Annual Meeting Abstracts: View

  • (Abst. 1.347), 2017
  • Predictive Analytics for Surgical Outcome Stratification in Epilepsy
  • Authors: Pue Farooque, Yale University School of Medicine; Joel Caplan, Rutgers University; Les Kennedy, Rutgers University; Dennis D. Spencer, Yale University; and Katie Bandt, Northwestern University
  • Content:

    Rationale: The ability to predict an individual patient’s outcome following epilepsy surgery remains challenging. Recent advances in predictive analytics may facilitate improved surgical prognostication. One such example of predictive analytics that may be applied to epilepsy is risk terrain modeling (RTM). We report the application of RTM to epilepsy and its utility in pre-surgically predicting surgical outcome.  Methods: All patients undergoing anterior temporal lobectomy for management of refractory temporal lobe epilepsy between 2001-2015 at Yale University were identified from a prospectively compiled database. Variables were selected based on their ability to be spatially defined in anatomic and/or radiographic space as integration of these variables with the RTM algorithm requires spatial representation of input variables. The variables obtained included clinical seizure semiology, clinical MRI and PET interpretation, ictal SPECT if performed, scalp electroencephalography (EEG) including both interictal and ictal findings and clinical findings from neuropsychological evaluation. Exclusion criteria included incomplete clinical records resulting in insufficient variable definition and/or post-surgical follow-up less than 2 years. Integration of collected variables with the RTM predictive algorithm was used to predict post-surgical outcome at 2 years.  Results: 275 patients undergoing AMTL for TLE between 2001 and 2015 were identified from the prospectively collected database. Seventy-nine patients were selected for inclusion based on availability of clinical data and duration of follow-up meeting or exceeding 2 years. Of these 60/79 (75.9%) were seizure free (Engel class I), and 19/79 (24.1%) were not seizure free (Engel class II-IV). RTM was explored for its ability to predict this outcome based on the spatial relationships between variables. Interestingly, only limited spatial relationships were identified using RTM to predict surgical outcome based on pre-surgical variables further validating the super-spatial networked relationships at play in epilepsy. Work is ongoing to further define these relationships using alternative predictive analytic techniques.  Conclusions: RTM represents a novel, non-invasive, data-driven, individualized tool for predicting surgical outcome in patients undergoing surgery for epilepsy. Larger studies including pooled, multi-institutional data will be required to expand its utility and diversity of its application to epilepsy.  Funding: There was no funding associated with this study