Predicting memory decline from left temporal lobe epilepsy surgery using preoperative fMRI: a multicenter study.

OBJECTIVE

While fMRI language laterality has been used to predict verbal memory after epilepsy surgery, supporting evidence is not yet definitive. The FMRI in Anterior Temporal Epilepsy Surgery (FATES) project was a prospective observational cohort study at 10 US epilepsy centers, performed to determine whether a multivariable model including fMRI language laterality can predict verbal memory outcome 6 months after left temporal lobe epilepsy surgery.

METHODS

This analysis includes 70 adults who underwent left temporal lobe epilepsy surgeries including hippocampal resection for treatment of epilepsy. Patients completed standard protocols, including preoperative fMRI language mapping with a semantic decision versus tone decision (SDTD) task contrast and preoperative and postoperative verbal memory assessment. Five memory measures, obtained from the Selective Reminding Test, Rey Auditory Verbal Learning Test, and Wechsler Memory Scale, were selected to cover a range of episodic memory tasks including word list learning, delayed list recall, and story recall. Multiple linear regression was performed using the preoperative memory score, duration of epilepsy, age at epilepsy onset, age at surgery, hippocampal sclerosis status, and fMRI SDTD laterality index (LI) as predictor variables for 6-month postoperative verbal memory change.

RESULTS

Across different memory measures, decline (defined using Reliable Change Index) was seen in 8%-28% of patients. Changes on all five memory measures were significantly correlated with preoperative score, epilepsy duration, and fMRI SDTD LI. Variance explained by linear regression models for each test ranged from 34%-41%, with fMRI LI independently accounting for 7%-25% of the total variance (all p < 0.05). Cross-validation accuracy for predicting change scores in independent held-back samples ranged from 0.54-0.75 standard deviations of the preoperative sample.

SIGNIFICANCE

We demonstrate that models incorporating fMRI language LIs from a semantic decision task trained to predict verbal memory decline after left temporal lobe surgery provide meaningful information to help inform patients of the risk associated with left temporal lobe surgery and are practical to implement in different hospital settings. This confirms previous limited evidence that fMRI-based preoperative language protocols can be used to predict verbal episodic memory outcome after left temporal lobe epilepsy surgery.