A machine learning (ML) model that combines clinical and genomic data has shown superior performance in predicting the effectiveness of CDK4/6 inhibitors in patients with hormone receptor (HR)-positive, HER2-negative metastatic breast cancer. The findings, presented at the San Antonio Breast Cancer Symposium (SABCS) 2024, suggest the potential for more personalized treatment strategies.
Predicting Treatment Response
While CDK4/6 inhibitors combined with endocrine therapy have improved outcomes for patients with HR-positive, HER2-negative metastatic breast cancer, responses vary significantly. Dr. Pedram Razavi, Scientific Director at Memorial Sloan Kettering Cancer Center, noted that some patients respond well, while others develop resistance or derive no benefit. The goal is to identify patients who will benefit from CDK4/6 inhibitors at the time of metastatic diagnosis.
"There's a huge need in clinic to identify patients who may or may not benefit from adding CDK4/6 inhibitors at the time of metastatic diagnosis so that we can think about escalation and de-escalation strategies in advance," said Razavi. "More accurate prediction of outcomes could also help some patients avoid unnecessary side effects and financial toxicity from escalated upfront approaches."
Model Development and Performance
Currently, clinical features like treatment-free interval (TFI) are used to identify patients at high risk of early progression. Razavi and colleagues developed a multimodal ML model, OncoCast-MPM, to improve risk stratification. The model was trained using data from 761 patients with HR-positive, HER2-negative metastatic breast cancer who received first-line endocrine therapy with CDK4/6 inhibitors. Tumor sequencing was performed using MSK-IMPACT. The model's performance was tested on a holdout cohort of 326 patients.
The model generated three risk groups based on clinicopathological features (CF) and genomic features (GF). The integrated model (CGF) identified four risk groups. Median progression-free survival (PFS) varied significantly across these groups. For the CGF model, median PFS was 5.3 months in the high-risk group and 29 months in the low-risk group. The hazard ratio between the high- and low-risk groups was 6.5-fold higher in the CGF model compared to the CF and GF models, indicating superior patient stratification.
Key Predictors and Clinical Implications
The ML model selected clinical and genomic factors associated with outcomes or resistance to CDK4/6 inhibitors or endocrine therapy. Genomic predictors of poor outcomes included TP53 loss, MYC amplifications, PTEN loss, RTK-MAPK pathway alterations, RB1 loss, whole genome doubling, and high proportion of loss of heterozygosity. Major clinical predictors included liver metastasis, TFI less than one year, progesterone receptor negativity, low estrogen receptor expression, and presence of visceral metastasis.
"All of these variables are potentially available when the patients are diagnosed with metastatic disease, making such ML models broadly applicable," Razavi explained. "The hope is to integrate these models in clinical trial design of escalation and de-escalation strategies potentially transforming how we approach treatment for newly diagnosed metastatic disease."
Future Directions
Limitations of the study include its single-institution design, retrospective data analysis, and potential referral bias. Razavi's team is validating the model using external data sets and plans to develop an online tool for physicians to input clinical and genomic data and receive patient-specific outcome predictions. This could lead to closer disease monitoring and the use of liquid biopsies and tumor-derived biomarkers to inform second-line treatment options and clinical trials.
