Validation of a multigenic model to predict seizure control in newly treated epilepsy

Kanvel Shazadi, Slavé Petrovski, Annie Roten, Hugh Miller, Richard M. Huggins, Martin J. Brodie, Munir Pirmohamed, Michael R. Johnson, Anthony G. Marson, Terence J. O'Brien, Graeme J. Sills

Research output: Contribution to journalArticleResearchpeer-review

4 Citations (Scopus)

Abstract

A multigenic classifier based on five single nucleotide polymorphisms (SNPs) was previously reported to predict treatment response in an Australian newly-diagnosed epilepsy cohort using a k-nearest neighbour (. kNN) algorithm. We assessed the validity of this classifier in predicting response to initial antiepileptic drug (AED) treatment in two UK cohorts of newly-diagnosed epilepsy and investigated the utility of these five SNPs in predicting seizure control in general. The original Australian cohort constituted the training set for the classifier and was used to predict response to the first well-tolerated AED monotherapy in independently recruited UK cohorts (Glasgow, n=. 281; SANAD, n=. 491). A "leave-one-out" cross-validation was also employed, with training sets derived internally from the UK datasets. The multigenic classifier using the Australian cohort as the training set was unable to predict treatment response in either UK cohort. In the "leave-one-out" analysis, the five SNPs collectively predicted treatment response in both Glasgow and SANAD patients prescribed either carbamazepine or valproate (Glasgow OR. =. 3.1, 95% CI. =. 1.4-6.6, p=. 0.018; SANAD OR. =. 2.8, 95% CI. =. 1.3-6.1, p=. 0.048), but not those receiving lamotrigine (Glasgow OR. =. 1.3, 95% CI. =. 0.6-2.8, p=. 1.0; SANAD OR. =. 2.2, 95% CI. =. 0.9-5.4, p=. 0.36) or other AEDs (Glasgow OR. =. 0.6, 95% CI. =. 0.2-2.0, p=. 1.0; SANAD OR. =. 1.9, 95% CI. =. 0.9-4.2, p=. 0.36). The Australian-based multigenic kNN model is not predictive of initial treatment response in UK cohorts of newly-diagnosed epilepsy. However, the five SNPs identified in the original Australian study appear to collectively have a predictive influence in UK patients prescribed either carbamazepine or valproate.

Original languageEnglish
Pages (from-to)1797-1805
Number of pages9
JournalEpilepsy Research
Volume108
Issue number10
DOIs
Publication statusPublished - 1 Dec 2014
Externally publishedYes

Keywords

  • Antiepileptic drug
  • Epilepsy
  • Genetic association
  • K-Nearest neighbour
  • Machine-learning
  • Treatment outcome

Cite this