Fig 1.
Comparison of biomarker levels among groups with and without multiple sclerosis.
(A) NFL (age adjusted) levels, (B) GFAP levels, (C) tau levels, (D) IgG index, (E) BPF values, (F) RNFL thickness, and (G) macula volume in control groups (HC, SC) and in patients with OD, PrMS, CIS/early RRMS (with disease duration <2 years; ERRMS), and established RRMS (disease duration ≥ 2 years; LRRMS). Dots represent the log of the biomarker values or the tissue properties measured in each subject. Boxes are the IQRs and horizontal lines are the median values. Brackets indicate pairwise group comparisons with corresponding p-values. The age adjusted log(NFL) was calculated, as follows: log(NFL)– 4.24–0.040 * Age (y), as estimated by linear regression of the HC group. The age adjusted log(tau) was calculated, as follows: log(tau)– 5.75 + 0.015*Age (y), as estimated by linear regression of the HC group.
Table 1.
Characteristics of the patients and controls.
Fig 2.
Correlation matrices for biomarkers.
Matrices include (A) all patients and controls, (B) only patients with CIS/RRMS, and (C) only SC and HC groups. Blue: positive correlations; red: negative correlations. Spearman rank correlations are shown as numbers in the cells. The variables are clustered are reordered using hierarchical clustering.
Fig 3.
Scatterplot of data projected to the first two principal components (PC1 and PC2).
PC1 and PC2 explain 27.7% and 15.7% of the variability in data, respectively (total 43.4%). The projections of the biomarkers onto the principal components are shown as arrows. The HC and SC groups are mainly separated from the RRMS and PrMS groups by the biomarkers measured in the CSF. The OD group overlaps with both of the other groups.
Fig 4.
Patients with previously undiagnosed RRMS (grey), the SC group (red), and the OD group (green) were classified by splitting the groups based on the detection of the indicated biomarkers. The class of the majority of patients in a group is shown at the top of each node. The proportions of individuals that belong to classes SC, RRMS, and OD are shown as fractions in the middle of each node. The percentage of the total number of individuals in each node is shown at the bottom of each node. This model makes two splits, the first is based on intrathecal IgG production, and the second is based on NFL levels (age-adjusted), with an optimal split in age adjusted NFL at 148 ng/L. Thus, individuals with no intrathecal IgG production and low levels of NFL were predicted to be SC (51%). Among the individuals that satisfied these two criteria, 81% were truly SC, 2% had RRMS, and 17% had some other disease. Only 6% of individuals exhibited no intrathecal IgG production and high levels of NFL. Of these, 8% were SC, 25% had RRMS, and 67% had OD. Finally, 42% of individuals exhibited only intrathecal IgG production. Of these, 4% were SC, 85% had RRMS, and 12% had OD. The calculation for age-adjusted NFL is: NFL– 203.3–15.9 * Age (y), as estimated by linear regression of the HC group.
Table 2.
Predicted and observed group membership using a classification tree model with intrathecal IgG production and NFL.
The formula for age adjusted NFL is given by NFL– 203.3–15.9 * Age (y), as estimated by linear regression in healthy controls.