Dugan, P; Carlson, C; Menendez, J; Flom, P; Knowlton, R; French, J
"Utilizing a decision tree model to predict outcome for patients assessed for epilepsy surgery with EEG, MRI and IQ as factors" [Meeting Abstract]
Epilepsy currents 2011 2011; 11(1):-
- Rationale: Resective surgical treatment can be curative in a large subset of patients with treatment resistant epilepsy. There is a need for a simple surgical grading tool which can be employed by the referring neurologist, ideally utilizing information obtained prior to diagnostic hospitalization. Our hypothesis was that a model using interictal EEG, brain MRI, seizure semiology and IQ could stratify patients with treatment resistant epilepsy with respect to their likelihood of achieving seizure freedom following assessment for resective epilepsy surgery. Methods: A prospectively identified cohort of 211 patients from the University of Alabama was combined with a retrospectively identified cohort of 193 consecutive patients at New York University presented in surgical multidisciplinary conference and either proceeded to surgery or were excluded as surgical candidates. All met inclusion criteria: age e18, focal epilepsy diagnosis e2 years, failed e1 medication, e1 seizure 3 months prior to admission, follow-up>6 months. Patients were classified as seizure free following resective surgery or not seizure free following resective surgery/no surgery. Pre-operative EEG, MRI, seizure semiology and IQ data were reviewed, systematically categorized (Table 1) and were utilized in a decision tree algorithm to predict seizure freedom. When p<0.05 was utilized, a simplistic dichotomous algorithm resulted. Therefore, to further explore, a relaxed, exploratory statistical significance level of p<0.25 was used. Nodes resulting in >50% of patients becoming seizure free were considered predictive of seizure freedom. Results: The overall seizure freedom rate was 46.8%. The exploratory decision tree analysis (Figure 1) resulted in two EEG groups: F, G, H (bilateral temporal, bilateral extra-temporal, bisynchronous; node 2; N=97) versus all others (node 3; N=307). For node 3, MRI was employed for further stratification: b (unilateral mesial temporal sclerosis (MTS); node 4; N=75) versus all others (node 7; N=232). Node 4 was further stratified based upon IQ: below 70 (node 5; N=11), above 70 (node 6; N=64). Semiology had no statistically significant impact. The model correctly predicts outcome in 62% of patients, yielding a positive predictive value of 55%, and a negative predictive value of 71%, with sensitivity of 74% and specificity of 51%. Conclusions: The relatively low overall seizure freedom rate is due to the fact that patients considered for, but ultimately not undergoing, surgery were included in the analysis, which better reflects the actual decision process for patients and neurologists. Of interest, the model fails to identify the commonly considered "best surgical" group of unilateral MTS with concordant interictal activity as a unique node. Notably, the main predictive factor was interictal EEG; other factors were only statistically viable with a more exploratory statistical approach. The positive and negative predictive values in this model are probably not sufficient for clinical use
Check for full text:
# 175847 (EMBASE:70830772)
This publication list a product of the NYU Faculty Bibliography.