Guideline-based strategies to identify severe cytokine release syndrome in COVID-19 and cancer immunotherapy using large-scale electronic health records.

Introduction: Cytokine Release Syndrome (CRS) is a life-threatening adverse event of cancer immunotherapies and a complication of infections. Predicting which patients are at risk for severe CRS would inform mitigation decisions and drug development, but requires large, reliably labeled datasets. Methods: This study evaluates the feasibility of disease-agnostic case identification of CRS patterns in large-scale Electronic Health Records (EHR) to generate high-quality cohorts of CRS-positive and CRS-negative patients. Results: Using the Optum® de-identified COVID-19 EHR dataset, we isolated 2.5 million patients with active COVID-19 and 171 individuals treated with the T-cell Engager (TCE) blinatumomab. Diagnosis codes for CRS were underutilized and provided limited information on severity. Instead, we implemented the consensus CRS grading guidelines, which identified 92,541 COVID-19 patients (3.7%) and 54 blinatumomab patients (31.5%) with grade 2 or higher CRS, respectively. Severe CRS COVID-19 patients showed heterogeneous inflammatory levels. Discussion: Our EHR-based CRS case identification strategy is suitable for risk factor analysis and developing CRS risk prediction models.