Figure 3.

Importance of individual cytokines in classifying patient groups. Plots represent models discriminating patients with early inflammatory arthritis who develop rheumatoid arthritis (RA) from (a,b) all other early inflammatory arthritis patients (early disease that progresses to non-rheumatoid persistent disease plus early non-crystal-related resolving arthritis plus crystal-related resolving arthritis) and from (c,d) established RA. Panels a and c show the relative importance of the cytokines in the overall classification. The vertical axes represent individual cytokines arranged according to importance. The horizontal axes represent the average decrease in classification accuracy seen when the values for each cytokine are permuted. Important cytokines are associated with a greater decrease in classification accuracy. The plots shown in panels b and d are metric multidimensional scaling (MDS) representations of the proximity matrices of the Random Forest models demonstrating the relationship between individual patients in the two models. The two axes represent the first and second MDS axes. Closed circles represent patients with early inflammatory arthritis who develop RA in both panels; open circles represent patients with (panel b) all other early inflammatory arthritis and (panel d) established RA.