Background: Immune checkpoint inhibitors (ICIs) have revolutionized the treatment of cancer with significantly improved outcomes, but these agents have a unique spectrum of toxicities known as immune-related adverse events (irAEs). The recommended treatment for non-endocrine toxicities is steroid based. However, a subset of patients (pts) is steroid-refractory and requires second-line immunosuppression. There is very little evidence regarding this population. In this retrospective study we report the 1) incidence 2) type of treatment used 3) natural history and 4) potential predictors of steroid-refractory irAE at a major academic medical center. Methods: The Research Patient Database Registry at Mass General Brigham was used to identify pts treated with an ICI from 1/5/2017 to 6/1/2019. Pharmaceutical records identified a subset of the cohort received a second-line immunosuppressive agent within a 15-month period after ICI. For pts with steroid-refractory irAE additional information was collected: demographics, ICI regimen, type/#/and severity of irAE, clinical characteristics, # of admissions, length of stay (LOS), amount and duration of steroid therapy, second line immunosuppression type, treatment discontinuation rates, response, and outcome of re-challenge. Multivariate logistic regressions were used to predict risk of refractory toxicity and study the association of different variables (age, sex, race, marital status, cancer and ICI types) with refractory toxicities. Results: We identified 61 pts (1.4%) with steroid-refractory irAEs (48 colitis, 4 myocarditis, 6 pneumonitis, 3 neurologic) out of the total ICI cohort (N=4,325). 60.7% received ICI monotherapy. 24.6% received ICI in the adjuvant setting. Median length of steroid duration was 68 days with max of 1135 days. Despite use of second line immunosuppression, 25.8% of pts were never able to discontinue steroids. Majority of pts (72.1%) had at least one hospitalization with median LOS of 7.5 days. 93.4% of pts permanently discontinued the ICI responsible for the irAE. Thirteen pts (21.3%) were later re-challenged with ICI and 7 (53.8%) of these developed a subsequent irAE. Anti-CTLA-4 therapy was associated with a 10-fold risk of refractory toxicity compared to PD-1 (p<.05). Best tumor response was complete response in 21.3% and partial response in 26.2%. Among different cancer types, melanoma was most strongly associated with refractory events (OR 2.97 in comparison to thoracic malignancy). Conclusions: Refractory toxicity is uncommon but leads to high rates of ICI discontinuation, frequent hospitalizations, and a long duration of illness with exposure to prolonged and high-doses of steroids. There is an urgent need for further investigation into predictive factors for steroid-refractory toxicity given that ICI is being used more frequently and in earlier lines of treatment.