Current Projects

A prediction model for pathologic complete response following neoadjuvant chemotherapy in breast cancer

The optimal patient characteristics to recommend neoadjuvant chemotherapy (NAC) among breast cancer patients is an active area of clinical research. This research team developed and compared several approaches to developing prediction models for pathologic complete response (PCR) among breast cancer patients in Alberta.

The study included all breast cancer patients who received NAC in Alberta between 2012 and 2014. PCR was identified by a surgeon who reviewed the pathology report after NAC. Three types of prediction models for PCR were built: 1) ‘literature model’: variables identified in through our systematic literature review (SLR); 2) ‘expert model’: variables selected based on oncologists’ opinions 3) ‘machine learning model’: all available variables from the data using least absolute shrinkage and selection operator (LASSO) methods. Each model was compared using area under the receiver operating characteristic curve (AUC).

A total of 363 breast cancer cases were included in the analyses, of which 86 experienced a pCR. In comparing the modeling approaches, logistic regression generally produced the highest AUCs. The expert model (SVM) and the machine learning model (random forest) produced similar AUC (0.796 vs. 0.801). The expert model included age at surgery, T stage, surgery type, response to trastuzumab, ER/PR status, HER2 status and geographical zone. Age and HER2 positivity were the strongest predictors.

This approach yielded a prediction model of modest predictive ability using routinely collected data. Despite having a slightly higher predictive ability, the machine learning approach may not be of value based on the additional burden of data collection. Several biomarkers of interest identified in the SLR were not available for this analysis.