of expert subspecialists,” Waintraub and colleagues wrote. – by Mark Leiser

For more information:

Waintraub SE, et al. Can the cognitive computing system Watson for Oncology with Cota RWE help oncologists deliver subspecialist-level care? Presented at: Chemotherapy Foundation Symposium; Nov. 7-9, 2018; New York.

Disclosures: Watson Health IBM and Cota Inc. supported this study.

NEW YORK — A cognitive computing platform produced ranked treatment options that were largely concordant with expert opinions, according to study results presented at Chemotherapy Foundation Symposium.

However, real-world patients with similar characteristics received nonrecommended therapies in approximately 20% of cases, suggesting a need for decision support, researchers wrote.

IBM Watson for Oncology — a cognitive computing point-of-care system — provides evidence-based, confidence-ranked treatment options for individuals with cancer.

The platform uses structured and unstructured data from medical literature, medical records, treatment guidelines, and imaging, laboratory and pathology reports.

The system draws on data from the Cota Observational Database, which contains clinical, treatment, outcome and demographic data from nearly 15,000 patients with breast cancer treated at 30 cancer centers in the United States.

Stanley E Waintraub, MD, FACP, breast cancer division co-chief at John Theurer Cancer Center at Hackensack University Medical Center, and colleagues aimed to assess whether Watson for Oncology with Cota RWE — a point-of-care decision support tool that ingests patient attributes from the electronic health record, facilitates comparisons among similar patients and ranks treatment options as “recommended,” “for consideration” or “not recommended” — could help oncologists deliver subspecialist-level care.

Investigators initially presented EHR-derived data from 88 postmenopausal women with breast cancer to three breast oncologists at John Theurer Cancer Center. Without guidance from Watson for Oncology, clinicians were asked to record their recommended treatment strategies for each case, including choice of chemotherapy or hormonal agent.

The breast cancer specialists reviewed a combined 223 cases.

Their treatment recommendations matched acceptable Watson for Oncology strategies in 87.9% of cases. These included 175 cases (78.5%) in which specialists chose Watson for Oncology’s recommended option, and 21 cases (9.4%) in which the specialists chose one of Watson’s “for consideration” options.

The specialists chose “not recommended” treatment options in 27 cases (12.1%).

Seven (8%) of the 88 cases accounted for 59% of nonconcordant responses, with at least two specialists disagreeing with Watson for Oncology in each case.

The characteristics of each of the 88 test cases then were matched to generate a comparison cohort of similar patients from the real-world Cota observational database.

Researchers determined 69.3% of matched historical controls were most commonly treated with Watson for Oncology’s recommended treatment strategy. Patients received “for consideration” treatment in 10 cases (11.4%) and “not recommended” treatment in 17 cases (19.3%).

“Thus, the historical patients — treated in both academic and nonacademic centers — had a trend toward more nonrecommended therapies compared to breast cancer experts at John Theurer Cancer Center,” Waintraub and colleagues wrote.

PAGE BREAK

Researchers then assessed the influence of Watson for Oncology with Cota real-world experience data support on clinical decision-making among clinicians who do not specialize in breast cancer.

A selection of cases were presented to solid tumor oncologists or hematologic malignancy oncologists at John Theurer Cancer Center, with or without the Watson for Oncology/Cota RWE decision-support tool.

Solid tumor oncologists reviewed a combined 105 cases. They selected a recommended treatment option 66.7% of the time without Watson for Oncology/Cota RWE support and 86.7% of the time with the support tools. The percentage of cases in which they chose a nonrecommended treatment declined from 24.8% to 8.6% with use of the support tools.

Hematologic malignancy oncologists reviewed a combined 234 cases. They selected a recommended treatment option 60.7% of the time without Watson for Oncology/Cota RWE support and 88.5% of the time with the support tools. The percentage of cases in which they chose a nonrecommended treatment declined from 24.4% to 3% with use of the support tools.

“When presented with [Watson for Oncology and Cota real-world experience] support, generalist oncologists might be able to improve their treatment to match that of expert subspecialists,” Waintraub and colleagues wrote. – by Mark Leiser

For more information:

Waintraub SE, et al. Can the cognitive computing system Watson for Oncology with Cota RWE help oncologists deliver subspecialist-level care? Presented at: Chemotherapy Foundation Symposium; Nov. 7-9, 2018; New York.

Disclosures: Watson Health IBM and Cota Inc. supported this study.

    See more from Chemotherapy Foundation Symposium