Abstracts

Rationale: A previous report identified a five-SNP combination with the capacity to predict seizure control in newly-treated epilepsy based on an Australian cohort where 96% of individuals were initially treated with carbamazepine or valproate. A k-Nearest Neighbour (kNN) supervised learning model utilizing the five-SNPs was found to significantly predict seizure control in two independent Australian replication populations (p=0.0004 and p=0.003). The current study aimed: i) to assess the broader clinical utility of the multi-SNP kNN model by investigating the ability of the Australian developmental cohort (training set) to predict seizure control in two UK newly-treated epilepsy cohorts (test sets), a Scottish (n=281) and a English (n=491) cohort, and ii) to assess the relevance that the five-SNPs might collectively have in predicting seizure control within each of these two external populations.Methods: The Australian developmental cohort was used to attempt to predict the Scottish and English population pharmacoresponse outcome. A secondary, controlled investigation, using a leave-one-out testing approach was adopted. In the leave-one-out approach, the kNN model uses the same five-SNPs identified in the Australian cohort, in conjunction with the leave-one-out training sets derived from within each of the UK cohorts, to predict each of the individual test cases from the Scottish or English cohorts, respectively. Given that the selection of the five-SNP combination was independent of the UK populations, this assessed whether the combination of genetic markers originally identified in the Australian population has predictive value for seizure control in UK newly-treated populations.Results: The Australian developmental cohort was unable to predict pharmacoresponse in either the Scottish or English newly-treated populations (p>0.05). However, the leave-one-out analyses show that the five-SNPs collectively provide a significant prediction in Scottish patients initially treated with carbamazepine or valproate (n=118; p=0.003), but not with lamotrigine (n=112; p=0.3), or other anti-epileptic drugs (n=51; p=0.8). Likewise, this observation was confirmed when the leave-one-out approach in the English cohort significantly predicted initial response of English patients who were initially treated with carbamazepine or valproate (n=123; p=0.008), but not with lamotrigine (n=97; p=0.06), or other anti-epileptic drugs (n=123; p=0.06).Conclusions: The Australian-based 5 SNP kNN model previously reported to predict seizure control in newly treated Australian patients is not directly applicable in populations from the UK. This might be due to a contribution of both genomic and cohort differences. However, under a controlled analysis, the five-SNPs identified as being important in the Australian population appear to have a collective influence in predicting seizure control for newly treated epilepsy patients prescribed either carbamazepine or valproate in both UK populations (p<0.05).