Stroke subtype classification, applying standard criteria to clinical data, can reduce the heterogeneity of ischemic stroke for genetics studies. To date, most of the replicated genetic loci for ischemic stroke appear subtype-specific. Increasing homogeneity of phenotype comes with a trade-off in sample size, which influences the potential for successful identification of new loci. We estimated genetic associations using the union and intersection of two widely used stroke subtyping systems to assess the influence of sample size and homogeneity on test statistics.
Methods: The Stroke Genetics Network (SiGN) study used both the Causative Classification of Stroke (CCS) and the Trial of ORG10172 Acute Stroke Trial (TOAST) classification systems. The CCS generates both a causal and a phenotypic subtype. Using all available data from case-control strata from the a previously performed GWAS, we focused on three main stroke subtypes: cardioembolic (CE), large artery atherosclerosis (LAA), and small artery occlusion (SAO). We estimated genetic associations with the union of the two CCS outputs and TOAST (eg called CE by at least one method) and the intersection (eg called CE by all three methods). Our modelling approach included a fixed-effects meta-analysis, i.e. pooling stratum specific estimates from logistic regression models using 10 principle components to account for genetic ancestry responsible for population stratification.
Results: The majority of confirmatory findings from SiGN (PITX2 and ZFHX3 for CE and HDAC9 for LAA) were evident in both models and in both the union and intersection of the subtyping systems. Some findings (TSPAN2 in LAA) showed up best in the intersection with stronger corresponding p-values and odds ratios whereas other findings (12q24 for SAO) appear stronger in all of the union analyses, but were also evident in intersection analyses. We identify 16 potential novel loci, 6 of which appear in numerous analyses.
Conclusion: Both expanding and refining stroke subtypes may help in the identification of additional stroke genetic risk variants and should be considered as complementary to a single classification system. We are pursuing replication of the novel findings.