Fig. 3.

Results of logistic regression with an area under the receiver operating characteristic curve (AUC) for alternative risk-scoring models in sEOAD. For this analysis, the APOE locus was defined as a 500 kb region surrounding the APOE gene, and the scores produced by PRSice for this model are based on the SNPs within that region; PRS represents the score produced for all SNPs present on both the NeuroX array and in the base data set. The relevant variables included sex together with the number of APOE ε2 allele and/or APOE ε4 allele. As shown in the table, a nonsignificant Hosmer-Lemeshow p-value suggests that the model is suitable for using as a predictive tool. Nagelkerke’s R2 can also be used to identify the best model for risk prediction; the higher the value of R2 the greater the predictive accuracy of each model. This approach identified sex, ε2, ε4 + PRS as the best model for calculating risk in our sEOAD cohort as the largest AUC value is produced from the combination of variables. Abbreviations: PRS, polygenic risk score; sEOAD, sporadic early-onset Alzheimer’s disease; SNPs, single nucleotide polymorphisms.