P07.09.A DEVELOPMENT AND VALIDATION OF A LOGISTIC REGRESSION-BASED NOMOGRAM FOR PREDICTING POSTOPERATIVE HEMORRHAGE FOLLOWING INTRACRANIAL TUMOR BIOPSY: A SINGLE-CENTER PILOT STUDY

Abstract BACKGROUND Stereotactic biopsy, while essential for intracranial tumor diagnosis, carries 5%-20% hemorrhage risk, with 10%-30% requiring surgical evacuation. Existing models using coagulation or tumor size lack clinical utility due to complexity (≥12 variables) or poor discrimination (AUC<0.75). We developed a simplified predictive tool using routine preoperative data. MATERIAL AND METHODS This retrospective study analyzed 69 biopsy patients (2023-2024) from a tertiary center, divided into derivation (n=48) and validation (n=21) cohorts. The primary outcome was radiologically confirmed hemorrhage (≥1 cm³) within 24h; secondary outcomes included surgical evacuation for hematomas ≥30 mL or neurological decline. We assessed 37 variables, including demographics, tumor characteristics (depth, diameter), and hematologic profiles. Analysis involved random forest preselection (top 20% variables), multivariable logistic regression (backward elimination, p<0.10), and nomogram development with 1,000 bootstrap validations. Performance was evaluated using ROC-AUC, Brier score, and decision curve analysis (5-40% thresholds). RESULTS The hemorrhage incidence was 35.4% (17/48) in the derivation cohort versus 47.6% (10/21) in the validation cohort, respectively. The final predictive model included six significant variables: serum chloride (OR=0.82), urea (OR=1.14, eosinophils (OR=1.28), albumin (OR=0.96), platelet width (OR=3.64), and deep-seated tumor location (OR=1.27). The model demonstrated strong discrimination in validation (AUC=0.83), outperforming conventional models by ΔAUC +0.18 (P<0.001). It showed good calibration (Brier score=0.12) and clinical utility at risk thresholds above 41%. CONCLUSION This validated nomogram enables hemorrhage risk stratification using six preoperative parameters. Correcting modifiable factors (e.g., electrolytes) and vigilance in deep-seated tumors may reduce reoperations. Multicenter validation is warranted.