2026 Proffered Presentations
S247: DEVELOPMENT OF A CLINICAL NOMOGRAM TO PREDICT TRIGEMINAL POSTOPERATIVE DYSFUNCTION.
Karen Salmeron-Moreno, MD, MSc; Karthik Papisetty, BA; Chris D Kim; Justin Maldonado, MD; Rommi Kashlan, BS; Hithardhi Duggireddy, MSc; Travis Atchley, MD; Josephine Buclez, MSc; Thomas McCaffery, BS; Jennifer M Kim; Derek Hu; Uday Thakar; Karthik Valiveti; Youssef M Zohdy, MD; Gustavo Pradilla, MD; Tomas Garzon-Muvdi, MD, MSc; Emory University
BACKGROUND: Meningiomas are among the most common intracranial tumors treated by neurosurgeons. Skull base meningiomas, in particular, present unique challenges given their proximity to critical neurovascular structures and cranial nerves. Trigeminal involvement and associated facial pain/numbness is frequently observed at presentation and serves as an important marker of postoperative outcomes. This study seeks to identify preoperative variables that may predict postoperative trigeminal dysfunction.
OBJECTIVE: To develop and internally validate a nomogram for postoperative trigeminal dysfunction using age and presenting symptoms.
METHODS: We retrospectively screened 172 patients with skull base meningiomas. The postoperative trigeminal nerve morbidity outcome was initially classified as Improved, Persisted, or New symptom, and binarized as event (Persisted/New) vs. no event (Improved). Predictive modeling was restricted to the 29 patients with available trigeminal outcomes (31.0% events). Predictors included age at surgery and presenting symptoms: trigeminal dysfunction (defined as facial pain, numbness and/or dysesthesias), visual field deficits, and sensory complaints. Missing data were addressed via single imputation (median for age; mode for categorical variables). Given the predominance of female patients (93.1%, 27/29), sex was excluded to avoid model instability; a sensitivity analysis including sex yielded comparable performance. A ridge-penalized logistic regression model was fitted, with penalty parameter selected via pentrace. Model performance was internally validated using bootstrap resampling (B = 200). Discrimination was summarized by the optimism-corrected area under the curve (AUC), overall accuracy by the Brier score, and calibration by logistic calibration (intercept and slope).
RESULTS: Among the patients with available outcomes, mean age was 50.8 ± 9.3 years (median 51; IQR 47–55). Symptoms at presentation were: trigeminal dysfunction in 86.2% (25/29), visual field deficits in 19.0% (4/21), and sensory complaints in 41.2% (7/17).
Model performance showed an apparent AUC of 0.839, bootstrap-corrected AUC of 0.757, and Brier score of 0.120. Logistic calibration indicated an intercept of +0.548 and a slope of 1.661 (Figure 1), indicating acceptable but imprecise calibration given the small sample. Multivariable analysis showed trigeminal dysfunction at presentation associated with lower odds of the event (OR 0.05, 95% CI 0.01–0.48). In contrast, age (+10 years; OR 0.45, 95% CI 0.11–1.85), visual field deficits (OR 1.41, 95% CI 0.13–15.6), and sensory complaints (OR 1.02, 95% CI 0.13–7.96) presented wide confidence intervals. These effect sizes should be interpreted as exploratory given limited denominators. A points-based nomogram was developed for bedside application (Figure 2). Figure 3 shows odds ratios with 95% confidence intervals; age is reported per 10-year increment.
CONCLUSIONS: Our 4-variable model demonstrated strong apparent discrimination and acceptable optimism-corrected performance (AUC 0.757) in predicting postoperative trigeminal dysfunction. Given the sample size, our findings are proof of concept; external validation is required.
