PLoS ONE (Jan 2019)
Early experience with Watson for oncology in Korean patients with colorectal cancer.
Abstract
BACKGROUND:Watson for oncology (WFO) is a cognitive computing system providing decision support. We evaluated the concordance rates between the treatment options determined by WFO and those determined by a multidisciplinary team (MDT). METHODS:We reviewed the medical charts of patients diagnosed with colorectal cancer who visited the MDT at a single tertiary medical center from November 2016 to April 2017. WFO classified the treatment options for specific patients into three categories: 'Recommended', 'For consideration', and 'Not recommended'. Concordance rates between the WFO- and MDT-determined chemotherapy options, and the factors that potentially influence the concordance rate, were analyzed. RESULTS:Sixty-nine patients with colorectal cancer met with the MDT from Nov. 2016 to Feb. 2017. The mean age of the patients was 62 years (range: 34-86 years), and more patients were male (47/69) than female. Of the 69 patients, 51 (73.9%) were diagnosed with colon cancer, of whom 46.4% received the same regimen recommendation from WFO ('Recommended') as they did from the MDT. After inclusion of the 'For consideration' category from WFO, the concordance rate increased to 87.0%. The concordance rate between MDT and NCCN guidelines was 97.1%, and that between the WFO and NCCN guidelines was 88.4%. The concordance rates between WFO and MDT were significantly lower in patients with stage II, IIIC, or IV disease (P<0.001), and the colorectal cancer stage was the only statistically significant factor discriminating between WFO and MDT. CONCLUSIONS:The concordance rate between chemotherapy regimens for colorectal cancer determined by MDT versus WFO recommendations was 46.4%. After including the 'For consideration' category from WFO, the concordance rate increased to 88.4%. Further modification and improvement of the WFO prioritizing algorithm used to recommend treatment may increase the usefulness of WFO in the clinic.