Article Text

Download PDFPDF

1317 Enhancing patient safety in melanoma treatment: harnessing machine learning for predicting immune-related adverse events
  1. Yili Zhang1,
  2. Shaked Lev Ari1,
  3. Samir Gupta1,
  4. Jacob Zaemes1,
  5. Neil J Shah2,
  6. Adil Alaoui1,
  7. Subha Madhavan1,
  8. Peter McGarvey1 and
  9. Michael B Atkins1
  1. 1Georgetown University, Washington DC, USA
  2. 2Memorial Sloan Kettering Cancer Center, Washington, DC, USA
  • Journal for ImmunoTherapy of Cancer (JITC) preprint. The copyright holder for this preprint are the authors/funders, who have granted JITC permission to display the preprint. All rights reserved. No reuse allowed without permission.


Background Immune checkpoint inhibitors (ICIs), particularly the combination of anti-CTLA4 and anti-PD1, have demonstrated efficacy in treating patients with advanced melanoma. Although ICIs, whether administered individually or in combination, can lead to immune-related adverse events (irAEs), the clinical factors that predict the risk of irAEs are still uncertain. The objective of this study is to build machine learning (ML) models to predict whether patients will develop irAE(s) for melanoma patients receiving Immune-Oncology (IO) therapy and find the corresponding important factors.

Methods In our single-center study utilizing electronic medical records, we identified patients with advanced melanoma who received anti-PD-1 and anti-CTLA4 between January 2011 and April 2018. Baseline demographics, laboratory parameters1 and derived ratios, treatment history, cancer-related mutations, and irAEs were collected. The evaluation of irAE types and grades was based on CTCAE V4.03. A total of 224 patients were included in the study cohort, comprising 138 patients who developed one or more irAEs and 86 patients who remained irAEs-free. Eight ML models were trained with 80% of the data with five-fold cross-validation and tested with the remaining 20% of data. The area under the receiver-operating curve (AUROC) was used to assess ML models. Important features affecting the model were analyzed based on the model with the best performance by SHAP value.2

Results We employed eight machine learning models, including logistic regression, support vector machine, bagging k-nearest neighborhood (BKNN), random forest, Bernoulli Naive Bayes (BNB), etc. The BKNN and BNB models exhibited the highest AUROC score, achieving scores of 80.65% and 79.37%, respectively. According to the AUROC curve shown in (figure 1), BNB was chosen as the superior model for accurately predicting irAE development. The most important three features selected from the BNB model are the utilization of nivolumab and ipilimumab therapy in combination, pretreatment ECOG score of 0, and blood eosinophil count, as shown in (figure 2).

Conclusions Our study leverages comprehensive variables to predict irAE development for patients with melanoma treated with ICI. The findings demonstrate the predictive capability of ML models, with the BNB model exhibiting the highest performance. These outcomes underscore the promising prospects of ML techniques not only in melanoma but also in other cancer diseases treated with ICI. To address the limited sample size, we will involve more institutions in the data registry. Our future direction involves incorporating additional variables, like pre-treatment auto-antibodies, to predict irAE grade and type, enhancing irAE prediction.


  1. Hopkins AM, Rowland A, Kichenadasse G, Wiese MD, Gurney H, McKinnon RA, Karapetis CS, Sorich MJ. Predicting response and toxicity to immune checkpoint inhibitors using routinely available blood and clinical markers. Br J Cancer. 2017 Sep 26;117(7):913–920. doi: 10.1038/bjc.2017.274. Epub 2017 Aug 24. PMID: 28950287; PMCID: PMC5625676.

  2. Singh R, Lanchantin J, Sekhon A, Qi Y. Attend and Predict: Understanding Gene Regulation by Selective Attention on Chromatin. Adv Neural Inf Process Syst. 2017 Dec;30:6785–6795. PMID: 30147283; PMCID: PMC6105294.

This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.