Multimodal machine learning model prediction of complete pathological response to neoadjuvant chemotherapy in triple-negative breast cancer.

Abstract

601 Background: Triple negative breast cancer (TNBC) is a biologically and clinically heterogenous disease, associated with poorer outcomes when compared with other subtypes of breast cancer. In early-stage TNBC, surgery with curative intent remains the mainstay of therapy. Neoadjuvant chemotherapy is often given prior to surgery and achieving pathological complete response (pCR) has been associated with improved long-term outcomes in terms of progression-free survival (PFS) and overall survival (OS). There is thus high clinical interest in the ability to accurately predict pCR status using baseline data. Methods: A retrospective cohort of 57 patients with early-stage TNBC treated with neoadjuvant chemotherapy was analyzed to develop a machine learning-based algorithm predictive of pCR likelihood at the individual patient level. Multimodal baseline data were collected including clinical, biological (e.g., histology, genomic profile including CDC2, CDC20, KPNA2, MYBL2, complete blood count, Ki67), imaging data (baseline PET/CT scan), the radiology report and clinical outcomes data (pCR, PFS, OS). For each patient, tumors were segmented in 3D by an experimented nuclear physician using the SOPHiA Radiomics platform. Radiomics features were then extracted following the IBSI standards and then combined with the other data modalities. A filter-based variable selection method was applied before training several machine learning algorithms. The optimization criterion was the Area Under the ROC Curve (AUC). Due to the small size of the cohort, a nested leave-pair-out cross-validation approach was used to properly estimate the model performance. Results: The best result was obtained with the SVM algorithm with a linear kernel, reaching an AUC of 0.82, a precision of 71%, a sensitivity of 71% and a specificity of 70%. The features with highest weight in the algorithm were a mix of radiological, radiomics, biological and clinical features, highlighting the importance of a truly multimodal analysis. Indeed, withdrawing a specific data modality (e.g., radiomics features or biological features), led to a decrease of ̃10% of the AUC. Patients were then stratified into two groups based upon their predicted pCR status. These two groups displayed a statistically significant difference in PFS (p<0.001), suggesting that baseline multimodal data analysis could help predict long-term outcomes. Conclusions: This proof of concept study suggests that machine learning applied to baseline multi-modal data can help predict pCR status after neoadjuvant chemotherapy for TNBC at the individual patient level, as well as stratify patients to inform long-term outcomes. Patients that would be predicted as non-pCR could benefit from concomitant treatment with immunotherapy, or dose intensification. This algorithm will be further validated in a larger, multicentric cohort.