Fig. 2
From: Alzheimer’s disease-associated complement gene variants influence plasma complement protein levels
ROC Curves demonstrate the AD predictive capacity of complement biomarkers. ROC curves were generated using multiple GLMs for each protein. AUC statistics for individual (A–D) or combined (E, F) analytes are shown for EOAD (A, C, E) and LOAD (B, D, F). 95% confidence intervals, calculated using bootstrapping with 2000 replicates, are included in brackets for each GLM. Proteins unadjusted (A, B) and adjusted (C, D) for age and sex were included for comparison. Linear predictors were selected based on results of stepAIC models of adjusted data for EOAD (E) and LOAD (F), with the best model based on highest AICs relative to model complexity plotted first (EOAD = Aβ40 + Aβ40/Aβ42 + GFAP + C1q; LOAD = Aβ40 + Aβ40/Aβ42 + pTau18 + NfL + GFAP + CR1 + C1q + Factor H for LOAD) and then 3 regressive steps from the stepAIC results plotted sequentially