1677P Risk factors predicting immune checkpoint inhibitors (ICIs) toxicity using machine learning computer algorithm

Although ICIs revolutionized modern oncology, a significant challenge remains to minimize immune related adverse events (irAEs), and subsequent financial burden due treatment delay hospitalization. There is paucity of data about predicting biomarkers for ICI-related toxicity. In this study we aimed use machine learning (ML) algorithms the profile ICI A computer algorithm was developed using SAP Business Objects BI Platform 4.2 extract electronic medical records (EMR) patients receiving identify irAEs. To avoid potential confounding effects, included only patient without concurrent chemotherapy, biological or hormonal therapy radiotherapy during period. The used all available on each patient; Namely, structured (e.g., numeric lab results) unstructured notes written by treating physician). code mapped irAEs drawn from database, as well ML models in Natural Language Processing (NLP) with MDCLONE Studio V1.1.2 tool. Association between variables were assessed chi-squared t tests. cohort 1617 solid tumors who received years 2010-2021 Division Oncology Rambam Health Care Campus. 1104 (68%) men, mean age 69 years. 910 (56%) grade 3&4 after treatment. Gender, BMI pretreatment derived neutrophil-to-lymphocytes ratio (dNLR) not associated higher general cohort. Younger PDL-1/CTLA-4 combination found be irARs rates (P=0.001 P<.001, respectively). subgroup analysis, young hepatotoxicity hematologic (P<.001 P=0.01, respectively), female gender endocrine toxicity (P=0.024) high low dNLR renal P=0.029, Using tools real-world setting, several key characteristics identified correlated tendency