Figure 2. Box plots of error rates for the full ordinal scale of mRS (mRS 0.6), considering mRS 0 to 3 as individual grades and collapsing mRS grades 4 to 6 (mRS 0.3,4–6), dichotomizing at various cut-points of mRS 1 (mRS 0–1, 2–6), mRS 2 (mRS 0–2, 3–6), mRS 3 (mRS 0–3, 4–6) and mRS 4 (mRS 0–4, 5–6).

van Swieten’s inter-rater reliability matrix used as confusion matrix. (p<.001 ANOVA; post-hoc testing shows that all dichotomization errors are lower than either full scale errors with mRS 0–4 dichotomization the lowest; p<.05).