24P Using real-world evidence data and machine learning to identify molecular biomarkers for patient response to immune checkpoint inhibitors in metastatic melanoma

BackgroundImmune checkpoint inhibitors (ICIs) are one of the most advanced cancer treatments with good responses in metastatic melanoma (MM). However, the response does not exceed 40%, while 7% to 15% of patients may develop hyperprogressive disease (HPD or early death) upon ICIs administration. Currently, oncologists lack effective and reliable markers to predict patient response or identify patients that will eventually develop HPD (early death).MethodsTo develop and validate biomarker panels for HPD (early death) and response/non-response (measured by RECIST 1.1) to immunotherapy in melanoma patients, a total of 273 metastatic melanoma patients treated with PD1/PD-L1 inhibitor or its combinations with CTLA4 inhibitor were selected for analysis. These included 210 patients from 4 past interventional clinical trials and 63 real-world evidence data from an ongoing observational retrospective study in 3 hospitals in Belgium and Austria (NCT04860076). The patients enrolled in the study were treated with Nivolumab or Pembrolizumab, or their combinations with Ipilimumab. We obtained molecular, immunohistochemistry, and clinical data and applied advanced machine learning methods to investigate predictive factors that are indicative of therapy response or/and development of HPD. We use ensemble machine learning to discriminate clinical and molecular characteristics and build a predictive model. The model was developed on an initial discovery cohort and then confirmed with high accuracy on a validation cohort.ResultsWe identified the signature of 10 genes which was able to predict clinical outcomes (RECIST 1.1 status and/or early death) for metastatic melanoma patients treated by ICIs with high accuracy. Depending on the classification, the model accuracy reached 73% to 78% measured as ROC AUC on an independent validation cohort.ConclusionsIdentified biomarkers can have the potential to aid patient stratification for immune therapy and personalized treatment options for metastatic melanoma patients for whom there is currently no complementary diagnostics available on the market.Clinical trial identificationNCT04860076.Legal entity responsible for the studyAsylia Diagnostics BV.FundingAsylia Diagnostics BV.DisclosureV. Siozopoulou: Financial Interests, Institutional, Advisory Role: MSD. A. Khmelevskiy: Financial Interests, Personal and Institutional, Full or part-time Employment: Asylia Diagnostics. E. Richtig: Financial Interests, Personal and Institutional, Advisory Role: MSD; Financial Interests, Personal and Institutional, Advisory Role: Amgen; Financial Interests, Personal and Institutional, Advisory Role: Bristol Myers Squibb; Financial Interests, Personal and Institutional, Advisory Role: Merck; Financial Interests, Personal and Institutional, Advisory Role: Novartis; Financial Interests, Personal and Institutional, Advisory Role: Sanofi; Financial Interests, Personal and Institutional, Advisory Role: Pierre Fabre. All other authors have declared no conflicts of interest. BackgroundImmune checkpoint inhibitors (ICIs) are one of the most advanced cancer treatments with good responses in metastatic melanoma (MM). However, the response does not exceed 40%, while 7% to 15% of patients may develop hyperprogressive disease (HPD or early death) upon ICIs administration. Currently, oncologists lack effective and reliable markers to predict patient response or identify patients that will eventually develop HPD (early death). Immune checkpoint inhibitors (ICIs) are one of the most advanced cancer treatments with good responses in metastatic melanoma (MM). However, the response does not exceed 40%, while 7% to 15% of patients may develop hyperprogressive disease (HPD or early death) upon ICIs administration. Currently, oncologists lack effective and reliable markers to predict patient response or identify patients that will eventually develop HPD (early death). MethodsTo develop and validate biomarker panels for HPD (early death) and response/non-response (measured by RECIST 1.1) to immunotherapy in melanoma patients, a total of 273 metastatic melanoma patients treated with PD1/PD-L1 inhibitor or its combinations with CTLA4 inhibitor were selected for analysis. These included 210 patients from 4 past interventional clinical trials and 63 real-world evidence data from an ongoing observational retrospective study in 3 hospitals in Belgium and Austria (NCT04860076). The patients enrolled in the study were treated with Nivolumab or Pembrolizumab, or their combinations with Ipilimumab. We obtained molecular, immunohistochemistry, and clinical data and applied advanced machine learning methods to investigate predictive factors that are indicative of therapy response or/and development of HPD. We use ensemble machine learning to discriminate clinical and molecular characteristics and build a predictive model. The model was developed on an initial discovery cohort and then confirmed with high accuracy on a validation cohort. To develop and validate biomarker panels for HPD (early death) and response/non-response (measured by RECIST 1.1) to immunotherapy in melanoma patients, a total of 273 metastatic melanoma patients treated with PD1/PD-L1 inhibitor or its combinations with CTLA4 inhibitor were selected for analysis. These included 210 patients from 4 past interventional clinical trials and 63 real-world evidence data from an ongoing observational retrospective study in 3 hospitals in Belgium and Austria (NCT04860076). The patients enrolled in the study were treated with Nivolumab or Pembrolizumab, or their combinations with Ipilimumab. We obtained molecular, immunohistochemistry, and clinical data and applied advanced machine learning methods to investigate predictive factors that are indicative of therapy response or/and development of HPD. We use ensemble machine learning to discriminate clinical and molecular characteristics and build a predictive model. The model was developed on an initial discovery cohort and then confirmed with high accuracy on a validation cohort. ResultsWe identified the signature of 10 genes which was able to predict clinical outcomes (RECIST 1.1 status and/or early death) for metastatic melanoma patients treated by ICIs with high accuracy. Depending on the classification, the model accuracy reached 73% to 78% measured as ROC AUC on an independent validation cohort. We identified the signature of 10 genes which was able to predict clinical outcomes (RECIST 1.1 status and/or early death) for metastatic melanoma patients treated by ICIs with high accuracy. Depending on the classification, the model accuracy reached 73% to 78% measured as ROC AUC on an independent validation cohort. ConclusionsIdentified biomarkers can have the potential to aid patient stratification for immune therapy and personalized treatment options for metastatic melanoma patients for whom there is currently no complementary diagnostics available on the market. Identified biomarkers can have the potential to aid patient stratification for immune therapy and personalized treatment options for metastatic melanoma patients for whom there is currently no complementary diagnostics available on the market.