Development of a laboratory-based nomogram for predicting clinical outcomes in patients with severe COVID-19 undergoing glucocorticoid therapy

Background While glucocorticoids remain cornerstone therapy for severe COVID-19, substantial heterogeneity persists in clinical outcomes. This single-center retrospective study sought to establish a predictive model integrating inflammatory biomarkers to guide risk stratification and personalized management. Methods We analysed 151 adults with WHO-defined severe COVID-19 receiving glucocorticoid therapy (December 2022–August 2023). Treatment non-response was defined as mortality during hospitalization, mechanical ventilation escalation, or persistent organ dysfunction. LASSO and logistic regression analyses identified predictors, with optimal biomarker thresholds determined using ROC curves. A nomogram was constructed and validated via split-sample testing (7:3 ratio) and 10-fold cross-validation. Results Ferritin >970.7 ng/mL and IL-10 > 4.79 pg./mL predicted glucocorticoid resistance (AUC: training set 0.779, test set 0.780). The nomogram incorporated diabetes, ferritin, and IL-10, demonstrating robust calibration (Hosmer-Lemeshow p = 0.84; Brier score = 0.182) and discrimination (sensitivity = 71.4%, specificity = 70.0%). Diabetic patients exhibited heightened inflammatory responses and poorer outcomes, exacerbated by glucocorticoid-induced hyperglycaemia. Conclusion This nomogram shows promising predictive performance and provides a potentially implementable framework for risk stratification and personalized management, which warrants prospective validation in larger, multi-center cohorts.