Objective: To develop and validate a dynamic nomogram for the early prediction of 90-day in-hospital mortality risk in critically ill ICU patients with diabetic kidney disease (DKD), thereby filling the gap in predictive models for this population. Methods: Data were extracted from the MIMIC-IV database and divided into training and validation sets in a 7∶3 ratio. Feature variables were selected using the Boruta algorithm combined with Least Absolute Shrinkage and Selection Operator (LASSO) regression. The nomogram for 90-day in-hospital mortality risk was visualized using R software, and a dynamic nomogram was generated using the DynNom package. Model performance was evaluated by the area under the receiver operating characteristic curve (AUC) and calibration curves. The clinical utility was validated through decision curve analysis (DCA). Results: A total of 405 critically ill ICU patients with DKD were included, of whom 88 experienced 90-day in-hospital mortality. The results showed that admission age, red cell distribution width (RDW), and delirium were independent predictors of 90-day in-hospital mortality. Multivariable Logistic regression indicated that all three variables were statistically significant, and a forest plot visualized the effect estimates with 95% confidence intervals. The AUC values were 0.825 (95% CI: 0.793~0.857) for the training set and 0.717 (95% CI: 0.592~0.841) for the validation set. The calibration curves demonstrated good prediction concordance, and DCA confirmed the model's significant clinical value. The dynamic nomogram was available online at: https://codepen.io/Thi-Giang-Ho/pen/vEKOQar. Conclusion: The dynamic nomogram constructed based on the MIMIC-IV database integrates three readily available clinical features to predict the probability of 90-day in-hospital mortality in critically ill ICU patients with DKD. Internal validation demonstrated robust predictive performance and good clinical utility of the model.