Journal Article

1094 Pre-treatment predictive modeling of immune-related adverse event risk in immune checkpoint blockade therapy: a multi-modal machine learning approach from a real-world setting (RADIOHEAD Cohort Study)

Markova, Kseniia L, Alesse, Julia E, Syune, Ambaryan, Zaytcev Alexander, Wang, Iris, Sokolov, Arseniy A, Yang EnJun, Liang, Samantha I, Vardazaryan Anna, Alekseeva Evgeniia, Shulga Polina, Fastovetz Dmitry, Brunovlenskaia-Bogoiavlenskaia Aleksandra, Terenteva Anastasiya, Sheehan, Elizabeth, Mulyukina Alina, Tarasova, Anna, Bezlepkina Svetlana, Conroy Tori, Aloyan Ani, Nazar, Arutyunyan, Tumasyan Diana, Golubeva, Olga, Shilov Evgeniy, Bolshakov, Evgenii S, Goldberg, Michael F, English, Noel, Lawless Aleigha, Frank, Alena, Kust, Sofya A, Spirin Danil, Krauz Ilya, Barykin Evgeny, Boland, Genevieve M, Connolly, John E, Vasileva Tatiana, Savchenko, Maria, Bolshakova Anastasiia, Ushakova Ekaterina

2025

Overview

BackgroundPredicting the risk of immune-related adverse events (IrAEs) in patients receiving immune checkpoint blockade (ICB) therapy is crucial for optimizing safety and treatment outcomes. While single biomarkers have been implicated in risk stratification, multi-modal data integration can greatly enhance prediction accuracy.1–3 Here, we developed a blood-based, multi-modal predictor for estimating severe irAE risk prior to ICB initiation.MethodsClinical annotations, HLA genotyping, and bulk RNA-sequencing data from peripheral blood were analyzed across three cohorts: cancer patients prior to ICB therapy (Cohorts 1 and 2) and individuals with inflammatory bowel disease (IBD) or healthy controls (Cohort 3) (table 1). Events with grades ≥3 were defined as severe IrAEs. RNA-seq data were preprocessed to evaluate features such as immune signatures,4 cell composition,5 immunotypes,6 and TCR diversity. For overall IrAE risk estimation, the CatBoost classifier was used to train predictors on clinical data,7 while a logistic regression model was trained on integrated clinical and transcriptomic data. A genetic-based model based on HLA alleles was also developed to predict specific IrAE types.ResultsA CatBoost classifier trained to assess IrAE risk based on clinical data from Train Set 1 predicted severe irAEs in Test Sets 1 (AUC=0.72) and 2 (AUC=0.59). Age, therapy type, and diagnosis were the major risk determinants. We improved the model’s predictive capacity by combining its output with transcriptomic data (12 preprocessed features) and training a logistic regression classifier on Train Set 1, subsequently achieving AUC=0.79 in Test Set 1 and AUC=0.63 in Test Set 2. Features related to immune regulation (such as CD4+ T-cell activation, myeloid-cell-mediated suppression, and Treg activity) were strongly associated with severe IrAEs. Finally, using train and test sets from Cohort 3, we developed a CatBoost classifier based on 46 HLA alleles associated with risk of and protection against autoimmune colitis. This genetic-based model predicted the likelihood of colitis with AUC=0.68 in patients with severe IrAEs from Cohort 1, revealing a threefold HLA-driven increase in colitis risk (OR=2.8).ConclusionsOur multi-modal, machine-learning approach provides a robust framework for assessing severe IrAE risk before ICB initiation. Early identification and monitoring of high-risk patients will enable physicians to mitigate IrAEs in a timely manner, thus improving patient outcomes.ReferencesAli O, Berner F, Bomze D, Fässler M, Diem S, Cozzio A, Jörger M, Früh M, Driessen C, Lenz TL, Flatz L. Human leukocyte antigen variation is associated with adverse events of checkpoint inhibitors. Eur J Cancer. 2019;107:8–14.Ye W, Olsson-Brown A, Watson R, Cheung V, Morgan R, Nassiri I, Cooper R, Taylor C, Akbani U, Brain O, Matin R, Coupe N, Middleton M, Coles M, Sacco J, Payne M, Fairfax B. Checkpoint-blocker-induced autoimmunity is associated with favourable outcome in metastatic melanoma and distinct T-cell expression profiles. Br J Cancer. 2021;124(11):1661–1669.Kim KH, Hur JY, Cho J, Ku BM, Koh J, Koh JY, Sun JM, Lee SH, Ahn JS, Park K, Ahn MJ, Shin EC. Immune-related adverse events are clustered into distinct subtypes by T-cell profiling before and early after anti-PD-1 treatment. Oncoimmunology. 2020;9(1):1722023.1–12.Bolshakov E, Vasileva T, Kust S, Frank A, Savchenko M, Wang I, Conroy T, Merriam NR, Markova K, Brunovlenskaia-Bogoiavlenskaia A, Ushakova E, Ambarian S, Mulyukina A, Shilov E, Arutyunyan N, Shchetsova A, Shulga P, Spirin D, Terenteva A, Goldberg MF, Lawless A, Boland GM, Sullivan RJ, Zaytcev A. Identifying a composite signature for predicting immune-related adverse events in advanced melanoma patients treated with immune checkpoint blockade. J Immunother Cancer. 2024;12:e144.Zaitsev A, Chelushkin M, Dyikanov D, Cheremushkin I, Shpak B, Nomie K, Zyrin V, Nuzhdina E, Lozinsky Y, Zotova A, Degryse S, Kotlov N, Baisangurov A, Shatsky V, Afenteva D, Kuznetsov A, Paul SR, Davies DL, Reeves PM, Lanuti M, Goldberg MF, Tazearslan C, Chasse M, Wang I, Abdou M, Aslanian SM, Andrewes S, Hsieh JJ, Ramachandran A, Lyu Y, Galkin I, Svekolkin V, Cerchietti L, Poznansky MC, Ataullakhanov RI, Fowler N, Bagaev A. Precise reconstruction of the TME using bulk RNA-seq and a machine learning algorithm trained on artificial transcriptomes. Cancer Cell. 2022;40(8):879–894.e16.Dyikanov D, Zaitsev A, Vasileva T, Wang I, Sokolov A, Bolshakov E, Frank A, Turova P, Golubeva O, Gantseva A, Kamysheva A, Shpudeiko P, Krauz I, Abdou M, Chasse M, Conroy T, Merriam NR, Alesse JE, English N, Shpak B, Shchetsova A, Tikhonov E, Filatov I, Radko A, Bolshakova A, Kachalova A, Lugovykh N, Bulahov A, Kilina A, Asanbekov S, Zheleznyak I, Skoptsov P, Alekseeva E, Johnson JM, Curry JM, Linnenbach AJ, South AP, Yang EJ, Morozov K, Terenteva A, Nigmatullina L, Fastovetz D, Bobe A, Balabanian L, Nomie K, Yong ST, Davitt CJH, Ryabykh A, Kudryashova O, Tazearslan C, Bagaev A, Fowler N, Luginbuhl AJ, Ataullakhanov R, Goldberg MF. Comprehensive peripheral blood immunoprofiling reveals five immunotypes with immunotherapy response characteristics in patients with cancer. Cancer Cell. 2024;42(5):759–779.e12.Quandt Z, Lucas A, Liang SI, Yang E, Stone S, Fadlullah MZH, Bayless NL, Marr SS, Thompson MA, Padron LJ, Bucktrout S, Butterfield LH, Tan AC, Herold KC, Bluestone JA, Anderson MS, Spencer CN, Young A, Connolly JE. Associations between immune checkpoint inhibitor response, immune-related adverse events, and steroid use in RADIOHEAD: a prospective pan-tumor cohort study. J Immunother Cancer. 2025 May 12;13(5):e011545.Ethics ApprovalThis study involves human participants and was approved by WCG IRB Protocol #20182579. Participants gave informed consent to participate in the study before taking part.Abstract 1094 Table 1Cohort description

• Publisher: BMJ Publishing Group LTD
• Subject: Cancer / Cohort analysis / Inflammatory bowel disease / Machine learning / Melanoma / Regression analysis