Background: Patients who receive high dose corticosteroids and thereafter become hyperglycemic are at increased risk of inpatient mortality and developing diabetes. However, it is difficult to predict which patients will become hyperglycemic.

Methods: A cohort of 19882 patients with 28845 inpatient encounters between Oct 2014 and Aug 2018 was selected to create a model to predict incident hyperglycemia. Patients in this cohort were 18 years old and had received high dose corticosteroids (20mg/day of prednisone equivalents). Patients who died during admission or were discharged to hospice were excluded. Model features included creatinine, bicarbonate, AST, ALT, and albumin results and total steroid administration dose in the prior 24 hours as continuous variables and the presence of a lactate result in this period as a binary indicator variable. The outcome of hyperglycemic events was defined as patients having a blood glucose value above 180 within 12 hours of steroid administration. A logistic regression analysis was used as the predictive method.

Results: A subset of 71837 steroid administrations were used to create the predictive model and 8278 were used to test the model’s performance (AUC of 0.737).

Conclusion: Several labs in the 24 hours prior to receipt of high dose steroids are associated with hyperglycemia. The presence of a lactate result was the most significant predictor of future hyperglycemia.


M.G. Simons: None. J.D. Futoma: None. M. Gao: None. K. Corey: None. M. Sendak: None. K.B. Whalen: None. F. Doshi-velez: None. A.S. McGee: None. T. Setji: None.

Readers may use this article as long as the work is properly cited, the use is educational and not for profit, and the work is not altered. More information is available at