PROJECT SUMMARY/ABSTRACT Immune checkpoint inhibitors (ICIs) have drastically improved cancer survival over the past decade, but this survival comes at the cost of a new class of immune-related adverse events (irAEs) characterized by inflammatory and auto-immune pathologies that occur and persist long after ICI discontinuation. These irAEs can have major impacts on long-term quality-of-life, but our ability to appropriately address them is limited by an insufficient understanding of irAE rates and severity profiles. Automated methods to identify and monitor irAEs could improve clinical care, biomedical research, and pharmacovigilance, however, irAEs are often only documented in clinical text and cannot currently be automatically extracted from the EHR at scale. The overarching objective of this proposal is to create applied informatics technologies for cancer surveillance research and survivorship care in patients treated with ICIs. Our central innovation is the development and clinical validation of natural language processing methods, particularly neural language models, that can handle the complexities of the EHR for irAE extraction using unstructured and structured data streams. In Specific Aim 1, we conduct a clinical trial of informatics-assisted irAE detection from the EHR, measuring feasibility and effectiveness in improving registration onto Alliance A151804, an NCI cooperative group irAE biorepository. This will be the first trial of informatics-based adverse event detection for cancer care and a major step toward clinical translation. In Specific Aim 2, we develop new methods to extract irAEs according to their severity grade for detailed and standardized computational phenotyping, and perform external validations. In Specific Aim 3, we optimize generalist large language models for irAE information extraction without task- specific fine-tuning, including innovative methods to tailor models’ diagnostic reasoning to each patient. This work is highly significant for developing, applying, and validating informatics methods that take full advantage of the EHR to support the long-term goal of improving quality-of-life and survival in patients treated with ICIs. This clinical translational work will be carried out by an expert team of cancer clinicians, clinical trialists, informaticians, and computer scientists.

PROJECT NARRATIVE Immune checkpoint inhibitors dramatically improve prognosis for many cancer patients but come at the cost of a new class of immune-related adverse events that reduce overall quality-of-life and the net benefit of treatment. This proposal seeks to automatically detect immune-related adverse event phenotypes from the electronic health record to support timely, data-driven cancer care that enhances survivorship. To achieve this goal, we will develop and clinically validate advanced informatics and artificial intelligence technologies, such as deep learning language models, to automate immune-related adverse event monitoring and data collection.