A machine-learning model incorporating both clinical and genomic factors outperformed models based solely on either clinical or genomic data alone in predicting which patients with hormone receptor (HR)-positive, HER2-negative metastatic breast cancer would have better outcomes from adding CDK4/6 inhibitors to endocrine therapy as first-line treatment, according to results presented at the 2024 San Antonio Breast Cancer Symposium (SABCS; Abstract GS3-09).
Although the use of CDK4/6 inhibitors combined with endocrine therapy has significantly improved outcomes in patients with this type of breast cancer, Pedram Razavi, MD, PhD, Scientific Director of the Global Research Program at Memorial Sloan Kettering Cancer Center (MSK) and presenter of the study, noted that the responses to CDK4/6 inhibitors vary widely. Some patients do well, others develop treatment resistance over time, and some derive no benefit at all.
Pedram Razavi, MD, PhD
“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 we can think about escalation and de-escalation strategies in advance,” Dr. Razavi said. “More accurate prediction of outcomes could also help some patients avoid unnecessary side effects and financial toxicity from escalated upfront approaches.”
Study Details
Using OncoCast-MPM, a machine-learning tool developed at MSK, they generated three models to predict progression-free survival with CDK4/6 inhibitors: one based on clinicopathologic features, another on genomic features, and one integrating clinicopathologic and genomic features. The models were developed using a training cohort of 761 patients with HR-positive, HER2-negative metastatic breast cancer who received first-line endocrine therapy with CDK4/6 inhibitor combinations and had tumor sequencing performed prior to treatment or within 2 months of the start of treatment with MSK-IMPACT, a test designed to detect gene mutations and match patients to treatments or clinical trials for actionable cancer targets. The performance of the model was tested on a holdout test cohort of 326 patients.
The models trained on clinicopathologic features and genomic features each identified three risk groups: high, intermediate, and low, with median progression-free survival of 6.3, 15.2, and 24.5 months for clinicopathologic features and 9.9, 18.1, and 23.1 months for genomic features, respectively.
The integrated model identified four risk groups including two intermediate groups positioned between low- and high-risk categories. The median progression-free survival was 5.3 months in the high-risk group and 29 months in low-risk group, with the intermediate groups showing median progression-free survival of 10.7 and 19.8 months. Of note, the hazard ratio between the high- and low-risk groups was significantly higher in the integrated model (a 6.5-fold difference) compared with the clinicopathologic features and genomic features models (3- to 4-fold difference), indicating a superior stratification of patients with the integrated model. Testing the holdout test cohort yielded nearly identical progression-free survival and hazard ratio results.
“All three models performed really well, surpassing the conventional clinical risk models based on a single or a few clinical features. But the power of the analysis shone when we started combining the clinical and genomic features together,” Dr. Razavi said.
The clinical and genomic factors that were selected by the machine-learning model were mainly those known to be associated with outcomes or resistance to either CDK4/6 inhibitors or endocrine therapy. Key genomic predictors of poor outcomes were biologically plausible alterations such as TP53 loss, MYC amplifications, PTEN loss, RTK-MAPK pathway alterations, RB1 loss, whole genome doubling, and high proportion of loss of heterozygosity. The major clinical predictors included liver metastasis, treatment-free interval less than 1 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 machine-learning models broadly applicable. 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,” Dr. Razavi said. “Knowing that a patient on first-line CDK4/6 inhibitors is in the high-risk group could prompt the treating oncologists to implement closer disease monitoring and utilizing liquid biopsy and tumor-derived biomarkers to inform second-line treatment options and clinical trials. This could put us one step closer to staying ahead of breast cancer.”
Limitations of this study include its single-institution design, retrospective data analysis, and potential referral bias associated with specialized cancer centers. To address these challenges, Dr. Razavi and his team are validating the model using external data sets and aim to develop an online tool where physicians can input clinical and genomic data to receive patient-specific outcome predictions.
Disclosure: The study was indirectly and partially supported by the National Institutes of Health, U.S. Department of Defense, Susan G. Komen, Breast Cancer Research Foundation, AstraZeneca, Sophia Genetics, Novartis, and Tempus. Dr. Razavi has served as a consultant or advisor to AstraZeneca, Pfizer, Lilly/Loxo Oncology, Novartis, Tempus, Prelude Therapeutics, NeoGenomics, Natera, SAGA Diagnostics, Paige, Guardant Health, Myriad Genetics, and Foresight Diagnostics; and has received institutional grant/funding from Grail, Novartis, AstraZeneca, Biotheranostics, Tempus, NeoGenomics, Biovica International AB, Guardant Health, Personalis, Myriad Genetics, and Foresight Diagnostics.
