AI May Further Expand Prediction of Immunotherapy Response

Machine learning may be used to predict disease progression in patients with advanced melanoma.

There is potential to predict immune checkpoint inhibitors (ICI) treatment outcome by using histology slides and patients clinical and demographic characteristics, according to a study from Cancer Clinical Research.

The purpose of this study was to create a more accessible way for oncologist to predict response to immunotherapy in patients with advanced melanoma.

Other recent studies that show other potential to predict response, but rely on biomarkers, scalability, and demand a high availability of resources, so these options aren’t always plausible.

“In this study we aim to develop a streamlined approach to pre-treatment prognostication by leveraging information immediately available through routine clinical care,” says the authors.

The research included two cohort groups, the first being a trained group of 121 patients who received treatment at New York University between 2004 and 2018. The other cohort is the independent group which had 30 patients who were treated at Vanderbilt University.

Researchers built a multivariable classifier that integrates neutral network predictions with clinical data, that was validates on two slide scanners. In the end the classifier correctly stratified patients into high versus low risk for disease progression. Vanderbilt patients with high risk had worse progression-free survival (PFS) than those with low risk.

\The results also showed that 50% of patients from the NYU cohort experienced progression of disease (POD) compared to 64% of patients from the Vanderbilt cohort. The majority of the NYU population received anti-CTLA-4 monotherapy and patients from Vanderbilt received anti-PD1-agents.

Researchers achieved their hypothesis; this study shows that there is a feasible way to predict immunotherapy response by combining neutral networks classification on histology slides with clinicodemographic information.

The researchers believe this model has potential to be integrated into clinical practice with more clinical trials on larger databases. They note that one of the biggest limitations in this study is lack of database, and that future studies should work on a larger scale.

The authors note, “In this setting, our fast and readily available approach could provide rapid first assessments to preselect candidates for treatment or identify those who require further analysis using complementary predictive models.”

For more news on cancer updates, research and education, don’t forget to subscribe to CURE®’s newsletters here.