Development of a prediction model using preoperative immune-inflammatory cell features for persistent central diabetes insipidus 2 weeks after surgery in patients with craniopharyngioma.

Background: Central diabetes insipidus (CDI) is a frequent complication after craniopharyngioma (CP) surgery. Although many patients recover within 2 weeks, a substantial proportion develop persistent CDI that requires long-term management. Systemic inflammatory activation may aggravate hypothalamic-neurohypophyseal injury and delay functional recovery. However, reliable preoperative predictors of early recovery remain limited. This study aimed to develop a multivariable prediction model for persistent CDI at 2 weeks after CP surgery. Methods: This retrospective study included patients with pathologically confirmed CP who underwent tumor resection between 2016 and 2025. Patients with preoperative polyuria, severe infection, immune disorders, or incomplete hematological data were excluded. Persistent CDI was defined as polyuria (>3,000 mL/day, urine specific gravity <1.005) with hypernatremia persisting beyond 2 weeks postoperatively. Multivariate logistic regression was used to identify predictors of CDI recovery at 2 weeks, and model performance was assessed using receiver operating characteristic (ROC) curves. To reduce overfitting and assess model stability, internal validation was performed using five-fold cross-validation. SHapley Additive exPlanations (SHAP) were applied to determine variable importance. Two-sample Mendelian randomization (MR) analysis was conducted to explore causal relationships between immune cell traits and CDI recovery. Results: This study finally included 108 individuals; 49 (45.4%) developed persistent CDI. Baseline demographic and tumor characteristics did not differ significantly between groups. Patients with persistent CDI exhibited significantly higher counts of preoperative white blood cell (WBC), platelet (PLT), neutrophils, and monocytes (all P<0.05). These markers were independently associated with persistent CDI (P<0.05). The four-variable model demonstrated strong predictive performance [area under the curve (AUC) =0.944] and remained stable across age and sex subgroups (AUC range, 0.820-0.882). SHAP analysis identified WBC and PLT as the most influential predictors. MR analysis further supported a potential causal association between elevated immune cell levels and increased CDI risk. Conclusions: This study developed a multivariable prediction model based on preoperative immune-inflammatory cell features for identifying patients at risk of persistent CDI 2 weeks after CP surgery. This provides a novel strategy for individualized perioperative management and early identification of patients at risk for persistent CDI.