<![CDATA[In traumatic brain injury (TBI), non-invasive proxies for mass effect are crucial. The optic nerve sheath diameter (ONSD) is used to estimate intracranial pressure (ICP), but its correlation with structural outcomes like midline shift (MLS) is poorly defined, particularly regarding the optimal measurement method (unilateral vs. bilateral). We prospectively enrolled 38 adult TBI patients who received both a CT scan and a bedside ONSD ultrasound within 24 hours. Data was re-analyzed to classify ONSD relative to lesion location (Ipsilateral, Contralateral) and to correlate these, plus the Bilateral Average (ONSD-Avg), with CT-measured MLS using Spearman's correlation. We used linear regression to assess quantitative prediction (R-square) and binary logistic regression (ROC curve) to assess clinical classification (AUC) for predicting MLS >5mm. A significant, positive correlation was found between MLS and Ipsilateral-ONSD (rs = 0.450, p = 0.005) and ONSD-Avg (rs = 0.383, p = 0.018). The Contralateral-ONSD correlation was not significant (rs = 0.210, p = 0.206). A Wilcoxon test confirmed Ipsilateral-ONSD was significantly wider than Contralateral-ONSD (p < 0.01). The linear regression model for MLS quantification was statistically significant (p = 0.015) but had a very low predictive power (R-square = 0.153). In contrast, the logistic regression model found ONSD-Avg to be an excellent classifier for detecting surgical MLS (> 5mm), with an Area Under the Curve (AUC) of 0.88 (95% CI 0.75-0.96). In conclusion, ONSD measurement is significantly affected by asymmetric, unilateral TBI pathology. The bilateral average (ONSD-Avg) is the most reliable screening method, as it compensates for unilateral pressure gradients. The low R-square (15.3%) confirms ONSD is a poor quantitative predictor of MLS, reflecting the non-linear pressure-volume relationship. However, the high AUC (0.88) proves ONSD is an excellent clinical classifier for identifying patients with surgical-threshold mass effect. ONSD should not be used to "quantify" MLS, but rather to "classify" patient risk.]]>
