Background: CDK4/6 inhibitor in combination with endocrine therapy is a cornerstone of treatment for hormone receptor-positive metastatic breast cancer. We developed a machine learning model that used clinical, genomic, and transcriptomic features to stratify patients based on response to first-line CDK4/6 inhibitor treatment and identify predictors of response.

Methods: Patients with ER-positive/HER2-negative metastatic breast cancer, biopsy samples from before or <30 days after initiation of first-line treatment, curated clinical outcome data in a breast dataset from the Tempus real-world multimodal database, and data from targeted DNA sequencing (Tempus xT assay) and whole transcriptome sequencing (Tempus xR assay) were included. Driver DNA alterations associated with breast cancer were extracted at the gene and pathway levels. Cell type proportions and cell states were decomposed from bulk RNA sequencing data using EcoTyper. Genomic, transcriptomic, and clinicopathological features (age, tissue site, de novo disease, stage, early recurrence) and treatment characteristics (CDK4/6 inhibitor, endocrine therapy partner, adjuvant treatment) were used as explanatory variables in an integrative risk stratification model that used the OncoCast algorithm to predict real-world progression-free survival (rwPFS) in patients treated with first-line CDK4/6 inhibitors. Risk-set adjustment was used to mitigate the effect of patients’ delayed entry into the cohort due to the timing of their Tempus sequencing tests.

Results: Overall, 251 patients were included. Patients were stratified into two risk groups (high-risk [median rwPFS 11.2 months, 95% confidence interval [CI]: 7.2-17.6]; low-risk [median rwPFS 18 months, 95% CI: 14.9-not available]; log-rank p<0.001). Features associated with the high-risk group included higher rates of TP53 alterations, >4 metastatic sites, and high expression of genes involved in unfolded protein response, cell cycle, metabolic, proliferative, and androgen response pathways. Features associated with the low-risk group included high expression of genes involved in early estrogen response pathway.

Conclusion: Our study shows how multimodal real-world data collected during routine care can provide valuable insights into the biology of response to CDK4/6 inhibitor in patients with metastatic breast cancer and help improve patient stratification.

VIEW THE PUBLICATION

VIEW THE POSTER