Watson for Oncology’s treatment recommendations for patients with breast cancer may largely accord with those made by multidisciplinary tumor boards, according to research published in Annals of Oncology.1
Despite rapid clinical advances, oncologists are spending about 4.6 hours per week maintaining up-to-date clinical knowledge, compared with 53 hours per week spent on care and administrative duties. Watson for Oncology, a clinical decision–support system, is designed to keep abreast of research likely to change clinical practice.
For this retrospective observational study, researchers evaluated whether Watson for Oncology treatment recommendations for 638 patients with breast cancer would have a high concordance rate with those made by a multidisciplinary tumor board at the Manipal Comprehensive Cancer Center in Bengaluru, India.
Continue Reading
Concordance was determined by how Watson rated the tumor board’s recommendation, which received 1 of 4 ratings: “recommended,” “for consideration,” “not recommended,” or “not available.” If Watson rated the tumor board’s recommendation as “recommended” or “for consideration,” the decisions were considered concordant.
The mean patient age was 52 years; 71% of patients had stage II or III disease; 41% of patients had hormone receptor–positive disease, 29% of patients had HER2-positive disease, and 30% of patients had triple-negative disease. More than 80% of patients presented with non-metastatic disease.
The tumor board made treatment recommendations between 2014 and 2016; Watson made recommendations in 2016. Without accounting for the time period in which the decision was made, the concordance rate between Watson and the tumor board was 73%.
Only 62% of Watson’s recommended strategies were, however, in concordance with the tumor board’s recommendation.
As the tumor board’s recommendations were made earlier, a second blinded review for the 27% non-concordant cases was conducted, which led to an improved overall concordance rate of 93%. Concordance was worse for recommendations regarding patients older than 45 years and was particularly low among those 75 years or older. Recommendations about patients with stage I disease were less likely to be concordant than those about patients with stage II or worse disease. Receptor status did not have a major impact on concordance.
The authors concluded that an “AI-based advisory system may have broad value in offering breast cancer treatment advice, particularly for environments where expert resources are not readily available.”
Reference
- Somashekhar SP, Sepúlveda MJ, Puglielli S, et al. Watson for Oncology and breast cancer treatment recommendations: agreement with an expert multidisciplinary tumor board. Ann Oncol. 2018 Jan 9. doi: 10.1093/annonc/mdx781 [Epub ahead of print]
