Statistical analysis.

We compared Ng levels between the different clinical groups and also according to the CSF AD biomarker profile. Because of small numbers, we excluded those with genetic AD from the statistical analyses, but no other data were missing or excluded. To compare demographic, CSF biomarker, and clinical data between groups, we used the nonparametric Kruskal–Wallis test with the Dunn test to correct for multiple comparisons and report p values both with and without correction for multiple comparisons. To further account for multiple testing, we set the threshold for statistical significance to p < 0.01. Fisher exact test was used to compare the distribution of categorical data across groups. Data are shown as medians and interquartile ranges for numerical data or as a percentage for categorical data. Correlations between biomarker data were assessed using Spearman rank correlation. The optimal cutoff point of CSF Ng to differentiate patients with AD from healthy controls was identified by selecting the CSF Ng concentration that produced the highest Youden³² index (J = sensitivity + specificity – 1). This cutoff point was used to calculate specificities against non-AD diagnoses as a whole group, as well as against each non-AD diagnosis. Statistical analysis was performed using commercial software (GraphPad Prism version 6.00 for Windows; GraphPad Software, San Diego CA).