Data were analyzed using the R statistical software (version 4.0.2) [27 ]. For each group, quantitative variables were expressed as the median with the interquartile range (IQR, Q1 and Q3). Groups were compared using non-parametric Kruskal–Wallis or Mann–Whitney tests according to the group number. Pairwise comparisons were adjusted with Bonferroni correction. Correlations were assessed using Spearman’s rank correlation coefficient (ρ). The distribution of categorical variables was expressed with percentages and compared using Fisher’s exact test.
For comparisons of biomarker concentrations between diagnosis groups, to avoid the influence of extreme values, outliers were identified using Rosner’s test and discarded (n = 6). For each group, normal distribution was assessed using the Shapiro–Wilk test and homoscedasticity through Levene’s test. The assumption of normality was not obtained in only two groups, IPMS-Shim and Simoa Aβ42/40 for the AD diagnosis. ANCOVA’s assumptions of linearity, homogeneity of variance, non-collinearity of the factors (variance inflation factor <5), non-influential observations, and normality of residuals were evaluated. The impact of diagnosis on plasma amyloid biomarker concentrations was evaluated using ANCOVA controlling for age and APOE ε4 status. Multiple comparisons of the means were achieved using Tukey contrasts with diagnosis as a factor.
Plasma cutoffs were computed using expectation–maximization (EM) algorithms for mixtures of univariate normal distributions [28 ]. Cutoffs were visually determined at the intersection of two normal distributions.
A receiver operating characteristic (ROC) analysis was used to determine biomarker performances. A predictive formula adjusted for age and APOE ε4 status was built using a logistic regression analysis. The best values for sensitivity (se) and specificity (sp) were computed at an optimal cutoff point. Youden’s index was used to determine this optimal cutoff corresponding to the threshold maximizing the distance to the identity (diagonal) line and giving and equal weight to sensitivity and specificity. The area under the curve (AUC) was compared using the DeLong test. All tests were two-tailed, and significance was set at α = .05.
For comparisons of biomarker concentrations between diagnosis groups, to avoid the influence of extreme values, outliers were identified using Rosner’s test and discarded (n = 6). For each group, normal distribution was assessed using the Shapiro–Wilk test and homoscedasticity through Levene’s test. The assumption of normality was not obtained in only two groups, IPMS-Shim and Simoa Aβ42/40 for the AD diagnosis. ANCOVA’s assumptions of linearity, homogeneity of variance, non-collinearity of the factors (variance inflation factor <5), non-influential observations, and normality of residuals were evaluated. The impact of diagnosis on plasma amyloid biomarker concentrations was evaluated using ANCOVA controlling for age and APOE ε4 status. Multiple comparisons of the means were achieved using Tukey contrasts with diagnosis as a factor.
Plasma cutoffs were computed using expectation–maximization (EM) algorithms for mixtures of univariate normal distributions [28 ]. Cutoffs were visually determined at the intersection of two normal distributions.
A receiver operating characteristic (ROC) analysis was used to determine biomarker performances. A predictive formula adjusted for age and APOE ε4 status was built using a logistic regression analysis. The best values for sensitivity (se) and specificity (sp) were computed at an optimal cutoff point. Youden’s index was used to determine this optimal cutoff corresponding to the threshold maximizing the distance to the identity (diagonal) line and giving and equal weight to sensitivity and specificity. The area under the curve (AUC) was compared using the DeLong test. All tests were two-tailed, and significance was set at α = .05.
Free full text: Click here
