Predictive Tool for Melanoma Could Guide Immunotherapy Choices

Author(s):

Dana-Farber Cancer Institute

Key Takeaways

A machine learning model predicts resistance to anti-PD-1 therapy using genomic heterogeneity and ploidy, validated in four cohorts.
The model employs a decision tree to identify patients likely to resist single-agent immune checkpoint blockade.
In a small patient group, 43% predicted resistant to single-agent therapy responded to combination therapy, suggesting clinical utility.
Future research may use this tool with clinical sequencing tests to guide therapy decisions and improve outcomes for melanoma patients.

Genomic heterogeneity and ploidy identify patients with intrinsic resistance to PD-1 blockade in metastatic melanoma.

David Liu, MD, MPH

In previous research, Dana-Farber researchers found that a tumor’s genomic heterogeneity, meaning the tumor’s propensity to develop new mutations, and low genomic ploidy, a measure of the number of chromosomes in the cells, predicts resistance to immune checkpoint blockade with anti-PD-1 therapy.

This study builds on that research by developing a predictive machine learning model that uses genomic heterogeneity and ploidy to predict resistance to anti-PD-1 therapy. The team of clinical investigators and computational biologists refined and validated the model in four cohorts including two clinical trials. The interpretable machine-learning algorithm employs a simple decision tree to robustly predict which patients are likely to resist immune checkpoint blockade with anti-PD-1 therapy.

Further evaluation in a small group of thirteen patients revealed that 43% of patients predicted to be resistant to single agent therapy responded to combination therapy, suggesting that the tool could benefit patients. In future research, this tool could be used in prospective studies with existing clinical sequencing tests to determine if it helps guide therapy choices and improves outcomes for patients with melanoma.

Immune checkpoint inhibitors have dramatically improved outcomes for patients with advanced melanoma. Options for patients include single-agent immune checkpoint blockade or a combination of two immune checkpoint inhibitors. The combination therapy yields higher response rates but with significant added toxicity.

There are currently no biomarkers to help physicians choose between the two options. This study describes a predictive tool that can be used to choose between these two options by predicting which patients are likely to fare poorly with single-agent therapy and therefore might benefit from combination therapy.

This research was funded by the National Institutes of Health, Bristol Myers Squibb, Doris Duke Charitable Foundation, and the American-Italian Cancer Foundation.