A machine learning model that incorporates clinical and genomic factors has outperformed models based solely on clinical or genomic data in predicting which patients with hormone receptor-positive, HER2-negative metastatic breast cancer will benefit from adding CDK4/6 inhibitors to first-line endocrine therapy, according to results presented at the 2024 San Antonio Breast Cancer Symposium (SABCS; Abstract GS3-09).
Pedram Razavi, MD, PhD, scientific director of the global research program at Memorial Sloan Kettering Cancer Center (MSK) and presenter of the study, highlighted the variability in responses to CDK4/6 inhibitors among patients. While the use of these inhibitors has significantly improved outcomes, some patients experience resistance over time, and others derive no benefit.
“There is a critical need to identify patients who may or may not benefit from the addition of CDK4/6 inhibitors at the time of metastatic diagnosis,” said Dr. Razavi. “A more accurate prediction of outcomes could help some patients avoid unnecessary side effects and financial toxicity from intensified initial approaches.”
The study utilized OncoCast-MPM, a machine learning tool developed at MSK, to create three models predicting progression-free survival with CDK4/6 inhibitors: one based on clinical-pathological features, another on genomic characteristics, and an integrated model combining both. These models were trained on a cohort of 761 patients who received first-line endocrine therapy with CDK4/6 inhibitors.
The integrated model identified four risk groups, with a median progression-free survival of 5.3 months in the high-risk group and 29 months in the low-risk group. Notably, the risk index between high and low-risk groups was significantly greater in the integrated model, indicating superior patient stratification.
“The three models performed exceptionally well, surpassing conventional clinical risk models,” Dr. Razavi stated. “The analysis's power increased when combining clinical and genomic features.”
The study's limitations include its single-institution design and potential biases. To address these, Dr. Razavi and his team are validating the model with external datasets and aim to develop an online tool for clinicians to input clinical and genomic data for personalized outcome predictions.