Researchers Develop AI Model to Predict ICI Response in Advanced Melanoma

November 20, 2020
Allison Inserro
Allison Inserro

Results of the artificial intelligence (AI) model were consistent no matter which immune checkpoint inhibitor (ICI) therapy patients received, suggesting that some biomarkers are not necessarily specific to the checkpoint target.

Immune checkpoint inhibitors (ICI) revolutionized the treatment of advanced melanoma, but there is no easy way to predict which patients will respond or be at risk for immune-related toxicities without RNA sequencing that is generally not available at the level of community oncology clinics. Researchers said this week they have developed a model using artificial intelligence that, if further verified on a larger scale, could be implemented in these clinical settings.

Findings from the proof-of-principle study were published in Clinical Cancer Research. The computational method combines clinicodemographic variables with deep learning of pretreatment histology images.

The researchers used data from a training cohort of 121 patients with metastatic melanoma who received ICI treatment between 2004 and 2018 at the New York University (NYU) Perlmutter Comprehensive Cancer Center. ICIs target cytotoxic T lymphocyte antigen-4 (CTLA-4) and programmed cell death 1/ligand-1 (PD-1/PD-L1). The patients were treated with first-line anti-CTLA-4 therapy, anti-PD-1 therapy, or a combination of both; the majority (about 64%) received anti-CTLA-4 monotherapy.

Clinical outcomes were recorded as progression of disease or response, which included complete or partial responses.

Computer algorithms called deep convolutional neural networks (DCCN) analyzed digital images of hematoxylin and eosin–stained slides of metastatic lymph node and subcutaneous tissue and identified patterns associated with treatment response. A response classifier predicted if a patient's untreated tumor would respond to ICI therapy or progress.

The DCCN response classifier was validated in an independent group of 30 patients with metastatic melanoma treated at Vanderbilt-Ingram Cancer Center between 2010 and 2017. About 54% of these patients received anti-PD-1 agents.

The researchers assessed the performance of the DCCN response classifier by calculating the area under the curve (AUC), a measure of the model’s accuracy, in both the training group at NYU and the validation group from Vanderbilt.

An AUC of 1 corresponds to perfect prediction, and the DCCN prediction model achieved about 0.7 in both groups.

To augment the prediction accuracy of the model, the researchers performed multivariable logistic regressions that combined the DCCN prediction with conventional clinical characteristics. The final model incorporated the DCCN prediction, Eastern Cooperative Oncology Group performance status, and treatment regimen.

In both the training and validation cohorts, the multivariable classifier achieved an AUC around 0.8.

In the validation cohort, the classifier could stratify patients into high vs low risk for disease progression, with significantly different progression-free survival outcomes between the 2 groups.

The results of the model, which were consistent no matter which therapy the patients received, suggest that some biomarkers are not necessarily specific to the checkpoint target, the authors wrote.

Class activation mapping, which can identify regions within the digital slides used by the DCCN, indicated that cell nuclei were a key predictor, as larger and more numerous nuclei correlated with progression of disease.

"These results suggest that ploidy might be one of the biologic determinants detected by DCCN," said study author Iman Osman, MD, a medical oncologist at NYU’s Grossman School of Medicine and director of the Interdisciplinary Melanoma Program at NYU Langone’s Perlmutter Cancer Center, in a statement.

More work is needed, included larger training and testing data sets and additional validation parameters, before the method can reach clinical settings.

But the authors said the approach has potential “to oncologists identify patients who are at high versus low risk for progression through immunotherapy."

Reference

Johannet P, Coudray N, Donnelly DM, et al. Using machine learning algorithms to predict immunotherapy response in patients with advanced melanoma. Clin Cancer Res. Published online November 18, 2020. doi:10.1158/1078-0432.CCR-20-2415