In the clinical trial, patient-reported outcomes around complications, postoperative pain, psychosocial well-being, physical functioning, fatigue, patient satisfaction and costs were assessed at 1 week, 3 months, 1 year, and 2 years postoperatively. The data set then used to train the algorithms was merely patient satisfaction after 2 years.
One major ongoing shortcoming of machine learning algorithms in many fields is that they are trained on data from primarily white populations. In the original clinical trial, however, race and ethnicity were key data points collected from the cohort and tracked with outcomes, which is rare, and means the subsequent machine learning algorithms presented in the abstract are less likely to suffer from racial and ethnic biases — a foundational need for the development of clinical tools.
In other clinical breast cancer settings, machine learning is also being tested as a tool to support better patient outcomes.2 Researchers in Switzerland have been developing an algorithmic tool to help physicians make data-driven, personalized decisions around which patients would most benefit from preventive treatments. The algorithm helps identify and target high-risk patients, while helping lower-risk patients avoid unnecessary prevention that could be costly, painful, and upsetting to go through. This is machine learning applied at the earliest, most cautious point of breast cancer care: prevention. The abstract presented at the ASCO20 meeting bookends the field by offering an algorithm at 1 of the final points of care: reconstruction.
Continue Reading
Ultimately, what the algorithms show is that the best outcome predictor is baseline satisfaction with breasts. But other informative predictors are whether the patient experienced “radiation during or after reconstruction, nipple-sparing and mixed mastectomy, implant-based reconstruction, chemotherapy, unilateral mastectomy, lower psychological well-being, and obesity,” according to the poster.
Group-level guidance often underpins clinical decisions around mastectomy and reconstruction when such a personal decision deserves a more personalized tool to figure out whether reconstruction is the right choice, and, if so, which reconstruction technique will result in the highest satisfaction for the patient.
References
- Pfob A, Mehrara B, Nelson J, Wilkins E, Pusic A, Sidey-Gibbons C. Towards data-driven decision-making for breast cancer patients undergoing mastectomy and reconstruction: Prediction of individual patient-reported outcomes at two-year follow-up using machine learning. Presented at: ASCO20 Virtual Scientific Program. J Clin Oncol. 38;2020(suppl):abstr 520.
- Ming C, Viassolo V, Probst-Hensch N, Chappuis PO, Dinov ID, Katapodi MC. Machine learning techniques for personalized breast cancer risk prediction: comparison with the BCRAT and BOADICEA models. Breast Cancer Res. 2019;21(7):75.
