Development of a Risk Prediction Model for Early Postoperative Seizures After Glioma Resection and Optimization of Antiseizure medications Strategies Based on Decision Curve Analysis

Technical Summary: Risk Prediction and Medication Optimization for Early Postoperative Seizures in Glioma Resection

Problem Statement
Gliomas are the most common primary malignant central nervous system tumors, with postoperative seizures occurring in 60% to 90% of patients. These events prolong hospitalization, increase costs, and risk secondary brain damage. Current clinical practice relies on empirical, often universal, prophylactic antiseizure medication (ASM) regimens due to a lack of individualized guidance. While previous studies have attempted to create risk prediction models, they are typically limited by small, single-center samples and fail to translate risk stratification into specific clinical medication decisions. Furthermore, existing validation efforts often suffer from "treatment confounding," where the widespread use of prophylactic medication in validation cohorts leads to an underestimation of the disease's natural risk, complicating the evaluation of intervention strategies.

Methodology
The study employed a multi-stage methodological framework:

1. Systematic Review and Meta-Analysis: Researchers searched Chinese and English databases (up to May 2025) for cohort or case-control studies analyzing risk factors for early postoperative seizures (within 7 days) in glioma patients. Fifteen high-quality studies involving 3,409 patients were included. Pooled odds ratios (ORs) were calculated using fixed or random-effects models based on heterogeneity ($I^2$).
2. Risk Scoring Model Construction: Based on the meta-analysis results, a scoring model was developed using the Sullivan et al. method. Eight independent risk factors were identified and assigned integer or half-integer scores to create a total possible score of 26. Factors such as "no prophylactic medication" were excluded to prevent circular logic in model construction.
3. External Validation: The model was validated against an independent retrospective cohort of 333 glioma patients from a single center (2019–2024). Discriminatory power was assessed via Receiver Operating Characteristic (ROC) curves and the Area Under the Curve (AUC). Patients were stratified into low, moderate, and high-risk groups based on the optimal cutoff determined by the Youden index.
4. Counterfactual Probability Calibration: To address the high rate of prophylactic medication use (96.7%) in the validation cohort, which masked the natural risk of seizures, the study applied a counterfactual probability calibration method. This estimated the natural risk of epilepsy in the absence of medication using the pooled OR for non-medication derived from the meta-analysis.
5. Decision Curve Analysis (DCA): DCA was utilized to quantify the clinical net benefit (NB) of three strategies—"universal medication," "no medication," and "risk-stratified medication"—across different risk thresholds (2%–20%). Sensitivity analyses were conducted using the 95% confidence interval limits of the OR to test robustness.

Key Contributions

• Evidence-Based Scoring Model: The study presents a simple, arithmetic scoring model based on high-level evidence (meta-analysis of 15 studies) rather than single-center data. The model utilizes routinely available clinical, imaging, and pathological variables (e.g., tumor grade, cortical involvement, hemorrhage), making it accessible without complex genetic or imageomic data.
• Correction of Treatment Confounding: A novel application of counterfactual probability calibration was used to adjust for the bias introduced by widespread prophylactic medication in the validation cohort, allowing for a more accurate estimation of natural disease risk.
• Quantified Clinical Decision Pathway: Unlike previous models that stopped at risk identification, this study translates risk scores into specific medication strategies using Decision Curve Analysis. It explicitly defines the net benefit of interventions across different risk strata.

Results

• Risk Factors: The meta-analysis identified eight independent risk factors: history of epilepsy, preoperative seizure frequency $\ge$1/day, preoperative motor deficits, low-grade glioma, tumor diameter $\ge$5 cm, cortical involvement, peritumoral edema, and postoperative intracavitary hemorrhage.
• Model Performance: The external validation demonstrated good discriminatory power with an AUC of 0.854 (95% CI: 0.782–0.926). The optimal cutoff was set at 6.5 points.
• Risk Stratification:
  • Low-Risk (Score < 6.5): Observed seizure incidence was 0.94%.
  • Moderate-Risk (6.5 $\le$ Score < 13): Observed incidence was 12.84%.
  • High-Risk (Score $\ge$ 13): Observed incidence was 40.0%.
• Decision Curve Analysis Findings:
  • Low-Risk Group: The net benefit of any prophylactic medication strategy was negative across all threshold probabilities, indicating that the harms and costs of medication outweigh the benefits of preventing rare seizures.
  • Moderate-Risk Group: The "risk-stratified" strategy yielded the highest net benefit. Universal medication provided only marginal benefits at very low thresholds (<12%) and resulted in net harm at higher thresholds.
  • High-Risk Group: Both "universal medication" and "risk-stratified medication" strategies showed significantly higher net benefits than "no medication," with overlapping curves suggesting that routine prophylaxis is strongly indicated.

Significance and Claims
The authors claim that this study provides the first comprehensive system for predicting early postoperative seizures that bridges the gap between risk identification and clinical intervention. By integrating counterfactual calibration with Decision Curve Analysis, the study offers a quantified, individualized decision pathway:

1. Low-risk patients should avoid routine prophylactic medication to adhere to the principle of "first, do no harm."
2. Moderate-risk patients should receive personalized, selective medication based on specific clinical thresholds, avoiding a "one-size-fits-all" approach.
3. High-risk patients are strongly recommended for routine prophylaxis to minimize the risk of missing those who would benefit.

The authors emphasize that this strategy aims to reduce the incidence of postoperative seizures while avoiding unnecessary adverse drug reactions and medical resource waste. They acknowledge limitations, including the retrospective nature of the validation data and the lack of prospective intervention studies to confirm improved patient outcomes, suggesting that future randomized controlled trials are necessary to further validate the model's clinical utility.